Ladies and gentlemen, please welcome Intel CEO, Brian Krzanich.
Good afternoon, everybody. Welcome to Intel's first Artificial Intelligence Day, or as we like to call it, AI Day. I think Diane and team have put together a really great day today to really allow for a set of conversations that will both describe a little bit about our vision of artificial intelligence, but also give you a little more detail about what really is encompassed when people use the terms of artificial intelligence, and what does that mean for our technology. With that, let's get right into it. Today, what I'm going to talk about is three things. The first I thought we'd do is spend a little time talking about Intel's vision around artificial intelligence, and really kind of explain what the different meanings of artificial intelligence are.
The second is talk about how AI is really becoming pervasive and how we think about that and how we're looking at that across our product portfolio and across the industries we serve. Finally, how Intel is really trying to lead and be a catalyst for AI and accelerate it, move it faster. We believe it can bring great attributes and great value to businesses and operations and the silicon we provide. Let's get right into the first topic for today. For us, we really look at this as that Intel is continuing to evolve and really change the way we look at products and the way we look at technology. We're really evolving to become a company that tries to power all the billions of smart and connected devices.
As you look at that, you really have to think about how important artificial intelligence is and how it becomes a critical component in those connected devices. Those devices really become just noise without some form of artificial intelligence to allow us to understand and really correlate that data. To start with, I thought we'd take a look at just how machine learning, deep learning, and artificial intelligence, those are terms often used, how those all are interconnected and what they really mean. If we could just roll our video really quick, it'll help us.
Artificial intelligence. Machine learning. Deep learning. For anyone who's not a data scientist, the terms associated with computing's next big wave can be pretty confusing. At the most basic level, the relationship between these three key concepts is fairly simple. Artificial intelligence refers to the broad category of computing in which a system is capable of learning directly from data, a capability made possible by the advent of data analytics, and applied either through a set of rules that are evolved over time through human intervention with the system, or increasingly, through the process of machine learning, in which the system expands its ability to process and use information without the need for additional human intervention.
While the broader concept of artificial intelligence has been around since the 1950s, it was the introduction of machine learning in the '80s that brought the first real wave of significant advancements to the field. As computational performance reached a tipping point in the last decade, a new wave of even more dramatic breakthroughs began to gather momentum. Deep learning is a branch or subset of machine learning that leverages those computational advantages and a set of new techniques, such as neural networks, to make sense of and learn from even larger amounts of data. The result has been an impressive acceleration in the progress of elusive artificial intelligence capabilities like image, speech, and natural language recognition that promise to have a significant positive impact on our everyday lives.
The emergence of deep learning also serves as a reminder that the technology and techniques for achieving the full potential of artificial intelligence are not static, but will continue to evolve just like us.
Hopefully that gave you some insight. Those are terms often used in industry, in the communication areas, but often not well understood, in my opinion, and not well, really managed. What we're looking at is really investing through our technology and through acquisitions to really build and fuel artificial intelligence across everything we do. As you look at this what we call virtuous cycle, what you see is that the things at the bottom are the things that are out there collecting the data. As those things continue to collect data and push that data to the cloud, it's the artificial intelligence that's going to occur on both edges of this ecosystem that's important.
Allowing the things at the edge to use smaller sets of data to make more compact decisions that are important and need to be done more real time, and those things in the cloud that can be done with a much richer and much more volumetric set of data, but that can actually change how those things at the edge then behave and change and adjust to the environment around them. We're leading these computer transformations through innovation with our Xeon products, things like RealSense that allow those things at the edge to see and view the world with computer vision, and FPGAs. To do that, really, data analytics is a critical component of that. That ecosystem is really meaningless without that analytics applied to it. All of these things coming together deliver an end-to-end AI solution.
The data, the connectivity, and the analytics all provide what's really the AI environment. We believe that only with Intel can you get a range of computer solutions in a data center from general purpose targeted silicon to computer vision silicon, memory and storage, and communication assets. All of these things can come together to form an ecosystem that allows you to put in a complete AI solution. With all the innovation that's occurring and the assets that Intel has, we really need to take the next step, which is, how do we plan to make AI pervasive? That's really the next question I want to address here. As you saw in the video, and as we all know who are in this industry, AI is really still in its infancy. The algorithms that are being used are changing every day.
Data sets are becoming richer and more complex, the number of things at the edge are continuing to be increased and added, and the diversity of those things that are collecting data and the complexity of them is increasing every day, allowing the algorithms as well to change. Intel Xeon processors with GPUs are being used for deep learning, which is a small but a rapidly growing subset, as you saw in that video, of machine learning. The GPU architecture, however, does not have a unique advantage for AI. It's not the only solution that's out there. As AI evolves, highly scalable architectures are genuinely needed in order to scale with the data sets and the complexity of the problems that are being asked for.
At Intel, we believe our architecture can solve for these larger models and offer consistent from the edge of the device down at the edges, all the way through to the data center. Intel's committed to delivering really the next generation of AI through products and experiences. To deliver our AI vision, we've done a whole range of internal innovations, and you see that in our Xeon products, our Xeon Phis, all the way down through our Atom-based products and our Quark-based products. All the way through that product set, you see innovation and you see AI intrinsic in all of those. There's various different types of learning engines in each one of those products. In addition to that, we've had a series of acquisitions that have been targeted towards broadening and providing specific solutions towards this AI environment.
The first one I want to talk about is Saffron. Saffron is a leading AI as a service provider organization. It became part of the Intel team about a year ago. They're providing AI solutions to many sets of industries, financial industry, investment industry. The next one is Movidius. Movidius is unique in that it's a pending acquisition as of today, but it is a leader in what's called computer vision. Think about it as AI applied to when computers are out looking at the world and understanding just what is it that it's seeing. Being able to discern between a person, a cup, a dog, anything. That AI is what Movidius is doing. Then finally, it's Nervana. Nervana is really the top end here. It's a premier leader in deep learning and neural networks.
Nervana joined us this year, and this is really designed to bring us to the very top of the performance around deep learning and machine learning in this kind of an environment. Since the acquisition, we've been very active in integrating and bringing Nervana in and understanding how to integrate it best into our product portfolio and our IA roadmap. I thought rather than just listen to me this morning, and to tell us a little bit more, we'd bring Nervana CEO, Naveen Rao, on stage here, and have him come and talk to us a little bit about what Nervana does. Naveen
Thanks, Brian.
Thanks. First, welcome to Intel, right? You've been here about a quarter or so, a little bit more?
Yeah, about three months now, almost.
Yeah. Just welcome, and welcome to AI Day.
Thank you.
Maybe if you could tell us just a little bit about what does Nervana do, and exactly what have you been trying to do in Nervana from a product perspective?
Yeah. Understanding artificial intelligence has been a quest of humanity for a long time. Combination of biological inspiration with computer engineering, what we know how to build in silicon systems, has been a holy grail, really, for computing. I think we're actually at an inflection point where we understand enough about silicon, we understand enough about concepts of the brain. We can bring these things together and really take computers in a new direction.
That's something different about you as compared to a lot of the other data scientists I hear, and that's you talk a lot about the brain. You're trying to figure out how the brain gets to the computer, and how the computer becomes more like a brain.
That's something you have a real passion about, right?
Absolutely. My background is I spent 10 years in Silicon Valley designing chips, and then quit my job to get a PhD in neuroscience to try to get a hint on that answer to that question. I think we've actually hit on something that's really interesting, that we can actually take advantage of properties of silicon by using some inspiration from the brain, not at very low levels of mimicry, but at a higher level, and actually move computer architecture into the next wave.
Tell us a little bit specifically around deep learning. What's next from Nervana in deep learning?
We're going to be talking a lot more about that in the deep dive talk. Intel is committed and is ready to invest in many different levels in the technology. It's going to be integrated across the entire portfolio. Today, we are going to be talking about our chip technology that we'll be bringing to the market next year, as well as integrations of this into various product lines. I think it's going to touch everything, and it's extremely exciting for me.
What you bring is both hardware and software, correct?
That's right. In addition to the hardware roadmap, we're going to be displaying our unified strategy across software to bring together a lot of things that have been happening in industry, and make them work well across all of Intel's architectures.
Okay. Thanks a lot. I think the audience knows a little bit more about Nervana. Hope everybody attends your deep dive, and thanks a lot for coming on stage.
Thanks, Brian.
Thanks, Naveen. The goal of all these acquisitions is to really make AI pervasive across all of our products and our software suite. You could kind of hear that from Naveen as he describes not only how he's going to bring his silicon to market, but how he's going to work with us on producing some of those same kinds of intellectual pieces of property in a lot of our products and the software as well. Let's go to the next step, which is how do we make AI accelerated? How do we really push it across the boundaries and into almost everything we go and do? Let's talk about that for a little bit. One way we're accelerating AI is through our product portfolio, as we just talked about.
What I'm excited to announce today is that the Nervana brand, which will include both hardware and software, is becoming a part of the Intel branding network. It's built for speed and ease of use with a foundation of a highly optimized set of AI solutions that enable more data and more data professionals to solve the world's biggest challenges faster. You'll see the Nervana brand now used extensively across our deep learning and machine learning solution sets. Intel's committed to leading, we believe, the AI computing era, and we have a threefold approach to doing so. Let's walk through that. First, we want to drive the AI computing era, and we're going to compress the innovation cycles through products like Nervana's products, but all of the products, our Xeon Phi. You heard about Saffron.
We have many more products, like I said, from Xeon all the way down through Quark, that we're compressing all of their innovation cycles to drive AI across the board. It's through breakthroughs through data ingestion and understanding how data can best enter the compute model. Building training and deployment models that really are in the software side of this that really allow the compute to go through and understand the decisions and the environment and make the artificial intelligence really come to life. Diane's going to provide you a lot more details of that and examples on how we intend to drive this through this space. She'll be up next, and she'll talk about this. Second, we want to democratize access to AI.
Intel's leading the charge for open data and data exchanges and initiatives, really make it easy to use tools across the board. We're going to try and train more and more talent and really broaden availability of people and skill sets to this. Doug Fisher will provide you, a little bit later, with a deeper dive during his keynote around these tools. Finally, we want to be the trusted guide for AI within the industry. Intel has a long history of really being the trusted supplier for compute, for data across the industry. We want to maximize that and maximize the net positive impact of artificial intelligence to our world. We're engaging with governments, business, and society throughout the area to plan for this coming really transformation.
AI will generally transform most industries that we touch today, we believe there's a chance to make that transformation positive, to provide insights and abilities to humans they've never had before. We want to be the trusted leader and developer of that. You'll hear from several other speakers throughout the day, I think, as I said at the beginning, Diane and the team have put together really a great day here to really not only understand what is artificial intelligence, but how does it apply to our products, how do you really think about the software ecosystem, what is it going to have and what are the impacts to our society. I think there's going to be some fantastic examples today.
The last thing I want to leave you with today is just one example of how we're looking at a problem holistically and solving it with artificial intelligence. It's no secret that online and the online world has brought society numerous benefits. I think we all think about those every day. Along with those benefits have come challenges, and probably many of you on this audience have been witness to, or sometimes victim of, some form of harassment in this online world. If you have kids, and especially teenage kids, you've probably witnessed it for sure down at that level. At Intel, our goal is to power the smart connected world, if that world isn't inclusive, if it's not safe for people to go into, we'll never reach our full potential. The business will be stalled. You can see hints of that today.
We know we need to bring the industry together to stop online harassment. About a year ago, we brought together a team of partners to form a group called Hack Harassment. This is an amazing opportunity to use technology, specifically artificial intelligence, as a tool to address harassment. Using the power of artificial intelligence, we're working together on building an open source classifier that will detect and help and deter online harassment. Think of a tool that before you post, looks at your post and says, "That could be harassment. That could be harmful to people who receive this. Are you sure you want to send it?" Right? Just that little comment and that understanding could prevent somebody from doing something that they probably don't really want to do in the long run.
I invite you to take advantage of talking to our experts today who are speaking here and really can provide a deeper and more complete understanding of how Intel is advancing artificial intelligence. I'm really excited to have been the one to get to open up our first Artificial Intelligence Day, I'm really excited about what artificial intelligence is going to bring to Intel, really about what it's going to do for the industry. To talk more about this and to really get me off the stage I'm going to invite Diane Bryant, Executive Vice President of the Data Center Group, on stage to talk to you more. Thanks, Diane.
You're welcome to stay off the stage. Okay, welcome to all of you, really thank you for joining us. This is a big opportunity for us to engage with you and to tell you about our plans around artificial intelligence. In preparing for today's talk, I fell upon this quote from Intel's founder, Bob Noyce, he said that up until now, we've been trying to understand the brain by using computer modeling. Going forward, if we're going to better understand the future of the computer, perhaps we should start looking at the brain for some clues. Very consistent with what Naveen just said. That was 32 years ago. We have made tremendous progress in artificial intelligence over that time. However, much remains to be done. The opportunities remain immense. A very tangible example is the good old cat example that you all love.
If you were to show a three-year-old child four or five pictures of a cat, on the sixth picture, they'll say, "Cat." Now you can contrast that with the most sophisticated image resolution processing system you have today, you need to analyze 1 million different images of a cat if you're going to get a 97% probability that the next cat will actually be identified as a cat. We still have a very long ways to go in getting to that elusive position of where we can replicate the human brain. However, despite that big gap, we have seen very large step functions in innovation before in our industry. Intel is known for creating and leading and driving these big step functions in technology innovation. If you look back, we did it in the move from mainframes, the old mainframe era, to standards-based servers.
We've done it again in moving from standard-based servers to cloud computing. Now artificial intelligence is going to drive that next big wave of compute that will, as BK said, will truly, we believe, transform the way all businesses operate and how people fundamentally engage with the world. It is the fastest growing workload in the data center. By 2020, there will be more servers running data analytics than any other workload, and we predict that the amount of compute cycles running AI will grow by 12x between now and then, just four years out. That's growing at twice the rate of the overall compute market. One of the biggest innovators in the area of AI, and clearly one of the largest consumers of compute capacity, is Google. We are very fortunate to have had a very long and fruitful technology relationship with Google.
I'm excited today to share the news of a new collaboration between Intel and Google that will help shape the next wave of data center innovation across the cloud and across AI. Please join me in welcoming a very fabulous leader and friend, Diane Greene, the Google Cloud Senior Vice President. Hey, Diane. Nice to see you. We just like it because we're both named Diane. We just like to hang out together. No, but truly, thank you. Thank you for coming.
Oh, gosh, it's my pleasure. Thanks so much, and it was so great to announce that strategic partnership today. We've had this long-standing relationship between Google and Intel around the technology accelerating things, and now to be bringing it to the business side and the enterprise, as we sort of join together in our focus on secure systems, open systems, and we both believe in multi-cloud.
Yeah, it is great. We've got a formalization of a very strategic alliance that spans both technology and business. It spans the move to the cloud as well as TensorFlow and improvements in AI, and really a comprehensive agreement. Thank you for that.
Yeah. I think that as we work together in the cloud, there's on-prem cloud, there's multiple public clouds, we can work together to make it so that people can use them all very easily. Google, as an example, did Kubernetes, which we open sourced, which gives you orchestration across all these clouds. Then we've enjoyed a collaboration with you guys to optimize it for the Intel architecture, improve virtual networking, prioritization of resource models, some real technologies there, and delivering these code optimizations. Also, we've been working with you on security, of course, all important to the enterprise, and deep collaboration around making Intel hardware and Google Cloud support a more secure world for us.
Yeah. The direct optimizations that we've been doing to make sure Kubernetes, the open-source solution that you obviously invented and now have released to the world.
making that run best on IA. It helps the broad community, but we truly do believe it's going to help enterprises deploy that next generation, developed for the cloud type of solutions like artificial intelligence. It's a big opportunity, a nice investment, nice collaboration with you guys.
Thanks.
Yeah.
Thanks. Then the other one we're collaborating on, and completely in tune with your day-to-day, is TensorFlow, which is Google's library. Google Brain developed this library for machine learning, and we've open sourced it, and now we're collaborating with you to provide accelerations with the Intel platforms over a range of neural network models so that it can work for both training and inferencing. Those joint optimization should make it really perform well on the Intel architecture. Thank you very much. Then finally, some integrations around IoT, Internet of Things, to enable enterprise customers to have these secure at the edge and deliver that.
Yeah. The work we're doing together to make sure that TensorFlow really runs best on Intel architecture and really have that highly tuned machine. We've committed to releasing the Intel architecture optimization to TensorFlow. The team is committed to getting it done by the end of the year, but just looking at my watch, that's not a lot of time. They may be working through the holidays on this one. No challenge too great for the Google and Intel engineers. I'm sure they'll get that done. If not by December 31st, maybe January 1st. Well, thank you.
I know how you're spending your holidays.
I'll be talking to you.
We have had the two companies, Diane's company and Diane's company.
There you go.
No, we've had this longstanding relationship. It lets us accelerate bringing these new technologies to market much faster, the joint engineering, the joint validation, and efforts over the years, and looking forward as we're developing for next generation Xeon. As we expand our engineering now, I'm really happy to be going to market and expanding our enterprise business relationship, bringing cloud to all companies all over the world.
To all companies on the face. Yep, we're all cloud, and we're happy to help you. It has been great working together. We obviously share a common vision around the democratization of technology through open, and it's a real pleasure to have you here. It's a pleasure to have the strategic alliance, and thank you so much, Diane.
Yes. Thank you, Diane. Great. Thanks.
When we work with big industry partners and thought leaders like Google, they challenge us, and we are better for it. In that same light, as we look at the area of artificial intelligence, the industry is still in a period of discovery, right? The pace of innovation in AI is massive. For us to remain deeply connected and informed of all the new developments occurring, we are creating the Intel Nervana AI Board of Advisors. It's made up of industry-recognized academics and industry leaders, thought leaders. Through the advisement and collaboration of these folks with the Intel Fellows and Intel Principal Engineers, the board is going to shape our future R&D strategy across hardware and software, delivering that Nervana platform. This board is a unique cross-disciplinary group of people that are deeply committed to changing the nature of computation.
If AI happens to be your domain of expertise, you likely know some of these folks, and we're certainly excited to be partnering with them. I personally passionately believe that Intel is best positioned to support artificial intelligence and the revolution that comes with it. I guess you'd be surprised if I said I wasn't passionately a believer about that. If you look as just a starting point, a data point to start that discussion, our estimate is that 2016 Intel processors will power more than 94% of all the servers deployed for artificial intelligence workloads, delivering the complete compute solution, no GPUs, 94%.
However, we know that if we're going to continue to lead and drive in a field like AI that is continually evolving, we need to be matching that pace of innovation, that very rapid pace of innovation that we all see happening in the world of AI today. Our commitment remains that we will run all workloads best on Intel architecture, and that means all methods of AI. It means all implementations, techniques, all algorithms from deep learning and the various forms of neural networking that Naveen was speaking of, to the many associative and statistical algorithms, everything, all methods of AI will run best on Intel architecture. AI on IA, as we say. There are clearly then many different approaches to artificial intelligence, but what they do all have in common is scale. The more data you can compute, the more accurate the model will be.
The more compute capacity you deliver to the solution, the faster the model's going to train, and the more real-time results you're going to get from it. We provide a full portfolio of products that span all artificial intelligent implementations. What we deliver is one architecture from training to scoring, from development to deployment, from general purpose to highly targeted algorithms. Let's start with the Xeon family. The Xeon E5 is the most widely deployed processor for artificial intelligence and machine learning. With each generation of Xeon, we have been optimizing Apache Spark to run on IA best. Apache Spark is the most commonly used machine learning platform today. Over the last three years, we've delivered a staggering 18X improvement in performance. Beyond Xeon, we enable customers to further optimize their AI solutions with the integration of a FPGA, field programmable gate array, with the Xeon processor.
That addition of the FPGA allows for very low latency, which is critically valuable in these real-time inference solutions, it also provides some flexibility in the precision that is used. You get a very efficient, a highly efficient compute environment. I am very happy to announce today that we are shipping Skylake. Skylake is our next generation Xeon processor. It is a preliminary production version of Skylake. It has a targeted feature set targeted to the largest cloud service providers and for their use, as well as some high-performance computing end users. With Skylake, we just continue that beat rate of innovation. One of the features that's included in Skylake that is targeted for artificial intelligence is the new instruction set AVX-512, which is advanced vector acceleration.
It's a new instruction, that floating point feature is going to allow a significant acceleration for inference. All the OEMs today are currently developing full-featured high-performing systems that are across the Skylake product line, delivering the solution for the broad range of customer segments, the broad range of workloads, they'll be ready when Skylake, as the formal portfolio, launches in the middle of next year. Next in the AI portfolio is Intel Xeon Phi. Scalability is key for machine learning, Xeon Phi is inherently a scalable system, a scalable architecture. We recently launched our second generation of Xeon Phi, which was Knights Landing. It delivers 31X reduction in the time to train when scaling to 32 nodes. It's very impressive, near-linear performance scaling.
Xeon Phi delivers a 50% increase in deep learning training over the standard Xeon processor family, a significant pop in performance. It also provides direct access to 400 GB of memory. Huge memory space, which is incredibly important in AI solutions where the size of the data set really matters. If you contrast that 400 GB of memory, the best-in-class GPU today is limited to 16 GB, dramatic difference. Xeon Phi is also a bootable processor, you don't have to deal with the constraints and complexities that come with an offload architecture like a GPU. After that, we come to our next generation of Xeon Phi, which is Knights Mill. I actually announced this product at Intel's Developer Forum a couple of months back. Knights Mill is going to further optimize for deep learning training.
With that processor, we deliver single precision as well as half precision support, that's going to deliver a 4X improvement in deep learning training. Amazing results, we look forward to launching this product next year. Given also that AI solutions are increasingly running on high-performance computing clusters, the scale value proposition, we're very proud of the TOP500 supercomputing results that just came out this week. Xeon Phi comprised over 80% of all the new system accelerator flops. If you look at all the co-processors and GPU flops, we were over 80% of those new flops. We're making tremendous gains and results with the Xeon Phi product line. If you look specifically into deep learning, it is still a small but emerging portion of the overall AI world.
To give you some representation of the size of the market, 0.1% of all servers in 2015 were dedicated to deep learning training. A very small number, but we all know it is a space of immense investment and innovation. It is a space that is rapidly growing. Intel is committed to delivering the significant advancements that the industry needs. Before the end of the decade, the Intel Nervana platform for deep learning is going to deliver a staggering 100X increase in performance compared to today's highest performing GPU-based solution, 100X. That's our commitment by 2020. I'm looking at Naveen, that's your commitment by 2020. He's acting like he's just hearing this for the first time. Our first step towards this is to actually launch the new product that we've code-named Lake Crest.
Lake Crest is based on that great technology that we acquired from the Nervana company. First silicon of Lake Crest comes in the first half of 2017, just around the corner. Lake Crest will deliver the best-in-class neural network performance at launch. I'm also happy to announce that in addition, we have added to our roadmap Knights Crest. Knights Crest will create a very tightly integrated solution between our best-in-class Xeon processor with our best-in-class deep learning engine. Those two come together into an integrated solution. Aside from silicon, as BK noted, we have other AI assets, a critical one being Intel's Saffron technology. We acquired Saffron company last year, and their CEO, Gayle Sheppard, she continues to lead Saffron. She's right here for you, so if you want to chat with her afterwards, you'll know who she is.
Saffron's software platform enables companies to identify critical insights in a very timely and effective and efficient manner, it does that employing a memory-based reasoning system. A great example of the impact that artificial intelligence and Saffron has had in a particular business result is USAA. USAA, if you don't know, they're a financial institution. They provide insurance, banking, investments, retirement products, and advice. They are owned by their nearly 12 million members, and those members are exclusively current and former U.S. military, and their families. Running the data science research and development of USAA is Robert Welborn. His areas of focus are on machine learning and high-velocity data problems. Please join me in introducing Robert to the stage. Robert. Thank you for coming.
Thanks, Ann. Thanks for having me.
Great to see you. Yeah. As a financial services firm, you are obviously swimming in data. It's a very data-rich environment.
Very much so.
Why don't you jump right in and tell us about some of the AI implementations you've done, obviously along with Saffron. What was the problem you were trying to solve?
Sure. At USAA, we are obviously very focused on ensuring the safety of our members' assets and on spending our members', our members own our company, spending their money very wisely and not wasting any of it. We spend a lot of time trying to understand and detect fraud.
If you think about it, in the U.S. in the last 12 months, auto insurance companies have paid off nearly $1 trillion in auto insurance claims. Industry estimates put 10%-20% of that $1 trillion as being fraudulent claims.
Wow.
USAA pays $billions every year in claims to completely valid members, and we do our best to make sure that we don't spend $0.01 on fraud. However, no system's perfect. We first took out the investment in Saffron with the intent of improving our rigorous fraud investigations and looking at your memory-based reasoning systems to augment our current workflow to drive complexity out of the system, as opposed to adding additional business rules and models. Within 10 hours of setting the first instance up, we found our first fraud ring.
Wow.
This ring was bilking our members out of $2 million every year.
Wow.
We were extremely pleased to be stopping that.
Wow. That is a quick result.
It was. 10 hours was good. If you think about it, part of the thing is we would never have found that had it not been for Saffron's unique holistic graph that essentially looks at the associations between extremely heterogeneous data sets.
It is very impressive, an attribute of Saffron is that you are continually getting data real-time, so that data is continually being refreshed, which is critically important when you are in the process of continually trying to outsmart the malicious hackers, right?
Exactly.
That real-time data and refreshing of data, your model never ages out. It is a great attribute of Saffron, congratulations.
Thank you
On your fabulous results. I heard you also leveraged the technology not just for fraud detection, but also for customer service or customer experience?
Absolutely. At USAA, we're consistently member service champs. We turned to Saffron to understand why is it that members are calling us, logging onto the website, or connecting to the mobile app. Just to give you an idea, we have 160 products, we have a lot of financial services products, and we offer 8,000 distinct services that are attached to that. That's like 160 factorial times 8,000 factorial different-
A lot.
Yeah. It's more than any possible logistic regression's ever going to be able to do. To give you an idea, if I hired 1,000 developers right now, they started building a model every second, we would outlast the life of the universe by about 3X. Rather than waiting for a second universe to cool and-
Right. You thought you'd use Saffron, right?
Yeah. We thought in about 10 weeks, we got our first set of results, which was much better-
Wow
than seeing whether or not humanity evolved again.
Right.
With our very first iteration of models, which is actually still impressive, we're still hitting about a 50% rate of guessing what people were going to do next.
Saffron improved the rate to 88%.
Wow.
It was a 70% improvement in 10 weeks.
Wow, that's pretty impressive.
Essentially, we've used Saffron so far to automate some of the current analysis that we already do and considerably step it up, as well as taking some of the superhuman analysis that mere mortals could not possibly do.
Just to emphasize, fraud detection, you got, checked that box, and customer satisfaction checked. What's next? What are you looking to do in the AI space next?
We're looking to tackle cybersecurity and phishing next.
Not-
Not the No trout.
No. With a PH.
PH.
PH. No trout. Okay. Thank you so much for joining us.
Thank you so much.
Very impressive. Appreciate it, Robert Scott. AI, also, as BK talked about, in the Hack Harassment space, AI plays a critical role in delivering societal benefit, we clearly are passionate about that. I'm personally passionate and fortunate to be working in a space where I can actually have impact in an area of societal benefit, which is the marrying of technology to the healthcare industry. Many of you have heard before in various talks that we've done about the investments we're making with Michael J. Fox Foundation and the use of patient analytics for better understanding of Parkinson's disease. We also have our mission around the collaborative cancer cloud, looking to deliver personalized cancer treatment through distributed analytics of genome sequences, working with OHSU and other partners. We're also working with the National Cancer Institute and the Department of Energy to apply high-performance computing for drug discovery.
We're also working with Penn Med on machine learning solutions that identify patients that are at high risk for hospital readmission. I am also happy today to be announcing a new partnership with the Broad Institute of MIT and Harvard. Broad, if you didn't know, they hold the world's largest number of full genome sequences, so worldwide, and our objective together is to optimize the hardware and software solution for genome analytics. The goal here is clearly to create new solutions that will promote biomedical discoveries. Let's hear from Anthony, their Chief Data Officer of Broad.
At the Broad Institute of MIT and Harvard, we use a collaborative approach to biomedical and genomic research in order to improve human health and advance the understanding and treatment of human disease. Every eight months, the size of genomic data sets doubles. This makes scalability a big problem. The data is often siloed by institution or infrastructure. This makes sharing very challenging. When you look at the next decade or so of both genomics and healthcare at large, we're going to see that the great breakthroughs are going to come from applying artificial intelligence to problems in healthcare. I am truly thrilled to announce today the creation of a new Intel-Broad Center for Genomic Data Engineering. This represents a five-year investment towards developing tools for processing genomic data, along with reference architectures for deploying them across multiple institutions.
We hope to create a model for other industries to break down barriers and speed up research and discovery that relies on complex and distributed data sets. Broad and Intel have already done amazing work together. For example, in a recent project, we reduced the time it took to genotype a 20,000 sample cohort to less than one tenth of the time it took previously. Our continued work together has the potential to help researchers everywhere drive new insights that ultimately can improve human health. I want to sincerely thank Intel for its partnership and support of this groundbreaking work. We look forward to a deep and productive collaboration in the years ahead.
Not often discussed as another area that's critical for the application of technology, and in particular artificial intelligence, is in the trafficking of children, a serious issue facing our country. In 2015 alone, over 460,000 children were reported missing in the U.S. The National Center for Missing & Exploited Children, or NCMEC, was established by Congress to address this horrific reality. With me to talk about our work together to apply artificial intelligence to this space is the CEO of NCMEC, John Clark. John? Hi.
Thank you, Diane.
Thank you for coming. I really appreciate you being here with us. Why don't you tell us a little bit about NCMEC?
Thank you, Diane. Good afternoon, everyone. It's a pleasure and honor to be here. I was listening in on the wings of all the great things that Intel is doing to make the world a better place with technology. The National Center for Missing and Exploited Children is a small nonprofit organization based in Alexandria, Virginia. Every day, about 330 dedicated souls come to work every day with a single mission to try to find missing children and stop child exploitation. The organization was established in 1984 by John and Revé Walsh, our co-founders, John Walsh from "America's Most Wanted" fame, he really designed it as a clearinghouse for families and victims and other folks affected by the issues of missing children and child exploitation. Next month will actually mark my first year on the job as CEO. Very proud of that.
I had the good fortune of working with another historic and legendary organization, the US Marshals Service, prior to coming to this organization. Through that relationship, I watched the center really develop and grow into this great single mission here of finding children and stopping exploitation. Today, we've recovered over 230,000 missing children. We're very proud of that. In fact, more children are coming home today than ever before. Again, hard work and dedication, but that's what gets it done.
It is remarkable. Technology has played a substantial role both on the negative, unfortunately, and then, of course, the positive. On the negative, making it easier to exploit children, also using technology to help rescue them. Maybe you could tell us a little bit about how technology's played in and what you're seeing.
Sure. Well, threats against children have evolved over the years, the internet's paved the way really for the explosion of what we're seeing now in child pornography, online exploitation, somewhat benign terms really for actual violent sexual exploitation of children. Again, hard to believe, but that happens. In response, the National Center for Missing & Exploited Children created a CyberTipline, which began in 1998. People would have a place to actually report these types of things. In a recent FBI operation conducted just a couple of months ago, they recovered about 82 children involved in the online sex trafficking situation. The youngest child there being actually just 13 years old. In 2013, the volume of our tip line increased to half a million reports, and right now, with just another about a month to go, we're at our 7 million report mark.
Wow.
That's a significant increase in that volume on the CyberTipline. In order for the National Center to respond faster and be better prepared, we obviously need to leverage technology, and that's one of the great things we're doing with Intel.
Yeah, that is a dramatic uptick in reports. Certainly, the cloud service providers have become a large supporter as they are voluntarily taking action to ensure that they are not hosting and promoting child pornography. That's a big driver of some of that uptick. What is your biggest challenge then with all these new tips coming in? What's your biggest challenge?
Well, despite advances in technology, we still, rather sadly, I would say, in our modern era, have a lot of stovepipes and a lot of manual use of that kind of analysis on the material we're getting through the CyberTipline. It's still kind of a cumbersome process that we're using. Analysts have to personally review all the data, and you just imagine with that volume, what that would take. Artificial intelligence and specifically, Intel's innovation, in this particular work area are key to address this issue of going from a manual system really to one that's using technology to enable it. That's going to be ongoing here over the course of time. By leveraging this particular artificial intelligence, our goal is to more quickly and efficiently process these reports.
An interesting thing that Intel's helped us do or will be helping us do is to decrease the amount of time it takes for us, which right now in some cases may take up to 30 days to actually go through that kind of volume. With Intel's help, in many of those cases, we'll be able to get it down to a day or two. Just imagine that dramatic change.
Response, yeah. Amazing response. Yeah. Our commitment through our partnership is that we will achieve this reduction from tip to action, tipping reported to action being taken, whatever that action is, from 30 days down to one day, and we will do that in less than a year. That's our commitment to you. We're happy to be working on that very important project. Why don't you tell us what additional opportunities should we be working on or what more can be done? What are you looking at?
Well, today, about 94% of the reports we get on the CyberTipline are originating outside the U.S. We think with artificial intelligence, we can enhance our global response, being able to triage those types of reports, particularly of issues involving child pornography and child sex trafficking. Artificial intelligence is going to help us really find what we might call the needles in the pile of needles.
Yeah.
Process, we believe, up to 20 times faster. That data, again, that we're always looking at to help us protect children even in a better manner. At the National Center, we're also inspired by the work that's being done by others in this field, but Intel is certainly taking the lead.
Thank you so much for being here. Thank you for the partnership, and we certainly realize what an important investment this is and what a great application of artificial intelligence, and we're thrilled to be working with you. Thank you.
It's a great honor. Thank you as well for all the help that Intel's doing.
Thank you for coming.
Thank you.
Realizing the potential of artificial intelligence, I hope you now clearly agree, is a statement of when, not if. We are investing here at Intel to deliver breakthrough performance, to deliver societal benefits, and to democratize the access. We are also investing to make you all look great. When you leave today, don't forget to pick up your hoodie. Very cool hoodie here. Specifically now to talk more about the efforts we are making in the democratization of technology and making AI solutions more available, let me introduce to you Doug Fisher, Senior Vice President of Software and Services Group. Thank you.
Thank you, Diane. Good afternoon. Being a software guy, I should be wearing that hoodie up here. Today, I'm going to talk to you a lot about what we're doing to democratize artificial intelligence. As I was thinking about what I was going to speak about, a thread came through, a theme that came shining through on what I was going to talk about, and it's called upstreaming. What we're doing in artificial intelligence in my organization is upstreaming the work we do. How many people know what upstreaming is? Have a definition? How many people are from a big city like New York, L.A., San Francisco? Yeah, it's not that same upstreaming. It's not the upstreaming where you're standing on a curb and there's three people waiting for the same cab, and you go up one block to get the cab before they do.
That's not the upstreaming I'm talking about. What I'm talking about is getting closer to the source, engaging a lot earlier. This sounds a lot like the work we've done for years in open source software. Being applied that same technique in artificial intelligence, moving upstream. When you move upstream, you concentrate the effort, you accelerate the innovation, then the rest of the world benefits. Making it pervasive, or as I say, democratizing it, making artificial intelligence available for more people to innovate and build solutions like Diane just spoke about. That's what we're doing. In order to do that, we have to unleash the value. Diane talked a lot about the great hardware platforms we're building. We need to unleash that value. We need to make sure that whatever we build, whatever solutions are being built, the value comes through. We also need to harness that knowledge.
We need to make certain that the next generation of innovators have that knowledge to scale those solutions to the broader market. In order to unleash the value of artificial intelligence, in order to unleash the full potential of artificial intelligence, we are unleashing the value of IA. That is what we're doing. Let me talk about that. Diane showed the roadmap of the amazing hardware platforms from Xeon Phi, to the new Lake Crest she talked about, to our FPGAs. All that marvelous hardware platforms purposely designed for specific AI workloads. We need to build capabilities on top of that. She talked about how prevalent we are in machine learning, and we have libraries of primitives that we build in software to take full advantage of that hardware. These primitives accelerate the capabilities on our platform.
Building these primitives accelerate the algorithms on top of that. We're extending our Math Kernel Library to add deep neural network capabilities to accelerate those primitives, to ensure that whatever's built on top of that is taking full advantage of whatever Intel platform is below that. This is a value we can add to accelerate the performance that you so need to solve all those problems just discussed. On top of that will be the deep learning frameworks, and those deep learning frameworks need to be optimized and accelerated and take full advantage of those deep neural network primitives. We'll engage with the most prevalent and important frameworks to our industry to ensure they're fully optimized, so that when a data scientist engages and that framework utilizing Intel architecture and Intel platforms, they're getting the most value.
We're unleashing that value, and they're getting the most benefit out of that platform. That's why we engage in all these frameworks. Let me talk about a few we engage with. The first one you've heard about, Nervana. Nervana, we're going to engage with to ensure that we optimize that framework for our platform, ensuring the performance of that is second to none. We're using it as a space where we can drive more and more innovation for the market. An example of that is our Intel Nervana Graph Compiler. If you're a data scientist, the complications of going deeper and deeper into topologies and more and more layers and hand-tuning that is very difficult for data scientists to do.
With this graph compiler, it helps simplify the ability to have more and more complex models, deeper and deeper layers, so they can build more sophisticated models. We're innovating in this space with this graph compiler, and we're going to open source that. It's going to be released the first of this year, 2017, in neon's release 3.0. It'll also be open source for other frameworks to take advantage of it. We're going to accelerate technologies in neon and allow the rest of the world to benefit as well in all the other frameworks. We're going to work with those other frameworks to ensure we take those technologies and infuse them in multiple frameworks. TensorFlow, you heard both Dianes talk about TensorFlow and the work we're doing there to drive the performance optimization in TensorFlow.
It's a key pillar to the collaboration alliance we just talked about. On stage a moment ago is that optimization that we committed that I'll be working over the Christmas holiday to get done. It's always good to go first. We're going to ensure that we deliver all those optimizations at the end of the year and then deliver this to the masses to take advantage of in 2017, tuning and optimizing on our Xeon and Xeon Phi platforms. The early work that they talked about is already showing fruits of our labors. We're already seeing tremendous performance improvements in engaging with Google and optimizing those frameworks, taking advantage of those software layers that I described. Another framework is Theano. Theano we're optimizing very similar to the other frameworks.
A specific example where we engaged with Kyoto University out of Japan, they're looking to figure out a way to use machine learning and deep learning to determine what drugs will be most effective for solving challenges they have in their society. They analyze that and predict the drugs and the combination of drugs that will be most effective. By engaging with them, we were able to improve their performance on our Xeon platform and the optimizations we've done in the framework by 8x, increasing their accuracy of their model to 98.1%. That's an amazing result, taking advantage of the capabilities we have in our platform, the optimizations we have in our software, and delivering value in their model. Caffe is another example where we worked with a company in China called LeCloud.
LeCloud is the largest online video provider in China, and their job is to consistently look for illegal video. They take pictures and run it against a model, and using deep learning, they analyze the pictures to determine if that video being displayed is illegal. What they do is then pull it off, and they were having trouble meeting the demands of all the video. As you know, the proliferation of videos is growing dramatically. We went in, and using our Xeon platform, we're able to increase performance by 30x when we optimize the Caffe framework for our platform. This is the kind of work we're doing to unleash that value of Intel architecture, engaging with these frameworks, and really delivering value and performance on our platform. The final one I want to talk about is Torch.
Torch is another framework where we engaged with a company that started in 2015. I met this company at IDF. Company's called Pikazo, and they're doing this very innovative thing with their technology. We started engaging with them. I participated in this application that they developed. When I learned what was behind it was so fascinating to me, I wanted you all to hear about what they're doing. Instead of me talking about it, I thought I'd invite the CEO of Pikazo up on stage to talk about what he's doing, what his application's doing, and how he used Intel to deliver a better experience. Let me invite up on stage Noah Rosenberg, CEO of Pikazo. Hey, Noah, how you doing?
Thanks, Doug.
We talked earlier, we talked yesterday about this. I met you at IDF. I'm very fascinated with what you're doing. It does an amazing thing, which turns individuals like me into artists with a couple of clicks. Why don't you talk about what you're doing?
That's great, thanks for. I think that's a great way of describing it. Pikazo's an app that lets anybody create art with just two taps. Maybe instead of talking about it, I thought we could show it. Doug, do you mind if we-
Okay.
-use the app on you?
All right.
All we're going to do is take a quick photo.
All right.
Now we ship Pikazo with hundreds of styles, but the coolest thing is that you can bring in any art. I've been staring at this background all day. I thought, "Wouldn't it be cool if we could paint you with this background?" I'm just going to take a photo of that. That's what's so cool about Pikazo is that it lets you become an artist by combining art that you see in the world around you.
Now you're actually compiling that.
Send that in. There. Yep.
Here's some examples of what you've done before. Why don't you talk a little bit more-
Yeah.
-about what's behind what we see on the screen here.
Yeah. In this example, what Pikazo does is, you can see how we've given it an image of the Golden Gate Bridge. Our neural network, powered by AI, running on the Torch open source libraries, is going to study that image, and it can separate out and discover what area here is the bridge, what area here is the water? What are the clouds? It does it by modeling the visual cortex. It's really looking at it the same way that you or I would. Then we ask it to look at the style. In this case, it's a Picasso painting. When we're looking at the style, we say, "Where are the brush strokes? Where are the colors?" While the neural network is sort of imagining or detecting those two images, it's able to mesh them.
We ask it, just make these pictures kind of similar. Now take a look. Is that better? Each time, the neural network re-detects it, and it goes through and it says, "Okay, this is a little bit more like both." It processes that image over thousands of iterations. It takes a ton of computation. In the end, you get this beautiful artwork.
You engaged with Intel. You already had a solution before that. You engaged with Intel, why? What was the challenge that you were coming? Go ahead.
Yeah. When we first got started, we launched exactly a year ago today. At the beginning, it took us an hour to render a one megapixel image because this is just such a new idea, and it just requires so much computation. We thought, gosh, we can't wait to let people become artists. We released it in the world, and we just let you do a little thumbnail image. People went crazy. They got addicted to this thing, and they were running it all day long. Our challenge became, people kept saying, "How do I get my art out of the phone? I want it on my walls." For this to be really art, it needs to get out into the world.
We were looking at the problem, we said, we're running on these GPGPU systems, and the challenge there is for our little thumbnail images, those required 16 GB of RAM. We were really trying to struggle about how are we going to get this bigger? We knew that on the CPU area, if we could get our Torch library optimized, that's where Intel came in. Your guys came in and helped work with our engineers to optimize us for the MKL framework so that the Intel platform could be processing these images. We went from a 1 megapixel image in an hour to now we can do a 12 megapixel image in just 5 minutes. Anybody can do that right from their phone on our free app.
That's fantastic. Maybe we have some other examples of work you did while you were waiting for the keynote. I can't stop. You're welcome to have your own image, Diane. BK, if you're here. This is fantastic. Really what you had is a problem that required exactly what Diane just talked about, is that limitation to 16 GB. You needed 2 TB of data. It went from an hour to a small photo to actually canvases in minutes.
Yeah. Absolutely.
That's fantastic.
We're excited today to be able to announce that we can now go to unlimited resolution. We've got an image that's 16K by 16K pixels thanks to the help from Intel.
That's great. How did I come out?
Let's take a look.
Blank. All right. The screen's not working.
Yeah, it takes a bit to load up.
You get the idea. I'm up there.
Okay.
All right. You get the idea. If you go back, you can see what you did combining my photo with this image.
There it goes.
Very cool stuff he's doing. What's really interesting is the computation behind that and the science behind building that algorithm and utilizing our performance capabilities. Thank you so much.
Thank you so much.
Thanks for being here. In addition to the frameworks I've talked about, Diane also talked about the importance of all the data, the massive data sets that are coming at us, and our need to continue to optimize Hadoop and Spark for in-memory computation. She talked about the 18X performance improvement we've already shown by optimizing Spark. We're continuing to do things in Spark, where we're announcing the availability at the end of 2016 of BigDL, which is integrating in deep learning framework techniques right into Spark so that you can have the same developers who are utilizing Spark now have deep learning capabilities built into that. The same programmers can take advantage of the environment they're already in. You heard earlier, this is an evolving space. All sorts of new innovations are happening. That's an example of what's happening today, all these new innovations around this space.
Once the data scientist has decided which framework they're going to use, we want to make sure that we provide them that framework on our optimized framework, optimized MKL-DNN, and our amazing hardware platforms, and then get out of the way and let them do what they do great. That's build these models. What we've built is our Intel Deep Learning SDK that I talked about at IDF. It will now be in beta next month and release the first of next year in full gold production. What this does is allows a data scientist to deploy their environment, help them build the models, accelerate their ability to do what they do best, and build the models and train that model.
It then takes that trained model and then adds optimization techniques as well as weight quantification to compress it, to allow it to be deployed for inference on our broad set of platforms. This tool chain goes from end to end across all of our architectures that you heard Diane talk about today, allowing us to give those tools that the data scientists so need to accelerate their ability to innovate. That's what we're doing to unleash, to upstreaming all this knowledge, upstreaming the work we're doing in frameworks. All open source. We're working upstream at the source of these investments. All of this is happening upstream. We also have to educate upstream. We can't just do all this work ourselves. We have to drive the education of what's going on. We have to drive knowledge upstream.
That's what I'm so proud today to announce, as Brian talked about, the Nervana brand is going to be across a lot of our assets, the Intel Nervana AI Academy. This is going to be a one-stop shop for all artificial intelligence developers and educators to participate. This is going to be a framework around we're putting programs to engage and educate and then scale. It consists of three main pillars, the first pillar being the Intel Developer Zone for Artificial Intelligence. Those of you who know our developer zone, we already touch over 20 million developers. We have over 7,000 ISVs. We're rated as number 2 in quality of development environment by Evans Data. It's a quality scale platform. We're going to take that scale and utilize it to engage with the artificial intelligence developers.
This is where they come to get those optimized frameworks, the tools, the libraries. Everything I talked about, this is where they come to get that information and then participate and engage in that community. We're also going to engage with innovators and individual software developers who show a passion for this environment, show a passion to solve problems and advance this capability, and we call these software innovators. We find out who they are, we engage with them, we train them, we give them access to tools, access to training, and then allow them to have a platform to communicate what they're doing. We get them into events. We allow them to demonstrate their capabilities. This helps us drive knowledge upstream so they can downstream that to a broader set of people.
If you become very good through our crowdsourcing capabilities, a software innovator can then become a black belt. That's who the community recognizes, not only an expert, but also is doing their part in growing and strengthening that community. They're recognized by their peers as a black belt. You can find all this on the website below and get engaged there with software.intel.com/ai. That's where all this community is going to be housed. We want to go further upstream from the current developers and focus on not only developers of today, but developers of tomorrow, because they're the data scientists of the future.
We're going upstream to students as well and focusing on them, and that's why we started a new program called the Intel Student Ambassadors program, where we identify, support, and recognize high-achieving students in academia who are focused on this area, who are passionate about it, who are advancing this technology and want to drive it and advance it and educate others. This is the community that's going to be the next generation of developers that are going to drive those capabilities on our Intel platform, our tools, and our capabilities to the market. With that, I'm very pleased to announce our first student Intel ambassador is Dan Iter. Come on up, Dan. Dan's from Stanford. It wasn't by design I pulled a Stanford guy up here. He was selected as our first student. Tell us about the program.
Yeah, definitely. I'm Dan Iter. I'm a graduate student in computer science at Stanford, and I first got introduced to the student ambassador program for my research. We were doing collaboration with Intel for parallel training of deep networks, and I thought it was a great opportunity to get early access to optimized tools, because for me, as a student, it's really important to be able to quickly prototype and run experiments. Actually, my favorite part of research is being able to talk to other students, share ideas, and finding real-world problems that can be solved with new technology. The Intel Student Ambassadors program was a great opportunity for me to get involved with workshops and training sessions on campus, and help build the community around AI.
That's fantastic. You've been engaging with us. How's it been so far?
It's been awesome. I love it. The tools are great. Got to get some access to Nervana, to some of the optimized Caffe and things like that. It's been awesome.
Fantastic. As a recognition for you being the first student ambassador, we'd like to present you with this plaque. Congratulations.
Thank you.
We have here, since you are still a student, this is a backpack with a hoodie. A lifetime supply of Top Ramen, I think is in there for you.
Thank you. Thanks a lot.
Obviously, we're going upstream to the top universities around the world, identifying people like Dan to help be ambassadors and scale the knowledge and really show that passion and put it into action. We're going a little further upstream. Stanford also has a program called SAILORS, which is Stanford's Artificial Intelligence Laboratory. The idea behind this is to reach a diverse and inclusive environment. As you well know, Intel is very passionate about diversity and inclusion, and this is an opportunity for us to reach underrepresented people in technology, in this case, technical females. We're reaching back into the 10th grade and finding technical females who want to take their passion and move into the technology environment in their education, but they don't have exposure to it.
They go through a two-week camp where we give them mentorship, exposure to these capabilities, and train them on what is possible to really infuse that into them so they have that passion to continue that forward in their studies as they go to university. We're going as far upstream as we can to get this knowledge built in, grown, and propagated to as many people as possible. That's not enough. We have to now take all the information like Dan and all these developers are providing and building. We want to have webinars, workshops, and then meetups. The workshops are designed to train as many people as we can on our capabilities.
The webinars are designed to propagate that information through coursework, as well as the meetups, which we're having one November 29th, right here in Santa Clara, where all these data scientist developers can meet up and talk about what they're doing and share information and collaborate and accelerate their effort by having people with similar interests in getting their solution built. Again, we don't think that's sufficient. That's awesome that we're doing that. We want to scale even further. We want to reach more and more people and educate more and more people with all the things we're doing and the great innovations we have at Intel. That's why I'm really pleased to announce our new partnership with Coursera. I don't know if you guys know who Coursera is. Coursera is the largest online educator in the world.
They have over 23 million students, or they refer to them as learners, thousands of courses. We're partnering with them to take all this knowledge upstream and drive it downstream through this massive opportunity to reach many, many people. I'd like to invite up the COO of Coursera, and some of you at Intel would know her, Lila Ibrahim, to talk about what we're doing.
Hey, Doug.
Good to see you again.
Yeah.
It's been a while.
Yeah.
We used to work together a few years ago.
I still have my Intel badge.
Welcome back.
Thank you.
Let's talk about what we're doing together. Talk about Coursera first, and let's talk about what we're doing together.
Great. Well, let me actually, if you don't mind, I'll start off with a quick background. I spent 18 years at Intel. I actually first joined as an intern, so I'm really excited to see everything that Intel is continuing to commit for education. I was a design engineer, an electrical engineer on the Pentium processor way back when. Through my time at Intel, I had a chance to continuously reskill and upskill, and I found that my career advanced, not only in the technical field, but then into sales, working with developers, and eventually into management. There were two things that I really learned. One is the opportunity that opens up when you have education and continuous learning about what's happening around you in the world today. The second, through all my work with the developers, was the power of the Intel developer and ecosystem.
I'm especially excited to be able to have my two worlds collide today.
Great.
What Coursera is, Doug gave a great intro for us. We are a tech-based education platform, the world's largest. We were founded by two professors from Stanford to kind of keep that Stanford theme going along, two Stanford AI professors, actually. Our model is basically universal access to the world's best education, to democratize education. We do that working traditionally with top universities from around the world, over 145. If you think of the top CS types of schools, Georgia Tech, Michigan, U of I, Princeton, et cetera. What we've recently done is we've recently just started working with companies, and we're excited because Intel is one of the first companies we're working with and the first one that we're working with on AI.
That's fantastic. Our partnership continues.
Yeah.
You start to scale. You're going to democratize, which is absolutely what we need to do. How many people you think you're going to reach?
Well, we'll see. First, we're starting with our AI course, a series of AI courses together, these will be put together in something we call the specialization, which you'll see at the end of Q1 next year, pretty soon. We will use that as a foundation on moving forward. If you think about unbundling a degree and then rebundling the key assets, that's what we'll be working with together with Intel, we expect the AI course to just be the first. What we're super excited about, too, is our learners care a lot about career advancement, the value of the Intel certificate on the Coursera platform, I think, will really help people.
Intel, in fact, is going to take the top 10% of the learners from the course that Intel will be developing with their expertise and our understanding of online learning at scale and be providing resume reviews for internship opportunities at Intel sites around the world. This is a first.
Fabulous. Okay. Well, thank you so much for your partnership, Lila.
Thank you.
Great to see you again.
Um-
And-
Sorry, one other thing. Diane mentioned.
The Bob Noyce quote.
Yes.
Having spent so much time at Intel, I'm reminded of my favorite Bob Noyce quote, which is, "Don't be encumbered by history, go off and do something wonderful." Sitting in the audience today, I'm really struck by that because I think together we have a real opportunity to unleash the power and the potential of the developer network around AI. Thank you.
Very well said.
Thanks.
Thank you so much. Not only do you need to educate, that's critically important. Build all these great tools, great platforms, educate everybody how to use it. What we also need to do is solve. We need to put this acumen to the test. We really need to solve real-world problems. To do that, we need to harness all of those developers out there, all those data scientists out there, and there's no better way of doing this than to have a competition. To harness those minds, you put a competition, you throw a few dollars at somebody, and you put a competition out there. You'd be amazed at what comes out. We're very pleased to announce that we're going to participate with Kaggle, which is a great program that pulls 600,000 data scientists into one community to help solve real-world problems. They're helping solve business problems.
Examples of ones they've worked on is Airbnb trying to figure out how to market to first-time reservists. Walmart trying to figure out how weather dictates what type of products they put into their store, as well as Facebook using it to actually source resumes for jobs. It also solves more important societal problems, as Diane described, or social economic challenges, whether it's endangered species, whether it's looking at rainfall, or whether it's trying to help people with epilepsy, using the data scientist, uploading data to them, and then challenging them to build models that help solve these real-world problems. I'm super pleased to talk about our new partnership with MobileODT. The fourth leading cancer in our world today is cervical cancer. That's the fourth leading cancer. It hits developing countries dramatically. 84% is in developing countries. It's a very curable cancer.
What we've done is teamed up with them. They have a mobile device that does soft tissue imaging. We're going to have those data uploaded, providing our Xeon platforms and all the tools I talked about, and then challenging data scientists to put a model together to help people detect and determine if they have cancer and what stage it's at and what the best treatment is. This is another example of how we're melding technology with health challenges and delivering better value to our world. I talked about the three pillars. I talked about the Developer Zone, the community where you engage, where you get access to all the tools and capabilities, where you become innovators and black belts.
I talked about how we're upstreaming and engaging with students, both in university as well as high school, to ensure that the next generation developers are accelerating their knowledge in this space. I talked about how we're scaling that education and how we're reaching the masses, how we're challenging the community to solve real-world problems, taking advantage of that technology we're unleashing. We're going even further upstream and continuing our research with the top universities in the world around machine learning. We're engaging them on specific areas where we think the advancement needs to continue. One of those areas you heard about earlier was security. We need to ensure that we look at security across a broad set of things, we're working with this set of universities on securing workloads.
A lot of these workloads and datasets are very, very confidential, and they want to figure out how to ensure that those will remain secure. We also want to use it for security analytics. You heard about it earlier with the USAA gentleman talking about using it to figure out fraud and other things. We want to make sure that whatever algorithms we build, they themselves are secure. We have various work streams with universities to ensure that we actually look at the future challenges so that we can continue to democratize and drive artificial intelligence to the masses. As I said in the beginning, to unleash the full potential of artificial intelligence, we need to unleash the value of Intel architecture, and that's what we're doing. Thank you very much.
With that, I'd like to bring up Doug Davis, Senior Vice President of Internet of Things, to talk about everything we are doing to solve real-world problems. Thanks, Doug.
That wasn't my original title slide, Doug. Doug obviously took that tagline to heart that said, "Everyone can be an artist." You obviously have a lot of time on your hands, Doug, see me afterwards because I have some projects for you. Thanks. I really have the opportunity now to take everything that you've heard this afternoon and hopefully pull it all together. I want to continue to work to convince you that the breadth of capabilities and technologies that Intel is developing really gives us the capability to be the leader in artificial intelligence for all the different kinds of environments that we've been talking about today. I think we all easily agree that artificial intelligence is the next big computing opportunity, right?
It's happening at a scale that will really transform society in much of the way that we've seen other transformations happen, the Industrial Revolution, the Information Revolution, and all of the changes that we've seen that came about as we went through those stages. I really believe artificial intelligence has the capability to be able to do that. I want to give you some examples as I talk over the next few minutes about what those look like. At the same time, artificial intelligence has been on the verge of a breakthrough for what? The last few decades, right? What's different? What's going to enable it to happen now? I contend that all of these transformations really happen when there's enough economic value across the board for them to really become mainstream, right?
There are a number of things that I'll contend are beginning to come together that makes this possible. The first is sensing, right? The ability to gather data through sensors. The cost of sensors has gone down 2X in the last 10 years. The cost of connectivity has gone down about 10X in the last 10 years. Thanks to Moore's law, the cost of computing has come down 60X in the last 10 years. This has really placed us at that economic threshold that we can pass through that will allow this kind of technology and the efficient deployment of the things that you've been hearing about today to become widespread and to make artificial intelligence a widespread capability. This will enable us to take and create solutions, to really accelerate solutions to problems that in the past would have taken hours.
We heard a few examples of that earlier, or days or weeks or months, right? I want to walk you through a couple of examples just to continue to illustrate this. One is in the way that fans view sports and the way that athletes train for and participate in sporting events. Right? Teams are now collecting data through clothing and equipment, and they're using that data through all different kinds of sources to compete better on the field, to be able to predict injuries, and to give us, the fans, a lot more statistics because we need more statistics when we're watching sports, right? We even see that NFL teams are now using 360-degree high-definition videos from a startup company.
One of the startup companies is called STRIVR Labs, they take this high-definition video, they capture images from different aspects in the field, offense and defense. It gives players then the ability to walk through scenarios and to experience them in a very lifelike manner. It helps them then to be able to respond to those scenarios when they're in a live game situation. This is designed by a former Stanford kicker Derek Belch. I think it's really just a great example of how artificial intelligence is just beginning to make its way into sports. Another one from the world that I have heavy involvement in is in the industrial space.
We're starting to see applications that are evolving because as technology is moving along so rapidly, our ability to aggregate and store and utilize vast amounts of data is now giving us the capability to collect almost infinite amounts of data to continually improve the quality or the capabilities of different products. It's also giving us the ability to create efficiencies through self-learning. Right? Taking actions based on a complex set of interactions within an industrial environment. The ability to bring machine responses to bear in solving problems that are very, very similar to how a human might respond to those issues. You heard earlier that we're moving a step beyond now data analytics, it provides a capability for factories to begin to make decisions without any human interaction at all.
As a result, it's going to make factories more efficient, pretty obvious, but also much safer as well. The other area that I think we'll see artificial intelligence really applied is to take kind of tedious or maybe even dangerous kinds of tasks that we have to do. Maybe think about firefighting or agriculture or mining, to be able to enable them to become automated. We're already seeing this today in large farming enterprises. Right? The planting of seeds all the way through harvesting of crops is done more and more by autonomously driven tractors.
Of course, we're starting to see some of these technologies now as we talk about autonomously driven cars, because I think almost all of us will do one of those tedious and dangerous tasks as we leave here today to go back to work or to go home or to go to the airport. We're going to get in cars, I contend, based on the statistics, that's a tedious and dangerous task that we're all going to do. Let me go into that in just a minute. There's a context that was touched on a little bit earlier around this virtuous cycle, this really began to evolve in early 2015. We started taking the various technologies across Intel and merged those efforts into an end-to-end architecture. Marketing genius, we called it the Intel IoT platform. It gives you that end-to-end capability.
It described the importance of things connecting through the network into the data center or cloud, then using analytics to extract information from all that data and to gain insights that you didn't otherwise have. The importance of having analytics throughout, from the edge to the network to the data center, so we could economically optimize how we do that work. This really put Intel at the forefront of how you create these kinds of end-to-end IoT solutions. Thus, we call it the virtuous cycle. If you also think about artificial intelligence and how you create these solutions, it's the same kind of model. In order to take a thing, connect it back to the data center or cloud, and to create the machine learning, deep learning capabilities to continually make this thing smarter and more autonomous over time.
As a technology leader, you saw this from Brian earlier. Intel's investing in the growth of AI, and we're committed to continue to drive forward and accelerate this transformation. We're going to build intelligence into every device and give those devices the ability to connect. We can have real-time communications over high-speed 5G networks. We'll have the infrastructure and algorithms to really create those true end-to-end AI solutions. Of course, driving some of the partnerships and the ecosystem that you've heard others talk about today, and the industry standards that are going to be essential to really make these things mainstream. I mentioned the autonomous car earlier. We think that's a great example of how these artificial intelligence systems will be deployed, and so we put together a little video to illustrate what that might look like.
For more than a century, the evolution of the automobile has stirred our imagination and ignited our passions. Never in that long history has the industry faced a more important moment of transformation. A transformation that promises to bring a whole new meaning to the phrase, "I wonder what this bad boy's got under the hood." The car of tomorrow will essentially be a data center on wheels, capturing and analyzing terabytes of data collected through hundreds of sensors and put to use by powerful in-vehicle processors. Of course, the technology that will enable fully automated vehicles won't be confined to just the vehicle itself. It will require a diverse array of flexible, sophisticated, and fully integrated solutions, from bumper to bumper inside the car to end to end of the entire automotive ecosystem outside the car.
The result will be a technological wonder on wheels that leverages the intelligent use of data to enable exciting new innovations, make driving safer, and completely redefines the concept of high-performance vehicle.
As I said, we think of automated vehicles as a great example of that virtuous cycle, also a great example of how you would create machine learning and deep learning systems. I got to admit, when I'm talking to friends I start talking about virtuous cycles and machine learning, deep learning as to how we'll create automated vehicles, they start looking at me funny. My favorite comparison is, creating these autonomous vehicles is going to be like teaching a teenager how to drive. Right? We load up this young man or lady with all of the learning and their driving instructor school and us parents are going to teach them about everything they need to know.
We put them behind the wheel, we get in the passenger seat, they start driving, almost immediately, they start experiencing things they hadn't expected, right? New data, new situations. We're going to continue to teach them and train, take that trained model that we've built over the years and infuse it into their brain and make them a better driver over time. Right? It's creating that kind of closed loop system. Of course, you realize that data center moves in the back seat as we get older, right? This is also the kind of solution that we'll create with automated vehicles. The car becomes a thing, right, we're going to do all kinds of real-time monitoring.
We're going to have radar and lidar and camera systems that are going to tell us what's happening around the car at any moment in time. Through high-performance 5G networks, we can overlay high-definition map information so that we know precisely where that vehicle is. We're going to fuse all that data together and provide the computing horsepower needed to apply the trajectory planning to tell the car where it needs to go next. Right? Of course, those cars are going to encounter unusual situations. We're going to collect vast amounts of data from millions of cars that are going to pick up anomalies that we can feed back to the data center and continue to train that model and push it out into the inference that will exist within the vehicle. Over time, those vehicles become better and better drivers.
They become more capable because of this end-to-end connected solution. Of course, we're going to continue to invest to make this possible. We're going to nurture these artificial intelligence capabilities for autonomous vehicles. We recently announced a strategic partnership with BMW and Mobileye to bring BMW's autonomous vehicles into serious production by the year 2021. We're really working together with the best capabilities from BMW, from Intel, and Mobileye to really deliver all of the necessary capabilities to deliver these complete end-to-end solutions that are necessary to make these a mainstream technology. Of course, at Intel, we've always believed the best way to learn is by doing.
We're putting sensors on cars as well and driving them around and gathering vast amounts of data so that we can create those data sets and work together with Doug's team to build the tools and capabilities and optimizations that are going to be necessary for these kind of solutions to be able to become mainstream. We're also working with Intel Labs to explore what the human-machine interface will be like in these vehicles. Right? We'll become passengers in those vehicles, but the way in which we interact with them will change, and we want to be able to have the capabilities to build that in and help us to interoperate with things that are becoming more and more autonomous around us. You might ask the question, why is Intel in the best position to fuel the AI transformation? I think you've heard it today.
We have a rich history of successfully leading computing transformations. We know PCs and servers, high performance computing, supercomputers, networking, and of course, the rise of the cloud. We've been the leader in developing open, flexible computing platforms over the years. We'll deliver the most complete set of machine learning and deep learning frameworks and tools that Doug's talked about, optimized libraries. We're working together with developers, data scientists, and even students to foster innovation and drive scale, but to make this wide set of AI applications more and more mainstream. We're going to engage the ecosystem to upstream all of those technologies, to release elements early so that we can start to drive scale out into the industry. We're going to use our technology leadership to integrate capabilities into our processors.
Diane talked about the exciting products that we have coming that are delivering phenomenal performance, density, and cost advantages in order to be able to support these kinds of workloads and technologies. All made possible by Moore's Law. We have an unrivaled breadth of processor capabilities, both in the Data Center with our Xeon and Xeon Phi products and the kind of acceleration technologies we can deliver there, but also the ability to put Xeon, Core, Atom, and Quark processors into all the different things to create these end-to-end solutions with a common architecture. Delivering then hundreds of millions of devices with Intel architecture technologies to create these end-to-end types of solutions. We can layer on vision IP, memory, storage, software, 5G technologies, and the list goes on.
We're very excited to have acquired the best deep learning talent and technology with Nervana Systems, as you've heard today as well. At Intel, our mission is to invent at the boundaries of technology. There's always work to be done. There are big challenges to overcome, and we'll work through those. Those are natural with any new technology transformation. The same will have to be done to fully unlock the potential of artificial intelligence. We're poised to take AI out of the labs and out of academia to cross over that economic threshold that'll make it mainstream, and to enable the amazing experiences that are possible for every person on Earth.
I want to thank you again for joining us today for the Intel AI Day, I want to give you kind of a sneak preview as to what the rest of the agenda looks like. I'm looking at my watch, 20 after. We're going to take a 10-minute break. It says we're going to start back up at 2:25. Let's make that 2:30 so that we have a 10-minute break, then we're going to start two tracks of panel sessions. I encourage you to get engaged with those. I think we have some very exciting topics, and I think you'll find those very enjoyable. Again, thank you.
We're trying to replicate our thought process by putting in a lot of different rules into the
Ladies and gentlemen, let's get started. Please get to your breakout session and your panel sessions.
This is our problem where
We'll get started right away.
If you can solve this problem.
Thank you very much.
All right, if you all can make your way to find a seat. I'm Kevin Huiskes with the Intel Data Center Group, and I'm excited to be here today to kick off our three exciting panels on artificial intelligence. Our first one up today is on precision medicine, and I'm really excited to introduce our moderator of that panel today, and it's Bob Rogers. Bob Rogers is our chief data scientist in the Data Center Group, and I'm going to turn it over to Bob to let him introduce the panelists today.
Thanks, Kevin. Hi, everyone. It's wonderful to have you here. Would you raise your hand if you thought the first half of the day was really, really interesting? All right. We'll have to ask him why he didn't think it was interesting. Thank you, because I agree, it was fascinating, and I was, like many of you, excited about the announcements that happened earlier. I am the chief data scientist for analytics and artificial intelligence solutions here in the Data Center Group at Intel. I am very pleased to be here to lead this discussion of healthcare luminaries. My background, quickly, I've been in analytics forever. The last 10 years I've spent really working hard to build analytics capabilities and technologies that can bring transformational value to healthcare. These four people are doing just that in their respective organizations.
We have luminaries today from the Mayo Clinic, from Kaiser Permanente, from Cigna, and from Penn Medicine. I'd like to invite them to come out onto the stage right now. If you think about where they're coming from, they've got different perspectives in terms of care delivery, insurance, and of course, combined delivery and insurance. We're going to get a very nice mix of perspectives. The other thing to tell you about here is that we've all decided to stand for this panel-
Thanks
rather than to sit, there's a couple reasons for that. John is leading us here in terms of transformation. He told me this morning that sitting is the new smoking.
It's not a bumper sticker, it's reality.
It's real. He said, "Bob, would you be willing to stand?" I said, "Sure." We all stood. I think that's a great idea. That's right. Anyone who wants to stand, you're welcome to stand. Back there, thank you for supporting us. The other thing is that this ensures that you all the way to the back, can see us, which I think is a nice benefit. I'm going to ask my panelists to introduce themselves and tell us why they're excited about AI in healthcare. Mike, go ahead.
Thanks for that introduction. Really pleased to be here. I spent 16 years in missile defense working for a DoD contractor. Ended up there as a chief data scientist, and spent some time thinking about how we could repurpose these technologies in different industries. When I came across healthcare data, seeing the challenges that clinicians and patients have with converging on diagnoses and then converging on treatment, trying to understand all these really complex variables, really understood that AI methods, building these data models, could give really insightful information to help this convergence on diagnosis and prognosis. Has been really happy to be at Penn for the past two years and building these kinds of solutions.
Very good.
My name is David Holmes, I'm at the Mayo Clinic. I've been at Mayo for just over about 20 years now, there's been quite a transformational change in the big data component out there, where imaging used to be the big data. I did a lot of imaging, we have developed a lot of techniques to deal with imaging and visualize imaging data. This change happened where we started to instrument everything well beyond imaging. We instrumented the people, we instrumented the equipment that we use that was itself collecting data, we started to instrument everything. Data changed dramatically about a decade ago, that's really the path that we're on with the big data.
The real value, I think that AI is going to bring to precision medicine is how do we sift through this really massive amount of very heterogeneous data down to information that's actionable? How can we really turn that into decisions, and how can we do it in real time? Because more and more, we need to make real-time decisions in healthcare to kind of force change as quick as possible. That's why I'm excited about AI in precision medicine.
Great, thanks. John.
I've had the privilege of working at Kaiser Permanente and building out a natural language processing team. We've had countless successes in mining the data and coming up with new insights that have been actionable. I won't even list those, but I will talk a little bit about what really excites me about the future. Something I've been talking about for about seven or eight years is NLP for NLP and CBT. Right? Obvious? No. Natural language processing for neuro-linguistic programming and cognitive behavioral therapy. The simple way of representing that is a sage on the shoulder. There's only so much influence we can have on our children. We know that there's a whole lot-
Of mindfulness and resilience in terms of habits that are developed early in life, and that those are critical to both lifespan and healthspan. It's not what life delivers us, it's how we react to it. What if we had continuous monitoring of all conversations that our children were having wherever they were, and a local analytics and feedback loop that helps them deescalate a conversation with a bully, or helps them invite people into a collaboration at an early age, and teaches them how to be mindful of the micro decisions we all make every day to be healthy.
One other thing I'll mention is, there's pretty good evidence that if children before the age of 5, and I spend a lot of my focus thinking about how we bring better health to children, but children before the age of 5 who get too many antibiotics, they end up at the age of 15 with an MRI scan showing that their gray matter is maldistributed. There's a lot of circumstantial evidence that that's a very direct impact of causing a dysbiosis, that is, changing the bacteria in our gut that have 100 times as many genes as the human genome, which then interacts with our immune system, which then interacts with our genome and how our brain is wired.
That pathway is so convoluted with so much data between the 100 times the genes in our bacteria that are in our gut and those that are in our own genome, interacting with the immune system and our genome and our brain and how it's wired, there's no way that the human brain can possibly wrap itself around that complexity and that large data set. Machine learning is going to play an absolutely critical role in helping us understand those kind of connections. Those are a couple things that I'm really fascinated by and that I believe are completely tractable with some of the data sets and some of the machine learning tools that we have today, and even more so every day with all the great work that's going on here.
Of course.
Hi, Cameron. Hey, Cameron O. Longtime technologist. I went into the health insurance field back in about 2008. Since then, I've been kind of labeled as a bit of an more innovation kind of focus group. We do a lot of interesting stuff really to help enable a lot of our, what we call our consumer health strategy, like the emergence of all these ancillary tools, devices, bringing those all to bear so that we can actually help support personalized experiences for customers. What's really interesting is we also own a lot of the how that data's now going to be carried across into their experience with the whole care provider ecosystem.
How do we actually bring their PCP, the specialist, with all that information that they've been using in their personal time, or the tools they use to kind of manage their condition or where they get their information. How does that now flow across, and how do you actually help affect the right decisions, the right steerage, at the point of care to obviously get the best outcome. The AI stuff is interesting for me for a lot of pretty pragmatic reasons. With all this data, this just massive increase of data, I heard it all from the last couple conversations up till now.
The mass amount of data, I can't hire enough people to actually distill that down into a lot of credible kind of points that I can actually influence at the critical moment with a customer to make sure I've got the best information available to drive that over and kind of identify where I have to route, make decisions, support people more effectively. It's really, AI helps us kind of get that signal-to-noise ratio into something manageable, right? I'd also say, too, that we have a lot of human processes. There's a lot of predictability that we look for in the healthcare space. We have a lot of intelligence, right? There's no lack of smart people in the space.
This concept of allowing them to work up to their license, to really do the things that matter, to use the best of their experience to bring the best outcome, and maybe handle or kind of offload some of the less burdensome things. Some of the more trivial kind of things that could be just a matter of education, right, through interactive self-support means. Helping people get access to care when your doctor went home for the night, but you need to get some information to make a better decision. It's 2:00 A.M. My child has sniffles. Getting people directed to the right type of care that they can actually get better benefit of, that's more convenient to them.
Could I just build on that a little bit?
Of course.
I agree 100%, I think that was a brilliant description of one of the biggest opportunities. The thing I'd like to add to that is that as we bring all of these data sets to bear to convert data into knowledge, into actionable information that is then steered, one of the missing pieces that we have in personalized medicine is knowing how every individual is motivated differently than every other individual.
Exactly.
Being able to do machine learning, either real-time or offline or both, to understand when we deliver the right message at the right time to the right person, it's how do we deliver it? Do we inject humor? Do we use an in-app avatar? Do we use some sort of in-app reward system? Do we use a text or an email for a different situation?
I like to think of the fact that personalized medicine is going to be most effective when we have the kind of AI and machine learning that really looks at the motivational complexion of each individual and personalizes the motivational approach to the extent, and I'm working with a vendor who's doing some of this, where we can profile people very quickly and then determine whether they respond better to humor or a very directive kind of, you should do this, or a set of three options, and talk to your family or talk to your doctor. That once we begin to know this, we can create these, what I call motivicons, that are not like emojis, but they're actual pieces of video or avatar-delivered messages or in a large spectrum within a motivational formulary.
Just like we have a drug formulary today, have a motivational formulary that is personalized to the task and the individual and their motivational complexion.
Yeah.
Can I take that one step further?
Please.
So
Mike will jump right in.
You bet.
Pile on this one.
You bet. We're just free flowing up here.
Yeah.
You're the catalyst.
Yeah. Traditionally, with the way we actually look at personalization around a lot of our customers has been the typical market segmentation approach, right? Taking a lot of our population, building the cohorts that have some shared representative characteristics. To really get here, I think this is where AI has a tremendous opportunity, is the concept of a segment of one, right? As truly as an individual, you can bring all that information to bear. Again, drive down that signal-to-noise ratio into something that you can make sure that the right information is really truly present. You're building more of a rapport with that individual because that's the key to actually drive people.
Absolutely
to make better decisions and build the patterns that lead to better health outcomes. All that stuff is around building rapport and building confidence and trust based on show me that you know me.
Right.
It's so ironic that population care is still becoming avant-garde in the era of personalized medicine. I mean, really, we'd like to bury it.
Yeah. No, I just wanted to add that, for our team, the barrier in creating these AI solutions isn't the technology, isn't accessing the data. We have great collaboration with Intel in getting us there. In terms of seeing outcomes for the patients, it's this grinding work that we have to get the right human factors interface. We have this solution that's sending out a detection when someone is at risk of severe sepsis or septic shock, basically when the body's about to collapse. Positive predictive value is between 40%-50%, it's tremendous. Comes out 30 hours early. We've been running for five months, we see no change in our outcome for our patients.
There you go.
Which is really interesting, right? Then you take something else that's also used in combating this, which is getting a blood culture and understanding what they're battling. The blood culture is called, and the clinician, before they get the result one day later, they've already decided. They've already decided on the full spectrum antibiotics or some other set of care. You have these two things that are kind of crossing, the prediction that comes out 30 hours early, too early. The clinical team goes to the patient, and they don't look sick, to the blood culture, which they call and comes too late.
We're thinking about combining the alert with the order for the blood culture to say, "Look, you're not going to get the results in the time that you want, so how about we give you this risk value when you order it?" This is the really important work, in making sure these things are effective, in making sure they scale at an enterprise level and become these bolt-on things, is how do you get that human factors interface, whether it's super personalized or it's targeted to a task? That's really the challenge that we have in the operationalization of all these solutions. If I may, I think an interesting aspect of what you're trying to do, which is change the system a little bit and what you've described in terms of trying to personalize to the patient.
One aspect that is sometimes not at the table, and I think you maybe can speak to this more than the rest of us, is we also want to be able to model a provider and their mentality-
Yes
how they're going to interact with the patient.
Yeah.
We've got some providers who are very high risk, high return, and some patients respond well to that. We have other providers that take a very conservative approach, and that works out really well. Only when you get that interaction as part of the AI model, where you're modeling the provider, you're modeling the person who's paying the bills, you're modeling the patient, can you really get a holistic view. That's really hard for us to do individually. We can do a little bit there, but we have to bring in data from all these sources, and that's where AI really provides a holistic model, including the provider view, and the patient view, and then of course, the payer view.
Yeah.
I think one of the greatest examples of that is 1,000 times a day, a 45-year-old woman walks in the office and has newly diagnosed diabetes and hypertension. You can take in all the data from everything you know about her, and you can do the mash-up between the universe of knowledge and big data and the universe of the N of one, and do the mash-up and say, "Well, here's the three options. Option number one, you're going to radically change your lifestyle, and you won't need any medication, and your diabetes and hypertension are going to go away. Option number two, take these two drugs and come back in two months.
Option number three is somewhere in between." The woman says, "Well, the reason I have hypertension and I'm eating too much and I've got diabetes now is because my daughter's getting married in three months and I'm going through all the wedding hassles. What I want is I want option number two, give me the two drugs, and I'll come back after my daughter's married and the in-laws are all out of town, and then I want option number two." In order to support her decision in what she wants to do, when you do exactly what you say and lay out what those three options are and allow her to drill down to the extent that she is capable with her health literacy to begin to understand what those options represent across the waterfront.
Machine learning is going to play a big role in both the construction of those options, as well as the ability to drill down into the reference information in the N of one.
Right.
Yes.
That's really where we're focusing a lot in this kind of dialogue of cost benefit, right?
Yep.
The AI can help with that, but it's a dialogue waiting back on the clinician or the patient.
Right.
That's a really important piece that we've identified. Initially, we just came out with like the blinky light, like pew-pew. We're going to send out these alerts and just insight and wonderful things will happen. We're just not seeing it to the level that we want. Not a big surprise.
Yeah.
We're really working to that next level, to provide those decisions being made.
Yeah. Interesting about the my perspective has been that traditionally a lot of the healthcare delivery stuff has been very quantitative. I mean, clear measured data, biometrics, labs, things that you can actually measure. Where I'm interested specifically, and I really think AI's got a tremendous opportunity, is more of this qualitative stuff.
The stuff that's more implicit or derived intelligence about how people are interacting with systems that kind of fall outside of the normal healthcare delivery place. People are going to Dr. Google to get a lot of their information. There's a lot of their financial information, personal details that we haven't captured or collected before. A lot of that information and kind of identifying causative patterns, using things like Saffron, things we hadn't even considered to look at because most of our traditional data modeling approaches or data science approaches have been the very kind of rigorous scientific approach of saying, "Here's an expected outcome.
I'm going to go after that and build models that kind of prove or disprove that." Where AI can help us, I think, is finding some of those new nuances in all this data to figure out how we, when we make a decision around a treatment of care, or when we have an opportunity to intervene, that stuff can help us shape it in a way that makes it potentially more effective or maybe get us to a better outcome by leading the patient.
There was. Go ahead.
I was just going to say, I want to throw a question out to the panel because I think we're converging on some ideas here, the N of one, the qualitative-
where AI can impact. There's a looming question that we're all going to have to face as data scientists. The traditional model, what works about the traditional model of study, is you've got a rigorous scientific experiment that you can validate and get to an outcome and say, "I know this is true because I've proven it." Now we're talking about the N of one. We're talking about doing qualitative things. How are we, as scientists, going to validate that in real time and provide that when we give some information to a clinician or a patient, we have some confidence around that.
Agreed.
We said it might be an N of one, but we still think we're pretty good. How do we control that risk of going off in the wrong direction?
That's why I know that our approach specifically is, so we can talk about all the wonderful things that AI can do someday, but because we're in a very highly regulated environment, we have to get it right. You can't get it wrong. You can't use predictive modeling to actually say, "Well, I've got about an 85% certainty, and I'm going to make a decision on a care path for that." Our pragmatic approach is really how do we actually, like I said, get down some of those more pedestrian, more prosaic things, take those out of the system so the people that are there are licensed to do this appropriately. They've actually got less noise. They can actually focus on making sure that they're doing the best care with the best information.
As these things build up, and we start to build some confidence, either by shadowing along virtual assistants, shadowing along with some of the people, the way that we're getting on the phone and supporting somebody in a disease management case or case management, and building and understanding some of those patterns almost by shadowing along. We can get to a point, potentially, where we can really start to do dress rehearsals on this stuff and figure out when it is the right point to allow these big critical decisions to leap over to a technology.
Yeah. I think it's a really important place to have publications, to have shared experience. We haven't worked that out yet. There's an area of just causal inference, like understanding, all right, so if we're detecting that someone is likely to have severe sepsis and there's some care intervene, and they didn't, what is that? We want these false positives to increase, but in a way that's Understandable good and safe, right? Understandable.
Yeah.
Expected, it's an important area of study that we need to contribute to make these things more scalable, more plug and play. I think it's important.
I think what's fascinating is in the field of statistics, which is one of the most challenging disciplines in math.
It has to be
There's three new ways of looking at data that are being statistically advanced in this new context. In mobile healthcare, the stage wedge statistics to be able to add people to a cohort over time and still be able to do valid analytics is one. The other is, to your point, single subject studies. You see what happens along the course of this N of one and start drawing conclusions about that one individual for which there's no one else just like them, unless they have an identical twin, and even then they're not the same because of the transcriptomic interaction with the environment.
The third one is, to your point about going from a correlation to the inference of a causation, and there's a whole new statistical field called data interrogation, where you look at the big data analytics and you find something interesting, and you can begin to query the data in different ways and different data sets without having to actually do prospective randomized controlled trials to actually be able to cross-validate, is this spurious, is this real, is it causal? Those three areas of statistics address each of those three different questions, and they're invented because of the new access to data that we have today, and our ability to become much more personalized.
We need to work together.
Yeah. Here we are. Intel needs to keep us in the room.
I think we're just having a private conversation.
Exactly. We're having fun. What we've obviously got an opportunity to create insights from data to understand context, which I think we can probably all agree that the concept of AI really is about understanding context. What's important to me.
Right
Given what's happened recently, what's happened in the past, who you are, where we are right now. How have you instrumented these insights into your organization? I'm curious about what technologies or approaches have you used to get to that point of having an insight from data, and then how are you instrumenting that? Particularly, if we think about AI as something that augments human capability, this is a three-part question. If you think about AI as something that augments human capability, which humans are we augmenting for each of your experiences? Where's the bang for the buck? Because you could be talking to providers or administrators, case managers, patients. There's a lot of opportunities. What's your experience instrumenting it and influence? Right. We've taken a couple of approaches.
A way to categorize it is we're looking at cases that are very rare, very impactful, and it's a burden on the clinicians to monitor these very rare, but when they occur, big deals for the patients. Severe sepsis, maternal morbidity, stroke, these are things that we're building models for and sending out these alerts, and we're starting very simply where we're just notifying the clinicians. In some cases there's a just be vigilant kind of a response, and that's the severe sepsis where we're very underwhelmed in the response. In others, we have for Venta, there's a spontaneous breathing trial that the respiratory therapist will go and conduct, and we've seen better results there. Things that are very actionable, that are in line with the care that is already being provided or intended to be provided are things that we find better results in.
Those are things that are interesting but really hard to manage, because these things that have very low prevalence are going to take For one case, it took five months just to say, "Oh, we missed the mark." Right?
Right.
Now we're going to dial the knob a little bit, and we're going to wait another five months. We're going to do that because it's worth it, but it takes a long time to converge. There's another set of problems where there's little decisions made all the time, and if we can just optimize them a little bit, where it's levels of care decisions, it's people that go around our hospital and just make sure things are documented correctly.
Almost at an administrative level.
At administrative level
where you get lots of little-
That's right. By looking at these places where there's many decisions made frequently, and we can just optimize a little bit how the humans are going through and creating these lists, then that's going to be a value to them. It's already in line with the work they're doing, and we can justify the value to the health system and the patient by the outcomes.
Right. Right. Yeah, we see that a lot, the opportunity to prioritize based on the very specific circumstances. David?
Yeah. In our case, while we're involved with things at the individual patient level and administrative level, Mayo has a bit broader of a strategy. We're really working towards what Mayo's always called the value equation, which is quality over cost. At one point in time, a decade ago or so, people thought the value equation was a single equation that worked in large groups of people. Now with individualized medicine, precision medicine, we want to make that value equation be really at the individual level. In order to do that, you need that individual level context that you're talking about.
For us, the context is really what is the personalized cohort that really mimics you the best? There's traditional ways to do that. We happen to use a graph computing sort of a model to build context. Graphs are very good because it deals with heterogeneous data very well, and you can build patterns pretty straightforwardly. We use that as we get our context from that, then we use that context to build out a model that's the value equation, quality over cost, for that person. Even though you might look like even your twin sibling, it's really not the case that you need the same care, whether that be a human factor component or be a genetic component or be a where you are in your life component. All those things are very different, the environment you're in.
It's building those models in real time that we need the AI for, because we can't do that value equation very quickly when we're doing a pencil and paper.
Right.
Two days ago, I had the privilege of spending the day with Vint Cerf, who's doing a number of global initiatives around, one is Innovation for Jobs, and the other is People-Centered Internet, to bring the internet to the darkest corners of the world to help bring people into the digital economy and to address the issues of loss of dignity with job loss. His objective of starting that initiative seven years ago was knowing that AI and robotics were going to displace so many jobs so quickly that we weren't going to be able to keep up.
Last week, at the Techonomy conference run brilliantly by David Kirkpatrick, there were a number of open discussions like this about how that's going to play out, and sort of a consensus among other people who have thought about this a lot that we have a problem maybe for the next 10 years of disintermediation and loss of jobs. In 10 years, to your question, who's going to benefit?
Everybody. The number of jobs that are not being done just in healthcare alone today is huge. The opportunities for whether it's the administrators or the analytics group themselves or the physicians themselves, the nurses, the patients, they all are in a position to have collaborative relationships with AI and robotics that'll help it. What I see is our big opportunity is to use augmented reality and virtual reality, which have been shown to accelerate learning dramatically, and this field is just exploding, as everyone here knows. To be able to try and apply AR and VR to how we integrate the AI robotics and the human experience into dyads that we create more jobs earlier and try and shorten that window where we're really staring all this job loss in the face.
The last thing I'll mention with respect to that is, I had the good fortune of getting to know Andy Grove before he died, and I think most people know he's from Budapest. One of the biggest problems in this country right now is if you go to prison, and you look at who's in prison, almost everybody there has a learning disability or a language disability. They get bullied in school, and they have problems learning and integrating into society. I think one of the coolest things about the collaboration of AI and machine learning in the space of natural language processing, but not exclusively, is being able to do early identification of where children with English as a second language have language issues for which today there are no adequate tools for early diagnosis and intervention.
By the time you find out that an immigrant child has a learning or a language problem, they've already been subjected to a lot of discrimination above and beyond their being children of immigrants. To the extent that we can apply this to bring out the best in our immigrant population, like Andy Grove from Budapest, I think we have a huge opportunity to expand the job market and keep people who would live a life full of dignity out of prison because of undetected problems at an early age. Machine learning, I think, is going to play a huge role in that.
That's great.
You had a big question. There was three parts to it, right? Take us through it.
You can answer any of the above questions.
Well.
I guess the question was, how have you instrumented.
Yeah
AI and who's being augmented in a meaningful way?
Well, I guess the answer is two parts. It depends on everybody, right?
Traditionally, healthcare has not moved very quickly. The way the information moves around. From my lens, working for a health insurance company, it comes in through claims. Claims may indicate something that happened a month ago, and I'm using that to kind of draw inferences about what's going to happen potentially in the future. As we move forward, we're starting to get in this business now where there's more focus on risk with the providers and how are we actually managing that risk, predicting behaviors, managing population health more effectively.
Using a lot of these latent data sources, the need for us to speed that up to make it relevant, especially as we're creating these alliances with providers, because we do manage risk. That's what our bread and butter is. Helping bring that kind of competence out to the front of at the point of care is very helpful. It's not going to serve us very well if we're still using all these latent sources where the information that's in the EMR is minutes old, and it contradicts what I'm sending. Right? When we talk about who we're going to automate, I don't know that it's necessarily a person in all cases. We talk about our end goal is making sure the customer experience is personalized.
We'd also like to make sure that the information is relevant to all the people that are supporting it, like treat all those users all the way through the value chain with the same level of personalization optimization. Workflow is king in about how you actually get people in, deal with them effectively, and get them to a better outcome. Using that same kind of level of inferencing and intelligence to kind of personalize how the information is delivered to a care coordinator that's sitting out at an accountable care organization, and they're making sure that they call out and get Linda in for her appointment because we just found out some information that she's got a potentially undiagnosed condition. You have to make the information as effective to them as possible so they can go about their day and still meet their value objectives.
Everybody in that whole value stream is pretty important. Traditionally, we have a lot of programs we do around clinical management, condition management, behavioral case management, helping people get referral for services so they can get in. All of the people associated with that, there's an opportunity for AI to automate, make sure where possible that we're actually staying within compliance, or making sure that we understand where there's variance and how to align to any standard operating procedures and practices, et cetera. I guess that's why I said it depends on everybody.
Right.
Can I come back on that just?
Yeah. Of course
real quick? I think the consensus here is it's around optimization, it's around personalization, it's around insights. It's not around a driverless health system, right?
Exactly, yes.
you hear things that.
We can get rid of that one.
We're going to get rid of doctors, we're going to get rid of nurses.
I'll get it first.
This is all optimized. It's a little bit like Andrew Ng, one of the founders of Coursera, made the statement that it's a little bit like worrying about overpopulation on Mars. We're a long way from even having to address that, and it's really about getting better outcomes and focus outcomes, helping people that really otherwise wouldn't have been helped, and unburdening our care teams to do exactly what they want to do. For all our solutions, we are pulled through by the clinicians. They're banging on our door to help them deliver better care.
Yeah.
One of the things that's come up a couple of times, is the human factors and the soft side and the social determinants of health, and we haven't really dove into that much yet. I think it's important to recognize that in the human factor side, the household we live in, the community we live in, all of the social factors that influence our health are profound. There's a ton of literature validating that. Yet, if you look at the complexity of what drives us from a social perspective, it's vast. We now have tons of data that are accessible to help understand that.
To the extent that we can link community health and the social determinants of health with personalized healthcare, I think we can really abandon this notion of population healthcare, where we classify people according to one or two diseases or one or two lab values.
Right
Begin to really look at this broad complexion of the human experience. To your point, we'll never have a driverless healthcare system as long as people live in communities that provide for very rich and profoundly deep experiences.
Yeah. Totally agree. We're getting close to the end here. We've talked a lot about different ways that data can drive understanding of patients, of disease, of different workflows for how to deliver care for even coaching. There's a lot of different things that we're doing and different things we can do. In your vision, if we're really successful with AI in the next, let's say, couple of years, not 10 years out, but in the next couple of years, what's going to be better about our healthcare experience as people trying to stay well?
Well, just real quickly, we're very much in the beginning. I think we're helping to converge right now, focusing a lot on diagnoses. I think in the next two or three years, we're really going to help with those cost-benefit decisions and personalizing those things. We're not there yet. There's a lot of these great challenges ahead of us. I think it is in the horizon of two to three to five years, we're going to start seeing that and seeing a lot better outcomes for patients.
Great. Better outcomes. David?
My view on it is, and it is just my view, that to the individual patient who walks into the hospital to talk to a doctor, it's not going to look that different. It's just that when they leave, the outcomes will be better.
Yeah.
It'll be better because so many things are happening behind the scenes, you don't know what's going on. In the end, you still ask your doctor, you say, "Doc, I don't feel well. Help me." The doc will do that. Then you just know that when you leave, you're better off than when you started. We have that now, we really do. It's going to be more precise.
It's going to be more effective. At the end of the day, it's also going to cost less.
Nice. John?
Today, 50% of the primary care docs in this country and across the world are experiencing burnout. The reason they're experiencing burnout is because in the digital age, you're confronted with every potential care gap that might have possibly existed unaddressed for the past 20 years, plus those right in front of you today. It's overwhelming, and people are overwhelmed. My view of how things are going to change is that we're going to use modern technology to restore ancient wisdom and allow doctors and nurses and pharmacists and healers of all sorts to do what they wanted to do when they went through their training to go into the healing professions. That is to be compassionate and caring and listen to the patient and empathic, and actually understand how this person is different, put yourself in their shoes and be able to understand it.
Today, our physicians, nurses, pharmacists are so overwhelmed with disintegrated data and disintegrated advice and incomplete synthesis, that they're burned out. They're really burned out. I see the difference in the experience of the person passing through the healthcare system as twofold. One, we'll try and keep them healthier in the first place by providing more mindfulness and resilience so they don't get sick in the first place. Secondly, if they do get sick, we'll have people on the healing professional team as well as the personal care team that has an opportunity to invest in understanding, being empathic, and being compassionate in care. I think the person experience is going to be vastly different.
That's great.
If I take a look at where I think it's going to be in a few years, I would say, I mentioned this a lot during the talk, but it really has to be personalized to drive better outcomes and drive value to people. I think it's going to become much more convenient. When you come in, people get the right level of care that's really needed for where they are. Helping our doctors or care providers and care practitioners work up to their license and do the things that merit their time, and getting the other stuff out of the system so that we can actually make the right decision, the right service. You're going to see a big cost benefit to that too, across levels. People are getting more information earlier. They're not burdening our healthcare system. They feel better.
The information that really informs who they are as an individual is being brought to bear at that critical point of care, so there's actually more confidence that what they're getting is appropriate. They're not doctor shopping and some of the behaviors that we see today. That's what I would say. It's going to be cheaper, it's going to be personalized, it's going to be more convenient.
More convenient. All right. I think this was really interesting. The way I would summarize, I think I'm going to steal from John and say that the opportunity for artificial intelligence in healthcare and precision medicine is not to make healthcare less human, but actually to augment and make it easier for us to make healthcare more human. I want to thank the panelists, and I think it's a really interesting conversation. Thank you.
Thanks, guys.
Thanks.
Good job.
Yeah. Great.
I don't want you to go far because we've got two more really outstanding panels on AI to come. We've got the autonomous world panel, the group has told me just take a five-minute break. If sitting is the new standing, you know what I'm talking about. Smoking. If that's the case, five-minute break. Back here for autonomous world in five minutes. Thanks.
Hello and welcome. I'm absolutely thrilled to be standing in front of all of you for our second panel discussion today. As we see more and more industries and market segments advancing towards autonomy, it's clear that AI is becoming the fuel for that innovation in autonomous things and autonomous systems. For this panel, we've assembled an extremely distinguished cast of characters who bring a diverse set of experiences, ranging from the automotive industry to the manufacturing sector, and the energy sector as well. We also have a guest from the academic arena as well. I'm Brian McCarson. I run IoT Strategy in the Internet of Things Group at Intel. Let me now introduce our four distinguished guests. We're not going to, by the way, just so you know.
We contemplated standing behind the chairs awkwardly and just holding them, figured that wouldn't work. Let me first introduce Cathy. Cathy is a manager for advanced analytics at Devon Energy. We're also lucky to have Jagannath here as well, who is the head of data services in the Digital Factory division at Siemens. We have Reinhard Stolle, who is here from BMW and is a Vice President of Artificial Intelligence and Machine Learning. Our other honored guest is Professor David Yoffie, who's here from Harvard Business School. Let's go ahead and have a seat. Thank you to everyone for coming. What we'll do to get things started is go through and have each of you speak a little bit about how artificial intelligence is playing a critical role in the evolution and advancements of autonomous things and systems in your respective fields.
Cathy.
Okay
We'll start with you.
My name is Cathy Ball. I was going to tell you a little bit about how artificial intelligence is changing the landscape of oil and gas. Particularly as some of that commodity prices really start to bottom out, you have to be cost-effective. You have to do things in new ways. The industry is actually starting to really adopt some of the AI. What we focus in is the autonomous drill bits, even down to the autonomy in the digital oil field, and getting the value out of that and getting past the hype.
I think I'll start off with trying to define what is autonomous manufacturing. I think at the very extreme end of it, I can imagine a scenario where you get a digital model that comes into a factory as an order. That digital model is then put into production by completely autonomous machines. You get the product at the other end to the right quality and specifications as the model defined it. Then you have logistics, autonomous logistics, shipping that and delivering it to the person who ordered that. Now, in its extreme, this would be autonomous manufacturing. I think there are a lot of foundational pieces that needs to be put into place for this to happen, and that's something I hope we'll discuss during this session here.
Fantastic. Reinhard?
Artificial intelligence is transforming our product in at least three main ways that I imagine. One is it makes autonomous driving possible, right? We have new ways of perceiving the environment of the car. The car has a way of interpreting the situation around it, and then finding its way through the traffic, negotiating with all the other traffic participants. That, of course, that's the major impact of AI. The second impact is personalization and adaptation. Providing context to any situation that the driver or the car may be in. Also user interface is becoming more natural, gesture, natural language, everything just being less brittle and more robust in a natural intelligence way of interacting. The third way is that the whole notion of mobility and ecosystem, system of mobility, will just change by the influence of AI.
Rather than cars as sort of driven by individual minds.
Trying to get from A to B, the whole ecosystem will be connected and use that connectivity to have an overall intelligent solution. You may think of intelligent car sharing and so on.
I don't have a similar story to tell. Obviously, as a professor, I look at this a little bit differently. Every year I give a lecture on what are the new technologies I think executives around the world need to be thinking about. I try to look at things that I think have impact within the next 18-36 months. This year, AI machine learning is the topic that we'll be talking about, precisely for all the reasons that the panel has described. The business problems associated with actually making money out of that technology is much more problematic than just saying it's a technology that's going to be important and relevant across the board. Let me just raise two or three questions, which again, I'm sure we'll come back to in for the context of the panel.
One is the problem of what I like to call looking forward and then reasoning back. Which is, we can look forward and sort of understand what this is ultimately going to do, but thinking about what the technology delivers two, three, four years from now, then trying to figure out, okay, what does that mean I have to do today? Because we aren't there in actually many of the technologies that we've been discussing in today's conference. Problem number 1 is how do we reason back to the actual strategic implications for today? The second question that I think we have to think through and understand is ultimately for this to be successful, it has to be a true platform and not a product. A platform means it has to have a degree of openness.
It has to have a degree of sharing of data, across firms, which is challenging, again. BMW wants to do something distinctive from Audi and wants to do something distinctive from Mercedes. Yet ultimately, for us to be able to deliver on many of these AI solutions, we have to figure out how do we make these not just products, but platforms, because their success depends on that platform structure. Thirdly, related to that, I think we have to think through the potential for network effects. In the end, AI is going to be successful and have the ability to drive the value we all hope if we can find ways that everybody benefits by virtue of everybody else doing the same thing.
Again, in a world where we try to do things in a proprietary way, driving those network effects become problematic, and that actually slows down the ultimate adoption and success of a technology such as AI.
Let me piggyback on that a little bit.
Yeah.
I think what you just mentioned about we are not there, is absolutely true. When I go into the manufacturing space, and those of you who've been in the manufacturing industry, control systems have been in manufacturing for the last 30, 40, 50 years. Manufacturing plants have always been generating data, huge tons of data. The problem is we never exploited the richness of the data. Either we didn't have the technology or we didn't have the compute power. If we did that, which is possible today, and I think that is one of the foundational pieces that we need to add. IoT brings in that foundational piece, connecting all those data sources, connecting those machines, gathering those large big data sets. When you do that and apply machine learning, deep learning, now you start getting production models.
Production models for different operating environments. Production models are a very important element for autonomous manufacturing. I think these are the foundational pieces that we need to put in place.
Excellent. On one of those principles that we just described, and you highlighting the 3 different categories of consideration. Autonomous driving keeps showing up as a breakthrough technology that's bringing autonomy and artificial intelligence closer to the consumer than ever before. We've all, any of us who flew here, probably flew in an autonomous airplane where the pilots are just there to observe and be around in case something goes wrong. The actual idea of ownership of an autonomous vehicle is now becoming in the realm of possibility as the cost of the technology and the sophistication starts to grow. Can you comment more in your field? What sort of challenges and obstacles do you feel like the industry needs to overcome in general, both locally and globally? What sort of approaches are you trying to take to overcome those obstacles?
Right. We are working on a very exciting but potentially world-changing problem, which is autonomous driving. The first challenge is that this is something that a couple of years ago, we still thought, or everybody thought this is sort of very many years out there, before it becomes reality. Suddenly, through these advances, that more or less came overnight in the last three to four years, especially with deep networks and most prominently image processing and computer vision. It becomes sort of reachable, right? Like Davis mentioned before the announcement that we said by 2021 together with Intel and Mobileye.
We will build this open platform that has some degree of openness, and we are hoping to build something that is adopted across the industry and becomes mainstream just because we believe the problem is so difficult that if each of us tries to solve it alone, it will just be harder and take longer. Some of the very difficult problem is bringing it from the research lab to a product, right? Not making it work some of the time or most of the time, but make it work all the time, right? We as a car company with a long tradition, we have a lot of expertise in functional safety, right? Making sure that it's actually safe and our customers can trust our product.
In bringing autonomous driving to that level where our customers really trust it, that is one of the challenges that we have right now. Just how do we convince ourselves that it works not just in the cases that we tried out and the ones we envisioned, but in all the cases, right? Another difficulty is that, what we call the cognitive architecture. There are so many different techniques in AI and machine learning is a subfield of AI, and there's so many different machine learning techniques, and they can be applied to many of these subproblems of autonomous driving. The question is, well, which one applies best to which of the subproblems, and how do we combine it to the overall cognitive architecture, right?
Now, the first time in, let's say, history of car making, we have this chance of not just building individual functions, driver assistance functions that we build in the car, but we give the car this overall broad competence of driving autonomously. That is just very exciting, but at the same time also quite difficult. Right? If I may add one more aspect of a difficulty, less from an engineering point of view, but maybe more from a development of an organization point of view. If you see how the whole world of mobility is changing, and we are now stepping one step back and saying, "What's the business we are in?" Right? You could say, well, we are in the business of building really wonderful and exciting and emotional cars, right?
Really the business we are in is providing exciting and wonderful mobility, individual mobility to our customers. In this new world of sort of transformed mobility, in addition to just building wonderful cars, there's this whole new world of services, and that one is driven by artificial intelligence, and for many years has been driven by software. 15 years ago, we founded a small software research lab at the time, right? Over the last 15 years, we have hired literally thousands of software experts like myself, and to prepare and to drive this transformation.
Let me make three quick points about your comments. Number one, it's really interesting to think about not just the capability AI delivers to do autonomous driving, but all of the other pieces in the ecosystem that need to be developed in order for it to occur. Again, for example, how do you deliver very high-definition maps that are updated continuously for every street in the world? This is an example of what I was describing earlier. If we don't have an integrated platform that is driven together by network effects, we can never actually accomplish this unless it ends up being in very limited and narrow areas. That was number one. Number two, it's important to also think about some of the behavioral challenges if we ultimately want to make money here.
I've given lectures over the last couple of years on autonomous driving, and I've had people from all over the world, people stand up and say, "This may be a great technology, but I will never give up my car." In other words, as great as the technology may be, there's a behavioral component that's going to have to change. That's a change which may be almost as difficult as the technology problem. How do we actually convince people that this is great technology and may be safer, but these people love sitting behind their wheel and driving? We have to think about how do we solve that challenge. I had a third one, but I can't remember what it was.
As soon as it comes back to you, just jump right in.
I think, to your point, I drive one of those autonomous cars, and I can tell you, when you sit on that wheel and you're in autonomous mode, you're always thinking, what if something goes wrong and it's going to wreak havoc on that street? I come back in that context with the manufacturing industry, one of the most conservative industries. You can imagine one of the most conservative industries getting to a complete autonomous mode. Yeah. It's also an industry where technology got developed to address individual applications. Therefore, standards are always developed after the technology was developed, and then different geographies had different standards. If you have to bring in an element of autonomous manufacturing, just like in autonomous cars, you've got to have rigid rules, rigid ways of how it is operated.
Otherwise, it is going to wreak havoc, and I think that's a very important aspect.
I'd like to just chip in a little bit from our industry, we're jealous, by the way, because we're out in the middle of nowhere. We don't have Wi-Fi, and we can't see where we're going, we're going drill bits under the ground. You can't see what's in front of you, and the measurements for the drill bit are 90 feet behind you. I'm greatly envious of everybody up here. I'd just like to tag along with what they're saying, and AI, and to get IoT accepted, there's a human component and a mechanical component. The human component you're looking at, I had one of my European colleagues say, "This IoT stuff, I don't want to hear about it.
I want to hear about sensors where we can take some information, provide analytics, and then take an action." It's all I could do not to laugh. Sometimes you just have to be really careful about how you speak, how you present it, because some people aren't quite ready for it. If you call it cyber operations, they're all over it. You just got to roll with it, try not to poker face smile it, and try to get them at the level that they can advance. The second piece on the human side for us is on artificial intelligence. We could have solved the problem using AI from the very outset, but the engineers weren't ready for that. We had to do a crawl, walk, run, starting with, "All right, we're going to have you tell us what the pattern is.
We'll mimic it, let the machine learn from it, till they're comfortable with that." Now they're like, "Okay, we trust that machine. I'll let it do it, and then see how that happens." We've gone from them telling us the pattern to the machine finding the pattern and doing it in seconds. This is just a huge improvement. Now it's let's beat the machine. It's just a lot of fun.
The human factors you're talking about there are allowing the emotional release of control from someone who traditionally their job was defined as being the person who made those decisions. Convincing them that they have the ability to either intervene if something goes wrong or to allow them to build up trust in the way things operate. There's certainly going to be technology acceptance challenges behaviorally and network effects, to your point as well, where there's other adjacent industries that we didn't quite realize were going to create some obstacles in doing that. Do you have any other examples from the industrial sector and manufacturing where you've seen us already make breakthroughs in AI, but perhaps we just weren't using the language?
No, there are major breakthroughs that have happened. Especially, we talk now in the manufacturing world about the convergence of the virtual world and the real world. The virtual world is where a lot of technology innovation has happened. The virtual world being, today, you can design a product part by part, piece by piece. Have it as a data model. You can simulate every piece of that. You can simulate the whole product and its functionality. You can design and simulate the manufacturing system that would be needed to do that. All this before a single screw is turned. The virtual world has evolved a lot. Now, this gets transitioned supposedly seamlessly to the actual real world of manufacturing and operations. This is where now the technology begins to innovate. This is where IIoT and those things come into play.
There's a lot of stuff happening out there. I think there are also a lot of foundational pieces which are still missing.
I wanted to get back to your comment about people will need to change their behavior in order for autonomous driving to happen.
Some people do, not all.
Some people. I can relate to that because I'm lucky enough to be able to get to drive a lot of BMWs, and it's something you don't want to give up immediately. I need to say, well, it's really a lot of fun. There are times when it's not fun, right? Last night, I was driving from Mountain View to San Francisco, and I said, "Oh, two hours should be enough." Two hours was not enough. Not a single second of these two hours was really fun having the ultimate driving machine. In these cases, I think people will be very quick to actually adopt autonomous driving because they get their time back. They can do their emails, they can relax, and so on. I think once the affordances are large enough, people will very quickly adopt it.
The other question is trust, right? I think that's the more important one. Will you actually trust the machine to bring you from A to B in a safe manner? I think that is our responsibility to make sure it works.
Also realize the machine's not always right, and get that expectation level down to say, "It will have times where it's wrong, but it will also do the self-learning and learn from its mistakes as well.
Again, this is my point about looking forward and reasoning back, which is, at what point are we going to be confident that it is 100% safe? We've seen, obviously, people trying to prematurely take advantage of the technology, and people die. I would worry about that certainly in both energy and industrial, as well as in.
the consumer space, such as automobiles. The critical question is: Are we doing the right thing at the right time? Not trying to jump too far ahead. Which today, there's so much enthusiasm and excitement about it, the natural tendency is to jump to the end state rather than say, there's so much work that has to be done. We can't prematurely try to adopt the technology before it's actually ready.
That transition, in many senses, is how you build confidence over time.
If I think about my own experience working in the semiconductor manufacturing industry at Intel over the past few decades. The basic precursors of artificial intelligence are simple rules engines that are if/then statements. Then over time, you start to realize, I have 100% confidence in that very basic rule engine, and I've discovered a few other areas where I could have 99% confidence if I added more complexity to it. So these precursors of automated process control have been in existence for quite some time. That allows, I think, people that are working in that environment to start to gain that confidence.
Where do you think that we're going to have challenges, this is for any of the panelists, where those breakthroughs may be so rapid and so quick that it's going to create either other network effects or social acceptance or adoption challenges that we would have to overcome? What role can policy, from a government level or standards from an industrial level in combination with academia, influence that?
I look at the democratization of some of the AI software and trying to say, how can you give common standards? That's currently missing. We're having to do cloud to cloud, hybrid clouds. We're doing deep learning on top of that. We're also trying to say, how do you get, we call it analytics in the fog, how do you get that closer to the source so you can make a faster decision where the latency is in the microseconds at a level that you can trust it? You can always solve the machine side. How do you solve the human side? That's a piece that if you miss that piece, you've only solved 50% of the equation.
I think in the manufacturing sector, there is no factory you walk into today which has only one brand of equipment. It's a very heterogeneous environment. Historically, in the manufacturing industry, these equipment have never operated with each other in any efficient manner. Interoperability is going to be a huge thing. The other aspect is, therefore, you need to have standards which enable that interoperability, because that's going to be how do you create autonomous manufacturing with keeping in mind that it's got to be completely vendor agnostic? That's a major challenge. The other thing that I mentioned, I think as you go across geographies, the standards are not only different, they're completely different. That's the other challenge.
I think the third challenge is, and I think this might be peculiar to the manufacturing industry, it depends on how advanced the particular company is. So there are different levels of adoptions today. That can also be challenging because you're never getting a complete picture of what's going on. The final thing on a social side, I think I remember, and I think most of us would know that when automation became a big thing.
30, 40 years ago, there was this huge problem about people losing job. Jobs are being.
Right
taken over by automation. You can imagine what happens with autonomous manufacturing. It's going to be even a bigger challenge.
Yeah.
I don't have much to add to that. I think if the value that this new world gives to people is large enough, then there will be a demand. For autonomous driving, if it works and if it's safe, people will just want to have it. Then everything else, like regulation and so on, that will just have to follow. For me as a techie, that is sort of a simple answer. Do the engineering first and do it well, the other things will have to follow. I think getting it to work and getting it to work safely, that is the main thing.
I have an example that I throw out. I'd be curious your reaction. Part of the problem, for example, in autonomous driving is there are 1 billion cars on the road today.
Right.
A lot of those drivers are not good drivers. There were surveys done in the U.K., which you might have seen, where drivers were asked if they saw an autonomous car, would their behavior change? The survey said, it was a very large percentage, I've forgotten, 40%, 50% said, if they saw an autonomous car, they would try to take advantage of it. They would try to bully it. They would make sure they got into the intersection first. There are real problems associated with the initial adoption. I've always said government can help in this. It can create standards. In the case, for example, of autonomous driving, the natural thing to do is to carve out a section of a city and say, "Only autonomous cars." I think a country like Singapore, for example, would be the natural place for this to start.
If you could have Singapore carve out 10 square blocks, then 20 square blocks, and say only autonomous cars will operate, and it works, assuming it does, you'll start to build the confidence. You'll build the trust. You'll start to create the opportunity for people to feel comfortable adopting the new technology, perhaps even faster. In the absence of that, you're going to run into these behavioral issues.
David, the question I always ask is, all of us sit on airplanes and fly, and it is always flying on its own. In the middle of nowhere, 30,000 feet above, we are happy to do that. I think it's also a matter of getting tuned to the idea to some extent. Today, all of us sit on a plane without thinking twice. The same argument should hold there, right?
I agree. We'll quickly adopt because it will work just like airplanes.
Yeah.
On the other hand, for driving there, it's a heterogeneous environment. That is different from the airplane. That notion of trying to just take advantage of autonomous cars, that I can completely mention. I remember when I was a student in the '90s, I went to all these AI conferences and robotics conferences, and that was the fun of any conference, when people were showing their robots, and you just did this, and they would fall over. It was just fun. I think that a natural reaction is to try to show that you're superior to the robot.
Remember, airplanes have air traffic control.
That's true.
It's a little bit different. There is a centralized control system.
Today. Today, in a big way, but much different when it started, right?
They had people. They weren't going by themselves, right?
Well, it'll evolve.
It'll evolve, yes. It just may take a little longer.
Exactly
to be adopted on a large scale.
I think a big issue coming up is what we're all doing to the communication networks, as you have to put in broadband to handle the big data or communication packets that you're sending through cell phone signals. As we get to terabytes of information going back and forth, and you are, how is the infrastructure going to have to change for this to happen?
Intel is figuring that out.
Thank you.
I think it's a topic for regulation, maybe also set up of infrastructure. It may also be a topic just for a lot of these dynamics in sociotechnical environments. It may also just be a matter of politeness toward machines. I could imagine that one way could be that you say, "Well, I take advantage of it, and I'll cut in front of this autonomous car." The other way could be that you say, "Well, there's an autonomous car. It may not see me. I'd rather sort of stay away and be a little careful." You naturally give the right of way to autonomous cars. I think it's probably more empirical that we'll find out how these things-
It was only a British survey, so Americans will be more polite.
Fran, I want to touch on the topic that you just mentioned, the data explosion. Data is going to explode. I can tell you, the manufacturing industry, if anybody's seen a simple what we call a drivetrain, a motor, a gearbox, and a load with a variable speed drive, can generate 2,700 to 3,000 data points. That's just one asset. Could have 10 of those, hundreds of those. Therefore, I think one way this is going to be addressed and one way autonomous manufacturing will also come into play is a lot more of edge analytics and edge computing. When I talk about edge, it's not just alongside assets, but in the manufacturing environment. Also, you have to start having computing at different points before you reach the cloud. I think that is really going to be kind of solution to the data explosion.
The ability also to operate within the clouds, hybrid cloud-
Exactly
proprietary systems, and do your analytics at the same time. Understanding it's not just an analytical problem, it's also a software and a technology problem.
Right. You want real-time, near real-time analytics.
Yeah.
That's the other thing.
In the case of the energy sector, you don't often have the luxury of an ethernet connection out in the middle of a shale field in some remote portion of the world. You have to rely on all that AI capability to be locally placed right where the decisions have to be made in order to really make that work, and to have the low latency requirements so that you don't damage the equipment as well. How is that going to affect your industry, do you think?
It's going to affect it quite a bit. You'll get towards that analytics in the fog. How can you get that decision done in milliseconds? If you operate it similar to a smart city or a smart grid, how do I connect those parts and those pieces and make an intelligent decision in milliseconds?
We're getting there, but you're also having to say, "Should I do it?" You're also adding Hadoop clusters and saying large amounts of data here, interacting in memory and database here, putting them all together, putting it back to actually tell the PLC to take an action at the oil field. You're going to have to get high-performance computing, high-performance analytics in cloud, perhaps in cluster in the Hadoop.
Ultimately, though, this is going to have to drive some sort of business or societal value proposition. In these other industries that we've been talking about, how do you see that value proposition manifesting itself?
You're asking me?
Yeah, sure.
Manufacturing industry is all about, if I may put it very simplistic, profiteering.
Also, of course, generating the goods.
Products required for a comfortable life. I think, the last I remember, the global manufacturing is close to $12 trillion or something like that. I always think about.
Hello, everybody. My name is Tony Salvador. I'm a senior principal engineer at Intel. I'm a social scientist. I'm one of their ethnographers. It is my honor to get to have known these people in the last couple of days. I say that because each one of them is doing work that is important and work that goes unvoiced. It's work that gives voice to people who have none. It's work that gives power to people who have none. It's work that gives strength to people who have none. They will each talk about what it is that they do. They will talk about the hopes they have for AI, what AI can do to help them do their jobs better, to help the people that they're trying to serve. Each one of them has a passion behind what they're doing and a reason for why they're doing it.
Just very quickly to introduce them, this is Sixto Cancel. He's got a nonprofit called Think of Us that helps foster kids work their way through the system. He'll describe about that. This is Michelle DeLaune. She's a Senior Vice President and COO of the National Center for Missing & Exploited Children. You heard John Clark, the CEO, speak earlier. This is Ina Fried. We have a little joke going back there. This is Ina Fried. She's a senior editor with recode.net. This is George Siemens. He's a professor at the University of Texas at Arlington, not to be confused with Texas A&M, which is also in Arlington. Each one of them will actually talk a little bit about what it is that's really driving their passion, why they're here, and why they think AI is important. We'll just start right here.
Hello. Good evening, folks. That's the part you kind of say something back. There we go. Oh, afternoon. There we go. There goes the confusion. My name's Sixto Cancel, I'm the CEO and founder of Think of Us. We seek to leverage data and technology, to really help young people in the foster care system heal, develop, and thrive. We've been developing this platform that allows young people to build their own personal advisory board, so that those adults in that young person's life can coach them through their adolescence then to be on their own.
When I think about the power of AI I think about the possibility of AI, I get really excited because I see it as the ultimate assistant tool to young people figuring out how is it that I go from being an adolescent in the foster care system, that a pathway is shown to me of, several different pathways are shown to me of how is it that when I turn 18, I have not found a family, I have not been adopted, that I have not only built the skills that allow me to be self-sufficient, also that I have chosen what my future is going to look like. Right now, we have about half a million young people in the foster care system. Unfortunately, when you are in that system, you do not know what's going on.
You do not have a lot of say in your case plan, what happens to you. When I think of AI, I see it as the ultimate advocate, the ultimate empowerment to young people in these systems being able to get the information that they need, that they can access the services that they need, make the plans that they need to, really focus on healing, developing, then thriving.
Thank you. Michelle.
That was a good one. Michelle DeLaune. You heard my boss earlier speak, the CEO of the National Center for Missing & Exploited Children. Thank you Intel for hosting us. The National Center for Missing & Exploited Children, we're a 32-year-old organization. We're a nonprofit, non-government, non-investigative organization that exists to serve families and law enforcement and all of those stakeholders who are looking to protect our children. We have three primary focuses and missions. One is to find missing children. As John mentioned earlier, we've been very successful. We found over 232,000 missing children and brought them home to their families. We've done a remarkable job in that area, which also that was the beginning, 32 years ago, founded as a result of a tragic abduction and murder of Adam Walsh.
In that time, over the 32 years, we have seen the scope of the problem grow. In addition, we've seen the advent of the Internet, that has brought a whole host of other issues along with it. Much of what we're going to be doing with Intel, with our new partnership, is focusing on how we can use AI to fight exploitation. We have the CyberTipline, which is receiving this year 8 million reports regarding children who are being sexually exploited online, primarily. Much of that is regarding individuals who are trading child pornography. We have 25 analysts working on the CyberTipline who are trying to handle 8 million reports. Clearly, as was mentioned earlier, looking for the needle in a stack of needles, in this case, every needle is a child.
We need to do everything that we can by leveraging AI and leveraging technology to stand out what is important, help us prioritize those leads, and make sure that nothing is slipping through the cracks simply because we're relying on humans. We need to be able to teach the machines to do much of what the humans are doing. We're thrilled to be here and be working with Intel.
Thank you. Ina.
My day job is covering this industry, and I've been covering tech for about 20 years now. I started at CNET. I've been with Kara Swisher and Walt Mossberg for about six years now, since we were All Things D and part of The Wall Street Journal. We went independent. We got bought by Vox Media. I write about this industry, but also Recode and Vox are partners with Intel on the Hack Harassment effort, which is really personal to me, as part of the LGBT community and having been a crisis line volunteer a couple of times in my life for LGBT lines and an LGBT youth line.
The amount of harassment that particularly LGBT youth get at a time when they're very vulnerable is a particular area of passion, as well as just seeing online harassment in general, whether it is women getting harassed for speaking out, whether it's people of color being harassed, certainly, and we may talk about this, in the environment we're in, I think that's only increasing. I love what Doug said at IDF. He was saying that technology created this problem, and it's really up to us as an industry to solve it. Certainly in general, we chronicle the industry more than getting involved. Harassment is personal to us. A lot of our writers at Vox come under attack for their political beliefs, their sports teams. I always joke, I've faced a fair amount of harassment, in my career, mainly around two areas.
A small amount for being a transgender woman and a large amount for covering Apple. I get a lot of just the craziest hate ever. I'm less concerned about the systemic issues around the last group, but I think AI hopefully can help us to create an environment where simply being a woman and talking about being sexually assaulted would not be an invitation to harassment the way it is today.
I think anyone who puts themselves out online is likely to face significant harassment. People choose to speak out nonetheless, but that shouldn't be the choice that anyone has to make to tweet, to speak out online.
Thank you. George.
I'm George Siemens. I'm with the LINK Research Lab at University of Texas Arlington. We have a guiding focus in our research activity, which centers on what does it mean to be human in a digital age. A number of areas we look at relate to work and automation. It relates to the experiences of success for all students. We're also focusing on future knowledge systems. What does a society look like in an age of AI, and how do we learn in that kind of a setting? A few things that are quite prominent is, we are likely the last generation that will be smarter than our technology. What are the social implications of that?
We've also seen, this has come up on numerous occasions on different panels and different points, which is we didn't expect it to happen this quickly. We're worried about 3.5 million truck drivers being out of work within the next 5-plus years. There's enormous social issues that are around it, but not even just the social issues, it's the learning needs that exist within that kind of a setting. Meaning, instead of a four-year relationship with a university where we go to school, get a bachelor's, we end up with something that likely looks more like a 40-year relationship with a university. What does that look like when we have this system that is prevalent in our lives, like the healthcare system is in our lives today, literally from birth to death, where we're going to be essentially learning beings.
My primary issue with that is not everyone has access to that system today. Less and less people are having access to it that actually need it most. If you look at the completion rates of college education based on income quartiles, those individuals that in the 1960s were sitting at around 20% access to the higher education level in terms of income, they've actually dropped slightly. The top quartile has gone up from around the 50%-60% level in excess of 90%. Basically, if you're born poor in America, don't count on education to lift you out of that poverty, even though that's a narrative that we have.
I'm very interested in what are the technological systems and how do they need to integrate with our social systems so that we can have those students be successful that right now are largely excluded from the system. In fact, the primary system that we know for elevating people out of poverty is now inaccessible to those that are in that position.
Wow. Thank you. Thank you all. You bring up a point right at the end there that I'd like to sort of lead off with, if you will. It's that notion of the technology solving a fundamentally human problem. I think each of you talked about things that are really fundamentally us caring for one another in one way or another, right?
Whether it's somebody who started off with a bad lot in life or somebody who has something to say and is being pummeled as a result of it, how do you start thinking about that integration of technology with what are fundamentally human social problems? Maybe we can start back with you again, and then I'd like to move to Michelle to think about that because you have some examples, I think, in how you're using technology and people to solve some specific problems.
Well, I think at this stage, we're very close to getting into a level where we can say really dumb things that 5 years down the road will be watched and will be made fun of by colleagues. I think it's always difficult in the AI space to forecast even a few years down the road. I do think that we need to turn, first of all, to the essential learning sciences literature and the effects that have the greatest impact on student success. Now, this could be a young student in the K to 12 system. It could be someone who's in higher education. It could be somebody who's been in the workforce and needs to come back to reskill. I think in all of those situations, the human actor is so central to our success academically.
Many of us have stories, I'm sure, of an missing my life, or someone went out of their way to come for me and it changed my life. One of the areas that we look at as a lab now is we're very interested in wellness and the human condition. I think cognitively, we need to just say, "Look, damn it, computers have won." We can't out-cognition our computing technology anymore, and that's only going to accelerate that gap. I think what's left, the last domain of humanity to stake its claim is actually on the feeling, the emotion, the awareness, the affect, and those elements. Our systems need to account for the things that technology does far better. I think it's always difficult in the AI space to forecast even a few years down the road.
I do think that we need to turn, first of all, to the essential learning sciences literature and the effects that have the greatest impact on student success. Now, this could be a young student in the K to 12 system. It could be someone who's in higher education. It could be somebody who's been in the workforce and needs to come back to reskill. I think in all of those situations, the human actor is so central to our success academically. Many of us have stories, I'm sure, of an missing my life, or someone went out of their way to come for me and it changed my life. One of the areas that we look at as a lab now is we're very interested in wellness and the human condition.
I think cognitively, we need to just say, "Look, damn it, computers have won." We can't out-cognition our computing technology anymore, and that's only going to accelerate that gap. I think what's left, the last domain of humanity to stake its claim is actually on the feeling, the emotion, the awareness, the affect, and those elements. Our systems need to account for the things that technology does far better. Missing my life, or someone went out of their way to come for me and it changed my life. One of the areas that we look at as a lab now is we're very interested in wellness and the human condition.
I think cognitively, we need to just say, "Look, damn it, computers have won." We can't out-cognition our computing technology anymore, and that's only going to accelerate that gap I think what's left, the last domain of humanity to stake its claim, is actually on the feeling, the emotion, the awareness, the affect, and those elements. So our systems need to account for the things that technology does far better. Cognitively, we need to just say, "Look, damn it, computers have won." We can't out-cognition our computing technology anymore, and that's only going to accelerate that gap. I think what's left, the last domain of humanity to stake its claim, is actually on the feeling, the emotion, the awareness, the affect, and those elements. So our systems need to account for the things that technology does far better.
To be able to access the internet and see these stories and understand that there was this very slight chance, this possible pathway, to be able to leave Bridgeport, and I saw someone do it, that encouraged me. When I think about what is the role of AI, at least in youth development in our work, it's to be able to say there is a couple different hands. The thing about all of us is that we're all born with different privilege. Privilege of being a male, white privilege, you name it. We can go in for days. You may be born with 12 cards, and this person may be born with only three cards to play. This person with three cards to play still has a pathway of winning the game.
It's just the person with 12 cards has a better chance of winning the game. The AI can help assist young people really engage in the right developmental opportunities, engage in the right steps to be able to play the cards that will get them to being self-sufficient, to being connected to education, to being connected to employment later down the road. Right now, what you have is a group of people beyond just foster care who don't see that pathway. I think that what we've learned in the last week is that there's a whole group of Americans who are hurting, both young people and adults, who do not have that type of sense of, this is where I want to go. I think that artificial intelligence can help us.
A very practical example of that is, for example, on our platform when we have a young person creating a budget, and the AI in the future being able to understand that budget and provide recommendations on specific goals around their money habits so that they're not going out for those pair of Jordans or PlayStation causing homelessness, which is things that we deal with.
Kind of like we heard before about personalized medicine, you're thinking about almost a personalized pathway for people to navigate through the system.
Absolutely.
It provides some sense of encouragement.
Absolutely
to them as well, and it helps the foster care system. When you're looking for missing or exploited kids, how does AI help to encourage actually the really hard work that you guys do? Looking for a missing child or an exploited child is not all fun and games. It's hard, emotional work. How does AI help you? How does it help you do your job better so that those kids are served?
Well, we're just scratching the surface at this point with AI. We do have some individual tools really that stand alone. What Intel is helping us do is integrate those that the analyst or the user is able to utilize the different tools in a much smarter way. At this point, we have many visions for what Intel is going to be able to help us with AI. Specifically, one really concrete goal that we have is decreasing the amount of time it takes for a report of a child being abused to come into us to go out the door. The volume is just ridiculous. Again, we're a nonprofit, and every year the numbers double. Last year it was 4 million, and we were pulling our hair out, and the year before that it was 2 million.
Now it's 8 million, we see no reason that those numbers are going to go down. Where we're looking for AI to be able to help is right now you have many analysts who are taking reports as they arrive, looking for information that may indicate that this one is at a higher risk to a child than another, trying to put those in packages and make the reports available to law enforcement in over 100 countries. While we're a U.S. company, we actually serve the entire globe because you have individuals exploiting children from everywhere. One of the concrete goals of what we really feel will be a huge success will be if we're able to take these leads that come into us, be able to prioritize using AI, have the system see things that I'm not going to see.
We talk a little bit about gut and where something just feels like this one might be more important than the other. We don't know how to define gut. We don't really know how to define why this seems like it may be more important. If we can teach an algorithm, if we can teach AI what to be looking for, really a learning model with feedback, be able to identify as cases are coming in which ones need to rise to the top. We don't want to make our jobs easier and say, "Well, great, took us a day. Here you go," and kick the can. We need to be able to make these reports better and prioritized for the law enforcement agency who's receiving them.
If I'm putting 300 reports on your lap this week, it'd be very nice if I could tell you which ones have the highest likelihood of a real child being abused. Those are concrete things that we're going to have to learn how to translate gut, translate clues that are there that we just really don't know how to identify into a tool that can really augment the human effort making a difference.
Again, sort of the idea of swarming almost, of being able to swarm around an idea with a particular set of technologies. I saw an article earlier that there was a student, I think it was at Baylor, who had been harassed, then 300 of her co-students actually gathered around her to escort her to class. I'm wondering, is there a way of thinking about how do you think about that kind of support in an online setting where somebody is just trying to voice an opinion or actually sort of reveal a little bit of an oppression or reveal some way that they're not being treated fairly? How do you think about that? How does AI help us? Could it, or maybe might it, or what's your hope for it?
Well, I think you point to what AI can't do, which is AI can't be that human support system. I mean.
Hopefully, AI can do things like recognize hopelessness or suicidality online. There are areas where AI can help or provide resources at the right time, I think social networks are getting a little better at that. Where I think the hope for AI around this is to identify hateful and harassing content. Far, social media is not doing a very good job of policing itself. Even companies that say, "We want to do a better job," too much is slipping through the cracks, and it's all reported after the fact, when the impact has likely been done on the person being harassed. It's still beneficial to remove that content, but it's probably had its primary impact on the primary person already.
It's important because the continued existence of it has a continuing effect on other people that might read it then be less likely to speak out themselves. I think one of the things, the Hack Harassment group is trying to work on this, is it useful and helpful if an algorithm can recognize before something is posted that it might potentially harassing? Is it useful to notify the person about to send that tweet, "Hey, this could be perceived as harassment. Are you sure you want to send it?" One, can the technology get good enough to identify it? Can it understand things like sarcasm and playfulness? Two, does it have an impact to share that with a person, or do people actually know whether they're harassing? I don't claim to know the answer to that.
I think the other piece is there a way to do it. One of my favorite examples, this was human moderated, but you can see where it would go into AI. When I was at CNET, we did have comments on our site, we had a way to block somebody. The most effective one that we had, blocked them so that their comments still showed up on their screen, but they didn't show up on anyone else's, which I thought was brilliant. This online troll somewhere in their little cave making their nasty comments. In their mind, their nasty comments are right there, and nobody else has to see them.
If you asked me what the world could look like with AI, I'd like to see a world where the AI is sort of ranking people on eBay, you know who to buy from because it's an online community, and things are ranked and rated. I think computers could have a role to play in really identifying, this person's only been on for three months. Basically making sense of whose comments are constructive and not. Obviously, when you're talking about speech, it's a very tricky area, and we don't have time to get into all of it, but an unpopular opinion isn't necessarily harassment. How do you dial the knobs? It's a very tricky problem.
Well, I think that's exactly where I want to go next. I think that We'll start over here at this end. You asked the question of what does it mean to be human in this digital age? What does our humanity actually mean to us, right? Especially all four of you are talking about having different relationships with the technology. Actually, I just noted on your profile page on Recode, you have a pretty extensive ethics statement, right? Of the companies you work for and don't work for and that sort of thing as you're reporting. I'm kind of wondering about what you think about the ethics of what it is that you're talking about. Where do you think it's going to go? Where does it have to go?
Recognizing protected speech versus harassment speech is an important thing. We have that in our living daily lives, but we don't have it particularly well on our digital lives. How do we start thinking about those kinds of issues? Maybe we'll start over there then move back this way.
Well, I think one of the scope of the challenge of integrating AI into our cultural and social systems is vastly underestimated right now. It will be a stunningly complex, contentious challenge. The reason it will be such a big challenge is that technology makes things explicit that are ephemeral in physical space. Meaning we have a conversation here regularly, it would be done. It's recorded now, it's available, which means it can be analyzed, you can code and tag different parts of the video, you can do natural language analysis of the conversations and conceptual understanding, a number of other things. That's only possible once it's been rendered in an analyzable format. That's what we have to face now. All of our assumptions, our legal system, what weight, which a judge would have as an implicit bias, that now has to be made explicit.
What weight do we put on race as part of a sentencing recommendation? When we have an individual that goes online and that has that little warning that comes up that says, "Maybe you shouldn't post this," there's a number of things at play that have to be identified. Everything that is now ephemeral and intuitive has to become mechanistic and explicit. We have to do that cover to cover. Pick one aspect of society from school to our social cultural institutions to our healthcare system, it's going to really make us confront ourselves because the implicit bias that we all function under has to be surfaced and confronted head-on.
I just want to emphasize, that's probably the biggest thing that I think we're not understanding in the AI conversation is that we have to have a long, deep soul-searching discussion with ourselves and what it means for us to be human in an age where we can pass a lot off to technology, but there's a lot that we probably shouldn't.
Yeah. The way you're talking about it, there's a relationship that you're building between the technology and the people, and the people and the technology. I think the way that you guys are talking about the technology, it's almost like a helpful tool, an additional tool, something that It provides a personalized pathway. If you have just enough information, it helps this one kid go through, or it helps your analyst identify, okay, that's where that kid is, right? That kind of thing.
Tony, can I just throw back at that a little bit?
Yeah.
Is there one area in society where technology doesn't become the alpha?
If you think about over a long period of time, technology always takes the alpha role.
I think that's interesting, when it comes down to healing trauma, I don't see technology, AI, being the thing that helps heal the trauma. That's why I think in our case, it's a tool because it's that human connection, that rewiring of the brain that happens from that interaction with the human that causes the healing. That's why I think in this, it may, but it may not.
Yeah. I'm with you in terms of, I think it's going to be very tricky. I don't even think we could tell the algorithms today. If I went around this room, maybe even if I just went around the five or six of us, I don't think we'd totally agree on what harassment is. I certainly know if we go in the room we wouldn't, and definitely if we go in the culture as a whole. I think, at best, it's going to be a very evolving role. It's not just a technology problem. I think technology can help, I think, certainly if the events of the last week have made me think about anything, it's like, what would the government consider harassing speech? Well, I think I know what the current government would consider harassing speech.
I think the next government might have a very different view that might differ greatly with some of the voices that are being harassed. I think that there's a long way before technology is able to, quote, "solve these problems." Even what I think we're more realistically talking about, where technology can be a tool to help solving them, it's still tricky, and it's not just a matter of Moore's law.
It's an interesting intersect where technology, or at least from my perspective, talking about child pornography, which is really a bad misnomer, in the sense that we're talking about child sexual abuse imagery. It's an interesting intersect where this crime has existed for a very long time. Technology changed the face of it. We've spoken with many victims and many of their families, and to hear the impact that technology has had on their healing is devastating. That individuals who experience a horrible traumatic experience of being abused, then have somebody photograph and memorialize the abuse and then share it online. That abuse never ends. That abuse continues every time that image is sent, every time somebody downloads the images. At the same time, technology has compounded this horrible event, technology is also the key to ending it.
It's this very strange circle that we've found ourselves in terms of utilizing new ways that the internet and basically all digital presence, what can be done in order to reduce the further trauma and allow for the healing of these particular children because, the problem existed before and technology has made it worse and it's also a big part of the solution. It's a challenge.
I think that's the parallel is the problem has grown significantly.
Yeah
those things existed in the non-digital world, their ability to magnify the ability of negative actions to have these outsized, longer lasting impacts, is tremendous.
It's certainly greatly magnified the problem. I think we're both hopeful that it can also speed a solution.
Well, I think, I'd like to go towards that end. If I think about George, you taught in the first MOOC, I believe. Right? The first massively online I don't remember what it stands for now.
Massive Open Online Course.
That's it. Courses.
Yeah.
Thank you. God, once something becomes a little term, it becomes a term, and that's it. You taught on the first one of those, and what you were doing, I think, was trying to make learning available to a wider array of people. What kinds of things do you think, and this is also adhered to you, what kinds of things do you think that people have to start learning so that we can actually start to address these kinds of issues?
Part of the challenge, I think, is the infrastructure isn't in place for the kind of reality we're starting to inherit. Meaning there's a lot of legacy pressures that we're bringing into this space with us. Give you an example. When I was at Red River College in Manitoba, this was in the late 1990s, and ours was the first college in Canada that went exclusively laptop. So the technology came in and did the work of old for the teacher. So instead of the projector slides, we had PowerPoint. What changed on that half of the classroom was everything. So I think that's the difficulty that we have is that the legacy inheritance of existing habits that can now be addressed with AI, we're not clear on what those might be.
I'm worried we're going to use AI to just keep teaching the way we've been teaching. One of the things I wanted with the first MOOC was to increase the capability of people to own their own learning. I felt that self-regulation, self-ownership, self-identity was critical, and it shouldn't be sort of given over to a faculty member to drive and shape for you as an individual. Now, as a result of some of those actions and activities, there's a whole set of technical skills that need to emerge. There's a number of infrastructure parts that have to be addressed, a lifelong portfolio of learning. We're developing a personal learning graph right now that's an attempt to have a stable identity that you own for life that captures what you've learned formally and informally. That's something that has to happen.
We have to have a school system that stops thinking in its current timelines, a university that stops thinking in three credit hours and instead says, "What do you know, George? We'll feed you content that fills the gap in alignment with the degree that you're seeking to pursue." That's not in place.
No, it sounds very similar to the kind of thing Oh, you're nodding. I think it sounds similar to the kind of thing you're thinking about, a personal pathway, a graph you called it?
Yeah, I like hashtag echo everything he said. I think you drove the point home. Yeah, I pass it on because you wrapped it up really nicely.
I think we have a couple of minutes left. I'd like to draw one thing I think that I heard just today in listening to you. It's that it's not only a human problem or human problems that are actually being resolved, but there's a notion here in the work that you're doing and your hopes for AI that you're trying to increase the kinds of compassion that we have for one another, not just in our own families, but across time and space, right? There's something there, I think, in what you're doing, thinking of it like AI is helping kids go through a particular program, or helping find missing kids, or helping stop Hack Harassment, or helping people learn, which is probably one of the more compassionate things that we do for each other.
I guess in the last couple of minutes, if you each had a little thing to say about how do you think that technology can help to increase compassion?
When I think of AI, I think of many different roles it can play, the first one being the ability for it to discover. Discover that pathway, discover something that we didn't know before, provide an insight. The second most powerful thing I feel like is the ability to intervene. Right? I lost two of my siblings, and I very much blame the failed systems, right? No young person should be able to grow up in a country that promises them that if you work hard enough, if you do your part, if you go to school, that you can be successful, and then have to die in their early 20s before that was even a possibility. To me, that point of intervention is so critical that AI has a role there.
We see an example in New York City with a young boy, six years old, came to the attention of the child welfare system six times, and no one in upper management was alerted, and that child died, right? There could have been some intervention. The last thing is, I think that when I think of the power of AI, it's the ability to have that systematic change based on the data. Discovering that intervention and then using all of that learning to figure out how does this create the new system.
It does something positive.
Sure. I think in terms of where technology can increase compassion is technology has the reach, technology has the ability to communicate and reach into corners that we never would be able to, and to tell them stories that they would not otherwise be aware of. Whether it be children who need homes, who need good, loving homes, whether it be in our case, in a situation of children who go missing or children who are sexually exploited, which most people are saying either that happens to somebody else in another country. For us to be able to communicate not only that the problem exists here in our backyard, literally in our backyard, but also that there's something that people can do because awareness empowers people to protect the children near them.
Where I see technology really is lifting the veil for the American public to recognize that this is not something that just happens elsewhere, it happens here, too.
The stories we tell ourselves matter.
Absolutely.
Just finally, Ray here really quickly.
Yeah, technology's ability, and I think we've talked mostly about AI, but virtual reality is one of the technologies I look at as a real tool for empathy, where you can literally step into someone's shoes. There's this online movie, "Clouds Over Sidra," that Chris Milk did that literally puts you in a Syrian refugee camp. It's a movie, but you feel some of that confinement. I think one of the keys to stopping people from harassing is making it clear that there's real people involved.
There's real people and there's real pain. There's also real possibility of doing really good work to actually get over all of that and maybe make us a better society overall. Thank you very much, panelists. Really appreciate it. A hand.
Wow, lots of hands.
Our time is up.
Done?
We have to go.
We have to go.
Get off stage.
One last round of applause for this great panel. I think we've all had a great and insightful and challenging panel to wrap it up. I hope that this was useful. On behalf of Intel, we want to thank you for attending. There's a reception with snacks and drinks out in the lobby.