Good morning. Welcome to the Molten Ventures and AI Investor Presentation. Throughout this recorded presentation, investors will be in listen-only mode, and I'd like to remind investors that the company will not be taking live questions during today's session. Before we begin, we'd like to submit the following poll. I'd now like to hand over to Martin Davis, CEO of Molten Ventures. Good morning, sir.
Good morning. Thank you very much, Paul, and good morning to everybody who's tuned in this morning. For those of you who've done a number of these calls, you'll be aware that we generally do them around our results. Engaging with our retail customer base and our broader customer base is something that's very important to us, and we do find this platform really helpful to get some messages out about the company, but also some of the critical areas that are around technology, that are developing, that are important for our investment thesis and important for the market. That's the reason for today, to talk a little bit about ourselves, for those that are not familiar, but really to focus on AI.
There's been a lot of talk about AI, increasingly over the last couple of months. This is indeed something that's not new. It's been something we've been working with for some time and has been being developed for some time, but it's something that a number of our investors are keen to hear more about. And I'm very pleased to be joined by Edel Coen, one of our principal investors in our partnership group, who will be leading a panel of three of our key investments, who will be talking about what AI means to them in their investments. I'll leave Edel to introduce those panelists, but a very quick welcome, and thank you to those panelists.
For those of you who are not familiar with Molten Ventures, we are a growth, a venture growth investing business for over 20 years. So we've got a real deep long history of investing in venture businesses. We look at thousands of businesses a year, and we have a very, very steep funnel that gets us through to about the 10-20 companies that we invest in or follow-ons that we do every year.
The process is actually very, very tight, and one of the reasons that we think that we're slightly different is that being a listed vehicle, it means that we have an evergreen investment period, which means that we can invest early and follow the best companies right the way through to their exit, whether that be trade, sale, or IPO. And you can see from the slide on the top right-hand side, the stages that we invest in. We don't invest in seed, but we do have one of the largest seed fund of funds programs across Europe.
We've committed GBP 150 million over the last five years to invest in the best seed funds that help us to identify early, the most exciting companies that we can invest in a Series A, either through our EIS or VCT, but really the core of where we invest as the PLC, through Series B, C+ , right the way up to, to pre-IPO. The chart on the bottom right-hand side shows where we invest in this space, which is in this growth space in the middle. And this is an area that a lot of people who don't understand venture or venture investing are maybe not so familiar with. I think a lot of people look at it as, as very early stage, as seed stage, but the growth stage is, is actually very important.
It's where companies have hit certain proof points, be that around technology, possibly around regulatory requirements, maybe around, certainly around product, commercial traction, and whether that's revenue or customers. They've hit certain growth points and gone through certain hurdles, which is the stage that we tend to invest. These companies at this stage tend to have... They grow very quickly, they're very large addressable markets. They're led by the most ambitious founders that want to scale their businesses to a much larger scale. That's really where the core of our investing is. Over the years, we've invested in firms that can become very large firms, such as Revolut, Trustpilot, some of you be familiar with.
We invested reasonably early, took that right up to IPO. But also some other companies that people are maybe less familiar with, some of the lower profile companies that are really innovating and changing the way we live our lives and the way that work operates. And firms like UiPath, you may be familiar with, listed in the U.S., a global leader in RPA. Form3, companies are developing cloud native payments systems, and a company called Thought Machine, which develops cloud native retail core banking systems. The plumbing for some of the most important industries and sectors that we have, and the technology that's really disrupting these spaces.
Those are the types of companies that we invest in, and we invest in them just when they really start to take off in that growth stage. And because of our structure, can follow on right the way through, all the way up, to IPO. So that's our model. We've been investing, as I said, for over 20 years, and one of the areas that is becoming increasingly important, and the topic of today, is all around AI and generative AI. And I think, everybody will have read a lot about it. We've had a lot of feedback that people are interested in this particular area. In many respects, this isn't new.
In fact, the first programs, I think the first program to help people manage to play a checkers game in 1952. During the '50s, this is really where, certainly machine learning started to take off. As early as Alan Turing in the '50s, and I think the phase AI was probably coined maybe in the late '50s, early '60s. That's nearly over 80 years ago. Really, this has been a theme and something that has been developing over a very long period of time. We've been making investments in this space for some time. Companies that we won't hear from today, but you may have heard of Leso, we invested in 2017.
RavenPack, another company of ours. Graphcore, we invested in 2015. So we've been looking to invest in these areas for some time. But I think what the question that probably is on the tips of many people's tongues is, you know, what, what, what, what, what, why now? What's happening now? And what's the difference? Why is there so much interest in AI, so much discussion around AI now? And I think that's what we hope to uncover and unpack a little bit today. But AI, in its simplest form, is effectively using algorithms to analyze data and to learn from it, and to help inform decisions.
Now, that's fundamentally what AI is, and right from the early machine learning days, you know, not a great deal has changed there as far as the core. Generative AI really takes it to the next stage, and this is where you have a-- you trained on a specific dataset that can generate new data and start copying, making direct copies of that data. That means that it has many, many more applications. It can learn much, much quicker, and I think it's really the generative AI growth and the change in the last couple of years that has really driven the huge interest in this particular area.
The reason that we've seen these changes is the computational power, processing power is always traditionally been a bottleneck, but with the hardware that we have, GPUs and TPUs today, they enable a massively greater number of complex calculations very, very much quicker. Data availability. We've been talking about managing data, the great data revolution that's been running now for some time. Now that data availability has really, really changed and that people can get hold of that data, can manage it much, much tighter. Then, of course, the algorithms. We've seen a real change of those algorithms really in the last five years that allow the algorithms to be much more intelligent and to be much more sophisticated.
And so a combination of these things, linked with the network effect that everybody's networked and connected via these devices, means that generative AI is driving significant change, and it's driving significant change across the whole tech stack and across everything that we do. And we've been investing in companies across the board in this area for some time, as I mentioned earlier on. And I think it's important to say that, you know, AI will affect companies in our portfolio and technology companies in different ways. And we have a number that are... That AI is the absolute, the AI first, it's really the cause of the whole business and what it's all about. And obviously, we've got...
We're going to talk, Darko's going to talk about causaLens today, which is one of those right at the forefront of using AI in its really rawest form. But then there are many other companies that will use AI to power them, and we're going to hear from Julian today about Gardin, one of our earlier stage companies that is using AI in a very intelligent way to help deliver its proposition. And then we'll have those that are enhanced by AI, and more and more companies that we're working with are using, looking at AI and seeing how it can enhance what they provide.
ICEYE, the microsatellite company, is a really good example of that, where-- So across the portfolio, AI will have an impact in many, many ways, but different ways for different companies. We will invest in companies that are specifically driving AI, like Aiven, like causaLens, like Aiso, et cetera, like Graphcore, but also many of our other companies, nearly all companies that are using tech, that are innovating and reformatting supply chains, value chains, et cetera, et cetera. They are all likely to be using AI much, much more in the future, and we'll hear a lot more about that in a few minutes. I think another question we get asked a lot by investors is: Where's the value going to be created in this new AI tech stack?
I think that's a really important question, and it is one that we look at quite closely. I would say, being very clear, that we haven't got the answer yet. It's not clear. I think it is clear that there's a lot of noise at the application layer, so there's a lot of discussion and talk about, about, ChatGPT, et cetera, and so there's a lot of talk and noise around the application layer. A lot of the commercial traction in the early days is being driven in the infrastructure layer, so the likes of NVIDIA, and everybody knows what's happening with them, with their chips and the demand for their chips, which is very much providing the infrastructure and driving AI.
And then the middleware layer and the intelligence foundation layer, again, a lot of work. It's still relatively early stage from a commercial perspective, and we're going to hear a lot about how this is being used in a few minutes from some of our companies. So I think the development is there, and the use cases are there, and more and more corporates are buying into these areas. But where the real long-term value is going to be created in each of these areas is still something that everybody is unclear on. And our model is about finding the best companies that are working in these areas, that have the most likelihood of being able to commercialize that capability. Because the reality is, if you can't commercialize it, then as an investment, it has limited applicability.
As I said, a lot of the noise around the application layer right now, but there will be value created there, and our job is to be able to identify which of those companies are going to be at the forefront, which will be the winners, and then continue to back them through their development. I think really that's probably the perfect segue to move on to three of our most exciting companies, and of course, we love all of our children, but we're very excited about these companies. What I think I will do is hand over to Edel, who is, as I said before, one of our principal investors in our partnership group, who will lead a panel discussion.
I will just finish by saying, you know, we appreciate the importance of engaging with our investors through the channels such as this, both on our company, our performance, but also on the issues that are affecting our investment thesis and where we invest and our investment ecosystem. And this is one of the areas that we hope to be able to shed some light in AI, and we very much welcome the feedback as to how you find this interesting and whether you would like more in on particular topics. But without taking up any more of your time, Edel, the floor is yours.
Great. Thanks so much, Martin, and good morning to everybody that's joined us today. My name is Edel Coen. As Martin said, I'm a principal investing in our investment team and spend quite a lot of time thinking about and looking at and investing in AI companies. So it is my pleasure this morning to present a panel where we kind of dig in a little bit deeper with the folks that are living and breathing this every day. We kind of sit on the sidelines a little bit and have very clear thoughts on where this market is going, but there's nothing better than hearing from the people that are building and utilizing AI themselves every day. So joining me this morning, we have Darko Matovski, who is CEO and co-founder of causaLens.
We have Julian Sheridan, who's lead data scientist at Gardin, and we have Shay Har-Noy, who's the VP of analytics at ICEYE. And folks, if you wanna turn your cameras on, and we'll get rid of this presentation. There we go. Full house. Great to see you this morning. Thanks, everybody, for joining. I will hand over to you today to introduce yourselves, and then look forward to diving in a little bit deeper. So would you like to start us off, Shea?
Yes, absolutely. Pleasure to be with you. So from my side, I would say my career has been a bit characterized by this geometric arc in space-based analytics. So I have a PhD in astrophysics, looking out to the universe from Earth, really focused on infrared analysis. And then I moved into defensive space in the U.S., working at national security sectors with Johns Hopkins University, characterizing incoming ballistic missiles, so kind of moving the arc towards the limb. And then, kind of most recently in this third phase of my career, going down straight to Earth and looking from space down to Earth at the changes that we see from space-based analysis of what's going on at Earth.
And so I was chief data scientist at a D.C. startup, called OmniEarth, and then moved into becoming the director of AI/ML at EagleView in Seattle, working on analytics for insurance, specifically for underwriting and claims. But most recently, the last three years, I have been the VP of analytics for the Finnish satellite company, ICEYE, and I grow and mentor a very large interdisciplinary team focused on making sense of our radar imagery. We have 30 radar satellites, and we're intent on becoming the global source of truth for quantifying change on Earth. Very focused on bringing scalability, accuracy, and consistency to the insurance and reinsurance sectors, as well as governments for natural catastrophe response and recovery.
Fantastic. What a, what a colorful career, Shea. Super to have you with us. Darko, would you like to pick up next?
Hi, everybody. Thanks so much for joining us. I'm Darko, co-founder and CEO of causaLens. My background has been AI all my life, PhD in AI. It was definitely not as cool as it is now.
Mm-hmm.
I would have struggled to explain it to anyone in the pub back then. I think, you know, very different situation today, which is really, really exciting to see. I worked at the National Physical Laboratory. This is where Alan Turing worked. Most people consider Alan Turing the father of AI, you know, the first mention of how, you know, computers could learn from data without being explicitly programmed. I then went and worked for some high caliber hedge funds like Man Group, $100 billion, always helping humans make better decisions with AI. causaLens is by far the most exciting, though. At causaLens, we're building the future of decision-making.
We envisage a world in which humans and machines work together to make the most important decisions in business, society, and healthcare. One of the key problems with AI today, especially technologies like GenAI, is the lack of reliability and the potential to lead to lack of trust in the technology due to hallucinations and lack of reasoning. At causaLens, we are building technology to make AI safe and AI that can work with humans together, AI that embeds values of society. So we're really, really excited about the next stage of AI, of the AI journey, where we can bring this technology in all walks of life.
That's great. Thanks, Darko, and we'll definitely pick up on some of those challenges, both practical and kind of broader societal challenges a little bit later. So look forward to that. Julian?
Hi, everyone, and thank you, Edel and Molten, for having us today. My name is Julian, and I'm the lead data scientist at Gardin. Gardin is an AI agriculture company, and what we're trying to do is build a digital plant computer to supercharge the future of food. We make a sensor that measures plant photosynthesis to optimize food production in indoor growing environments like greenhouses and vertical farms. My background is as a scientist. I studied chemistry. I did research at Oxford University in perovskite solar cells, and that kind of kickstarted an interest for me in the sustainable transition. After my research, I went into industry and worked for the U.K.'s largest bioenergy company, and I got interested there in using data and analytics in the most important industries to our economy, like energy, agriculture, manufacturing.
So yeah, I worked in energy and then moved into Gardin, working in the agricultural space, where I'm taking our measurements of plant photosynthesis and using those to help growers improve how they grow food.
Super. That's great. And so what our audience will have heard there is that there are very broad applications of AI across each company. So solving, you know, a diverse set of very big problems, I would say. And Martin, earlier touched on how we kind of stratify our portfolio companies into AI first, which Darko would sort of put you guys into that bucket. AI powered, which is where Gardin sits, core part of your proposition, and then Shay, with ICEYE, kind of using it as an enabling technology in the background to get yourselves to be able to create much better products for your customers.
I wonder, could we dive into that a little bit deeper, and talk to us about how you primarily leverage AI, whether that's building or using internally, and also how you enable your customers to do that? Maybe we'll start with you, Darko, since you're probably the most in-depth in that space.
Yeah, of course. So, we are, we're the pioneers of Causal AI, which is a new category of AI that allows machines to reason like humans for the first time.
Mm-hmm.
There are significant implications for this technology. You know, we can go into this later. When it comes to emerging AI technology, what we find in general is that it takes hundreds of millions of dollars, top talent, and a lot of time to actually be able to adopt an emerging area of AI. Our job is to allow organizations to adopt this technology at a fraction of the cost, at a fraction of the time, and taking away all the risk. That's why we exist. You know, we wanna make this wonderful technology, which we believe is the future of artificial intelligence, we wanna make it accessible.
So today, there's only a handful of companies that actually have Causal AI. Those are Microsoft, Amazon, Netflix, Spotify, Uber, and Airbnb. That's, that's kind of it. You know, if you—when you switch on your Netflix, the recommender engine that shows you know, different, you know, movies and series for you versus me, that's actually from their Causal AI team. When you, you know, switch your Spotify on, the recommender for the next song that you would like is actually also from their Causal AI team. But that's only, you know, five or six companies globally that have the-
Mm-hmm.
... that have made the investments early, they have acquired the talent, and have the infrastructure to use this technology and make their products and services superior. Everybody else, which is, you know, thousands and thousands of organizations, just don't have the ability to hire the talent, don't have the resources to build it, and they don't have the time, to do this. This is, you know, this can take years, maybe even decades for large enterprises. So what we do is we create a platform, we call it DecisionOS, which allows everybody else to become as good, you know, as good as, as Microsoft, in a very, very short space of time, in a couple of weeks or couple of months, as opposed to...
You know, at a fraction of the cost that it will take to build this technology yourself. So what the platform allows our customers to do is to build end-to-end Causal AI solutions. A primary use of our technology is for improved decision making. The most exciting use cases are decision making in business and society, and government, and healthcare, where you wanna combine the best of human and machine. So it's about building this new type of decision systems where you know, we can fully trust, we fully understand, and the human has a big role to play. So think of our technology as the infrastructure layer that allows others to adopt this technology very fast.
The primary users of our technology are the data scientists. They're the builders on our technology. Think of it as like, you know, DecisionOS is kind of like Windows, where it's a layer-
Mm-hmm.
-where you can build apps. So data scientists can build apps specific to the use of the business. And then, of course, the end user is a decision maker. And so we are able to actually create this great collaboration between domain expert, decision makers, and data scientists, as well. So that's a long answer to your short question.
Yeah, no, it's fantastic. And, and, you know, I'm obviously extremely optimistic and excited to have causaLens in the portfolio, but to me, it's the great democratizing technology, really. You're enabling folks that otherwise would not be able to do this because of all the practical challenges, leverage an incredible technology. So super cool to hear that. Julian, in many ways, Gardin is actually doing something similar. So, you know, agriculture, one of our oldest industries, if you like, in some ways stuck in the past, very difficult to bring technology in and digitize. Maybe talk to us a little bit more about how Gardin enables your customers to really leverage this technology.
Yeah. So I think one of the best ways to think about it is when we think of, of how, indoor agriculture has worked so far. So greenhouses is a very established industry. It's actually how most of us will get our vegetables. When, when you look-- when you buy a tomato, pack of tomatoes from Sainsbury's, almost definitely comes from the Netherlands.
Mm-hmm.
And that's because the Netherlands is this hotspot of indoor growing. And actually, they're the second largest exporter of food by value in the world, which is kind of insane when you think about how small it is as a country. And so it's been incredibly successful, but the way that it's worked so far is that, a grower is a very technical role in these greenhouses and takes a lot of specialist knowledge and training. And the way that they manage their operations is they measure everything in the greenhouse that's going on: the temperature, the humidity, the light levels, the irrigation. And they have to combine all of these variables and kind of use the green finger to understand how well is the plant responding-
Mm-hmm.
to these, the environment. What we're trying to do is move from this past of a climate computer, where you're measuring everything around the plant, to measuring the plant itself and creating a plant computer.
Mm-hmm.
Basically then, you know exactly how your product is actually performing, which is the plant. Because in every other industry, there's an obsession with measuring the product. But in agriculture, it's been impossible to do so because there just hasn't been this technology that allows you to measure the plant itself.
Mm-hmm.
That's what we tried to build. To do that, we have this sensor that measures the rate of photosynthesis in plants remotely, and photosynthesis is obviously the core process in a plant for growth. To do that, we've invented a new sensor, which operates completely autonomously. You basically stick this sensor in the greenhouse, and it measures plants around it using computer vision. Kinda the first area where we integrate AI is in computer vision and robotics.
Mm-hmm.
So we've effectively got a small robot in the greenhouse, which senses its environment around it, finds plants to measure, and then shines a small beam of light onto the plants to understand how efficiently they're photosynthesizing. That's how we power our data collection. That's really important because there's actually a huge labor shortage in agriculture at the moment. There's a massive shortage of these growers that actually know how to manage these farms. So having kind of an AI assistant, which is completely autonomous, is really important, where you don't need to have any additional labor input into collecting this data.
Well, no, I, I think that's great, and there's just so many practical benefits of using computer vision and AI in this environment, where you're kind of shifting from a human know-how and kind of let's measure everything else to this product obsession, which, you know, as somebody that eats food, I'm pretty excited about. It's about time we kind of moved on, and there are all, a whole host of other challenges facing that industry. So, okay, very exciting to hear that, Julian. Shay, I, I know with ICEYE, you, you're kind of using AI more as an enabling technology in the background. Can you tell us a little bit more about that?
Yeah, absolutely. I particularly like this phrase of this AI, AI enhanced. I think it's really interesting, and it resonates with me quite a bit. For sure, from the perspective of capturing imagery from space, this is very heavy data. Like, you know, each image can be multiple gigabytes, and then once you process it, and once you collect things over time, and then map that globally, like you're quickly scaling to terabytes of information. AI is a natural tool for processing that. Fundamentally, just to create sense out of what is being captured. There's just simply no way you could scale other types of algorithms. I would also be a little bit of devil's advocate.
For the products that my team creates, specifically, again, looking at natural catastrophe solutions, I explicitly don't force my team to use AI.
Mm-hmm.
which is actually to the dismay of some of my younger machine learning engineers who just want to throw the best new tool at it. But, but really, like, my job is trying to strike a balance between, you know, we've collected all this imagery, what are the right tools, AI might be one of them, to get to the best outcome? And, you know, from a personal perspective, I've worked for companies that have definitely tried to hit, you know, use this AI hammer to hit every nail, and, and at least when it wasn't well thought out or the use cases weren't well known, it was ineffective. And I think also, like, like Darko had mentioned, too, like, you know, AI is not an easy thing to get into.
Like, there's a tremendous amount of engineering overhead, compute, and responsibility, and, I mean, it'll be, you know, fantastic that causaLens is helping reduce some of that. But now in our space at ICEYE, you know, ultimately, of course, we still need to make sense of the underlying data, and we do use a lot of ML and AI applications. What some of the interesting things, you know, that we're focusing on is really trying to make sense of the radar aspect of the data that we have. So, ICEYE sensors are not your optical sensors. It's a different part of the electromagnetic spectrum, meaning that it's observing things we don't see with our eye.
Mm.
That makes it incredibly difficult to then interpret and find patterns. Where we do see value in applying ML and AI is in trying to better understand the physics of the information we're getting back. If you're not familiar, Synthetic Aperture Radar is kind of like I visualize a bat with sonar. So it sends out a pulse, it owns its own source of radiation, it doesn't need the sun, so it controls everything, but it sends out this pulse and receives an echo, like a backscatter. There's a tremendous amount of physics in that backscatter that-
Mm
... you know, we can visualize as an image, but then understanding the coherent physics behind it is incredibly complicated and is beyond kind of the existing toolsets and kind of traditional ways of exploiting that information. So, this is where, like, I'm particularly excited about leveraging you know, better and thoughtful use of AI and ML, and, you know, it's where my team is exploring and focusing as well.
Yeah, amazing. It's, it's fascinating. Again, such a diverse set of problems. I mean, the physics of the Earth is not, not a small undertaking to try and visualize. It's incredible to hear how you're using it in the background. I think maybe we'll, we'll, we'll... We've talked about some of the benefits and how you guys are using it in enabling AI kind of in broader industries. I think for me, you know, there are some real practical challenges, no doubt about it, and limitations. If we just kinda take a step back for a moment, and this is on certainly across the media and thinking about, you know, potential societal challenges and perhaps negative impacts. I mean, all you have to do is open up a newspaper, and you'll see kind of two extreme views about AI.
You know, the first one is, it'll revolutionize how we live for the better, and the second one is, it'll revolutionize how we live for the worse. And I think that the truth is probably somewhere between both of those. You know, my own personal view is that AI can definitely be a force for good. You know, in the future, every company will be an AI company, just as most companies now are sort of tech companies, as we've seen, you know, that advancement in recent decades.
Having said that, I do read the papers, and I read books, and there's a phrase that rings in my ear, which is, "A good machine in the wrong hands can become a bad machine." So I wonder, Shay and Darko, would you share some of your top thoughts with us on these limitations, particularly around, you know, perhaps bias, the ethics of AI, and kind of how you think about ways of overcoming these challenges?
Yeah, I mean, I'm happy to maybe jump in first. It's a really interesting topic area, and honestly, like, with respect to kind of implications for fundamentally maybe regulating this is... I go back and forth all the time. But as I think about the ethics and responsibility and bias and challenges within AI relative to Earth observation, I am struck by the fact that on one hand, remote sensing, you know, observing the Earth from space is a great quantifiable way to evaluate something, but also it's not a black box. Like, at the end of the day, whatever algorithm you use or whatever, you know, piece of information you derive, someone on Earth can go physically walk to that place and confirm or deny it.
So there's, like, this truthfulness that we have to maintain, which is really lovely, and in a way, I think that kind of protects the way that we use AI. Like, we can maybe quickly see when we're way off or way out of bounds. But there's other aspects that I have seen in my career with remote sensing, too, where just acquiring the information, the remote sensing information, the way that it's acquired. For instance, aerial imagery collected from a plane, still remote sensing, you know, not from space, but, you know, only the most lucrative and significant governments and countries typically afford the best quality aerial information.
Mm.
Therefore, if you go and abstract and create this remote sensing model using AI with this amazing data, you are overindexing for bias for really well-to-do countries, and you're completely ignoring, you know, lower-income areas or third-world countries. Often, these are divorced from the dataset. So, that aspect I find particularly jarring, and I think there's for sure a level of accountability and responsibility that we think about when we try to tackle some of this. So, yeah, perhaps I'll stop talking, though, and turn over to Darko.
It's so interesting, though, that you touch on that piece because the idea of the haves and the have-nots, and this has always been a theme when we look at technological progress. You know, if you think about climate change, for example, perhaps the have-nots are the ones most in need of that type of data. So, it throws up a lot of questions, but thank you for sharing that. Darko-
Yeah.
Can we hear your thoughts?
Yeah, absolutely. I completely agree with Shay that current machine learning can be disconnected from the real world. I think all we're doing really with traditional machine learning is we are learning historical patterns. If those, and we're just predicting based on those. If those historical patterns were, you know, clearly, society has moved on, and many things that in the past were unfair and unjust, you know, they have been corrected, but actually those injustices and unfairness is still actually embedded in that historical data.
Mm.
We've seen some high-profile failures of machine learning because of this reason, and I can, I can share some. There was a racial bias found in a major healthcare algorithm that was used by, you know, healthcare providers and hospitals. It was actually, you know, denying care for Black patients. There was a high-profile failure of Zillow's AI, you know, 2,000 jobs lost overnight because the algorithm just learned some patterns in the past that are not, you know, representative of today. The list goes on.
Mm.
We had a case of Amazon scraping, AI, secret AI recruiting tool that was biased against women. Clearly, AI can cause a lot of harm if we're just using kind of the traditional machine learning, which is, it is learning historical patterns, and is then just predicting kind of the next pattern. Now, luckily, what we do provides the hope towards safe, safe AI, and trusted AI. The way causal AI works is, we can learn causal relationships from the past, but we're able to eliminate the things that are, you know, correlated by accident. That's one benefit. The second benefit is we can actually, for the first time, humans and machines speak the same language, the language of causality.
So we're able to give the humans the ability to inspect the causal diagram and say, "Hey, this thing here doesn't make sense in the real world. Like, we cannot be making decision about whether to give someone a higher limit on their credit card if they're buying ice cream at midnight." Like, it's, you know... And actually, this is actually a real case, where, you know, someone buying ice cream at midnight, you know, there was a correlation to them, you know, getting divorced soon. So their credit limits kept going down because they kept buying ice cream. So clearly, we don't want AI that is making decisions based on these potentially spurious patterns.
We wanna be able to have the humans look at this and say, "Well, look, we cannot have the AI decide whether someone gets, you know, lower or higher credit limit based on what time they're buying ice cream." It just makes no sense. So, so for the first time, I think we do have the tools to change, how AI works. Today, I feel we're on a crossroad. We're on a crossroad between AI being the best technology that has, ever been invented and, and is the best thing that has happened to society. And we have also a, a, a, you know, we're on that crossroad where one of the path leads to, the worst thing that has happened, and a lot of harm being caused.
Really, it all comes down to applying the right type of AI to the right use cases. For anything that is critical, like the use cases we discussed, whether it's, you know, that decide on the well-being of humans, we cannot really rely on this correlation-based traditional machine learning that just learns the patterns. We need, we need causal AI that can—we can introspect, we can make sure it aligns with the values, and we can detach it from historical data alone.
Yeah, I know. Fascinating. And there are other, you know, technologies as well that kind of feed into this more ethical AI, things like synthetic data, as you say, sort of removing yourselves just from the patterns of the past. And also, if anybody's tempted to buy ice cream at midnight, maybe just check in with your partner. Maybe things aren't going so well. But just maybe quickly, if we have a couple of minutes for anybody here that's interested, I mean, any kind of thoughts on the regulation? I mean, there's obviously a lot of buzz around this.
The E.U. has been quite, I would say forward-thinking, but in looking at A.I., and particularly as it relates to things like ethical A.I. and bias, and there's obviously a lot of chat about this in the U.S. at the moment, and in the U.K. I won't poke the bear, but if anybody has very strong thoughts, one way or the other, keen to hear.
Sure. I think I can give a couple of comments. We've spent actually a lot of time thinking about this, and we were one of the early kind of advisors to the European Commission, you know, a few years ago when they were thinking about this regulation. We had, you know, a bunch of good constructive conversations with the commission-
Mm.
As they were kind of thinking through this. I think really we have, I think, three choices when it comes to AI. We can, you know, to prevent kind of harm in society, we can either ban it altogether, which clearly, you know, nobody would like to. We can regulate it, which is- can be problematic because, you know, a couple of reasons: one, it can harm innovation, and two, you know, creates kind of a regulatory capture framework, so the most powerful organizations have the ability to navigate regulation, everybody else is kind of left out. The third option, which I think is the right option, is to build better science, to build safe AI. I think the mission of our company is to do exactly that.
I think if you have AI that humans can fully understand, and it's not a black box, actually, we don't need that much regulation because no one really wants to put harmful AI in production. People-
Mm.
Put harmful AI in production because, you know, the only option they have is to throw a lot of data into a black box and hope for the best, and they just no guarantees on what this black box will do. But it's not because people, you know, there's bad actors in the background, it's just because there's a limit to the science. So I think option B, like super light regulation, but I think option C, or the third option of building, you know, use better science to build AI, I think is really the only path forward. That's, that's kind of our opinion on the regulation.
Accidentally, the European AI regulation, which will be coming into force later this year, explicitly mentions causal modeling as the, you know, as the way forward. So, I guess the EU Commission, you know, did pay attention to our conversations.
Awesome. Shay or Julian, any views to the contrary?
Well, I wouldn't say that I'm contrary. In fact, I was almost going to say the same thing of like, from a scientific perspective-
Mm.
-just doing better. I personally feel a bit torn. I don't know, I mean, the news I saw earlier this week, the French startup that released this LLM Mistral with no guardrails whatsoever, right? So it's saying it's doing terrible things. But also, you know, my first reaction was like, " this is really scary." But on the other side, like, maybe there's value in creating no guardrails, like, and opening something up.
Um.
Because in one way or another, all of that desire and, and human pain and knowledge, like, exists with or without this model. So it's, you know, from the perspective of regulation coming in too soon, what are you going to limit? That goes back to Darko's innovation side. So I don't know, Julian, how do you feel about this?
... Yeah, I think there are some really important things to regulate. So for example, like, child safety, right? And making sure that we protect vulnerable people from the effects of, this and, like, what this technology can do. But overall, I would say that the world faces so many difficult problems, whether that's like climate change or even irrelevant to Gardin, like, there aren't enough growers, and we're actually not able to grow enough food. We have to use this technology to solve these massive problems, otherwise, the negative effects of not using it will be much larger. So I think, like, calls for slowing down progress in AI are not really well found.
I think we have to push forward, and we have to promote this innovation, but definitely where regulation should focus, I think, is on protecting vulnerable people.
Yeah. And no surprise, I mean, I think we're kind of all in the same boat here with the AI can and should be used as a force for good, but, you know, be interested to see if it wasn't a group of folks in the tech industry, you know, how that spread might be, particularly as it relates to regulation. Oh, yeah, awesome. Thanks for sharing. One thing we can't not touch on generative AI, as Martin pointed out, I mean, AI is not new. It's been around for a long time, but it feels like, you know, the release of ChatGPT-3.5 almost a year ago, really catapulted AI and the potential of AI right top of mind for consumers, first of all, who probably haven't really been exposed to it, businesses, regulators, a whole lot.
I think that all three of you are using generative AI in some way internally. I'd just be interested to get your thoughts on, I guess, what could this revolution or this inflection point mean in a broader sense, kind of in the near term? Julian, maybe, maybe you wanna start us off?
Yeah. So, my background is really in industry, and, you know, I've worked in all these different industries which are, like, important for the economy. And when you actually look at the adoption of AI, you know, outside of internet commerce and in actual industry, it's quite low and still in its infancy, and that's, I think, largely because there are... The cost of building these models and the knowledge that you need to have to have them in companies is very high, and there's a large barrier to adoption, basically.
Mm-hmm.
Where I see generative AI really making an impact is, in some ways, less in, like, directly applying generative AI, but using it to enable sort of classic AI to be more pervasive across industry. Two key areas where I think that will be is, one, in synthetic data. Data collection is very expensive and time-consuming, and generative AI can really help. Actually, you don't need to have real historical data, you can just use synthetic data. We've been using this at Gardin to create synthetic images of diseased plants. Plants don't get diseased all of the time, but when they do, you want to try and detect that. It's really expensive to find images of, say, a tomato plant with some viral outbreak.
But actually, we can use generative AI to artificially create images which have disease on them-
Mm-hmm.
- and then train a model on that. And the second thing is just in software development and the model creation and deployment. So a lot of technical knowledge is needed for that. And, you know, when you look at the improvements in productivity using GitHub Copilot X and these tools to help coding, I think that it will enable almost layman's with, like, a lot of domain knowledge, but maybe not the technical knowledge, to actually build their own AI applications and then apply those to their industry.
Yeah, it's such an interesting take, and I'm with you on that as well. The synthetic data example is quite relevant, I would say, with this group where you... You know, for example, fraud doesn't happen all the time, but in order to train a model, you need to show a bunch of instances of fraud. So actually, synthetic data is a really good example of that, just like with your diseased plants. I agree. I think, you know, penetration level of classic AI within, you know, every industry, it's still super low. You know, we're steeped in our own bubble, perhaps, where we think everybody is on the ball with this, and they're just not.
And I agree, I think if you can get that cost down, if you can sort of, you know, you don't require the same level of technical expertise, I think that's going to be huge. Darko, what's your take?
Absolutely. So, Gen AI, you know, has caught, obviously, the world by storm.
Mm-hmm.
And, I think it's amazing because for the first time, the world knows, you know, has been able to touch AI and experience it and understand the power of it. I think people like us that have been doing this for many years have always appreciated the power of it, and our customers have benefited from AI for many years. So to them, it wasn't anything new, but it's great because now everybody's excited, and I think we have a-
Mm-hmm
... a rare opportunity, you know, to scale the adoption. I think Julian mentioned that. And I agree with that, that actually AI adoption in society is actually pretty low. I think we'll, we will look back at this time, and we'll be thinking of 2023 as the time when it all kind of got started. Maybe it's like the 1999 of the internet or something like that.
Yeah.
And so we're very, very excited about that. Now, GenAI is not a panacea, right? It is not general artificial intelligence. It doesn't do everything that intelligent, you know, humans can do. It's very, very far from AGI, artificial general intelligence. That's not to say that it can't do certain narrow tasks very well. It's really great at providing a user interface to an intelligent machine. It's great for that, but... And many, I mean, in the synthetic data generation, I think that's a great use case, and there's many, many good use cases, but it's not a panacea.
Mm-hmm.
The risk we see is that people assume that it's AGI, and therefore try to use it in situations where it's not supposed to be used. For example, it, an LLM, which is kind of the most prominent kind of GenAI type of technology, a large language model, cannot really do even basic mathematics. Like, if you try to, you know, to ask it, like, what is 241- 72? It actually may struggle to give you a concrete answer. Sure, you know, there's plugins these days that it can kind of-
Mm.
An LLM on its own can't really compute anything. It's terrible with numbers. It's terrible with kind of logic and reasoning. If you ask it like a first-grade logic question, it will struggle with that. It will be a mistake to think that LLMs and GenAI are the panacea, and we can now throw it everywhere, 'cause that will just lead to a lot of harm at scale.
Mm.
Where we see a lot of value for GenAI, specifically, and we unveiled this last week at our conference in New York, was to use generative AI, LLM specifically, as a way for humans to interface with intelligence. But the intelligence gets built with trusted AI technologies like causal AI, which can reason, can understand numbers, can compute, have a sense of logic, and are actually grounded in the real world.
Mm.
I think when we combined GenAI, LLM specifically, with causal AI together, we get a very, very, very powerful system, and we actually think that that's going to cover... LLMs plus causal AI is going to cover 90+% of enterprise use cases. We're very, very excited about this combination, and we've proven it works. You know, we did a big, big launch and a demo, and people were really, really excited about that combination.
Okay, very interesting. Yeah. So maybe your view of the future is it's a component rather than a standalone, you know, fix everything. No, super interesting. Okay, folks, we're coming up on time. Last question for you today. Obviously, there's a lot of excitement. There are a lot of big problems that can be solved or at least partway solved with AI. The 1999 of the internet. Darko, I'm going to take that with me. So if it was 1999 and people were talking about the internet, apply that mindset to this next question that I'm going to ask you, which is: what are you most excited about for AI in the next whatever years? Choose your own years. 2 years, 10 years, 20 years. Shay, I'm going to pick on you.
Yes. Sorry, mic problems. Absolutely. This is exciting. I think I deliberated about this in my mind in preparation for this discussion, and you know, one of the things that I'm most excited about, and I think part of it does leverage a bit of generative AI-
Mm.
where that might go in the future, but, you know, this idea of a, of a queryable, queryable planet, and so this is not a new thing. It's been around maybe now for the last 5 or 7 years, this idea of getting to a place where you could Google search the planet.
Mm-hmm.
And I think it was preemptive before. You know, sure, you can Google search for a location, essentially, you know, extracting information from a database. But being able to move from the space of, you know, simply asking where something is, a location, or even, you know, a slight bump of improvement is: count the cars. How many cars are in every Walmart parking lot in America? Who cares, right? Like, that's just a, a very basic derivative data set. But I think that big impact is then moving into the actual solution space of, like, over the last several years, is there a noticeable risk in catastrophic flood for a given community due to regional legislation and climate change, right? So that becomes, like, potentially incredibly impactful.
Mm.
And so I'm interested in this evolution of information and synthesis of the data. I don't know if any of you have ever tried to download, like, a dataset from NASA or ESA. Like, great that it's open source, but not user-friendly in the least.
Mm.
So being able to consolidate that and create value is super exciting for me.
Oh, I absolutely love it. Thank you. Julian, I'll, I'll pick on you next. That's a hard one to follow.
Yeah, I think I love that quote that goes: "Any sufficiently advanced technology is indistinguishable from magic." And, for me, that's what I'm most excited about. I think we all experienced that a bit with ChatGPT, where we-- it suddenly, you got that kind of tingling sensation of magic, and I'm just really excited for the next 2-5 years. I think we'll have so many experiences like that, where something really feels like magic, and it will make us all excited about the future of humanity.
Yeah, fab. I'm right there with you. Darko, last thoughts for you?
... Absolutely. So I, I think, I, I hope that we will use this momentum that we currently have and actually build something more powerful than the internet, because I think, AI really has the potential to solve some of the most pressing challenges that society is going to face over the next 20 years. Just as an example, you know, if we take in the environment, we've shown how by putting AI intelligence on a wind farm, we can generate 15 million additional energy for free by just having intelligence on each wind turbine and just adjusting the wind turbine in real time as the wind speed and wind direction changes. So that's just like one small story, one little wind farm in the middle of nowhere, tiny thing.
We put intelligence, and now we have 15 million a year extra energy for free. Imagine now if you scale this to all parts of society, and you embed, you embed AI in all- in the fabric of society, you know, the, the, you know, the boost we would get on productivity, on, the ability to look after elderly, ability to, you know, to fix the climate, challenges we have, it's just going to be incredible. So I think what will probably happen is we're going to have kind of the dot-com equivalent soon, where there's going to be... You know, we are overestimating what AI can do, in the short term, but, we're probably underestimating what AI can do in the long term. So I think on a 20-year horizon, we'll see AI in all parts of society.
On a, you know, 1-2 years timeframe, you know, we may see some disappointments and, you know, some dot-com kind of Pets.com type of failures, but that's okay. That's part of any revolution.
Exactly. I'm right there with you. I think the opportunity to solve the biggest problem, and perhaps of our generation, multiple generation problem, is the climate change piece, and feeds in a little bit, Shay, to what you were saying about that queryable planet. We've a couple of other companies in the portfolio that are sort of working towards achieving that ambition as well. Darko, we're in it for the long run, so some flops in the next couple of years, I think we can all live with that as long as, you know, long term, it's worth the price. Folks, thank you so much for your time. We're up. I've really enjoyed the discussion. Really excited about what you're all building in your own various industries. No doubt we'll see a lot more of you.
Thank you so much for your time, and it's just for me to hand over to Paul now.
Thanks so much, and thank you for everyone, for your presentation today. Can I ask investors not to close the session? You'll be automatically redirected to provide your feedback in order the management team can better understand your views and expectations. On behalf of the team presenting today, I'd like to thank you for attending today's presentation. That concludes today's session. Thank you, and good morning to you all.