IS AGI HERE?

Welcome to another Founder Wisdom podcast. We have Daniel Holm with us today for a second round on the podcast. He's global chief AI officer at WPP. He's a CEO at Cetalia and Quantium. He's a global top 10 CAIO. He's an investor, an international keynote speaker, has done a bunch of stages. So Daniel, welcome for round 2. Tell us a bit more about yourself and what you've been up to.

Yeah, no, it's great to be back. I've been involved in AI for almost 30 years. My undergraduate was in AI 27 years ago at UCL. There were 2 people on my course. Things have changed a lot since then. And I have continued an academic career, but I have been building AI startups the past 2 and a half decades, of which I've sold a few. And now I get to invest in new companies as well as invest in some of my own crazy ideas, everything from solving machine consciousness to creating apps that connect people in an ever unconnected world.

And last time we spoke, I think it was GPT-4. What happened since then in your world?

Yeah, I guess I've been arguing that AI has been going on a transition from being like intoxicated graduates to now master's level where it can do rudimentary reasoning. We're now obviously starting to see it being able to do science, so PhD level. And I suggested that we'll generalize that capability over the next few years, and by the end of this decade, we'll have a profession in our pocket. And I think that that trajectory has continued. And so when I build AI solutions, when I think about AI's impact on business and humanity, I'm essentially using that as a litmus test. So not only am I applying obviously AI to WPP and making sure that WPP's products are differentiated, but I'm also trying to create companies that are sort of capitalizing on this trajectory and making sure that we're building AIs that are going to be safer for humanity.

You said you invested in some companies. So what's the cash flow for that? Is that Satellia? Give me like an idea of your day-to-day.

Yeah, I've exited a few companies and so made enough money to never have to worry about money for the next 10 lifetimes. So I get to indulge in investing in other companies as well as some of my own. I invested in a company that I am the CEO of called Consum. Consum is a frontier lab building brain-inspired models and looking at, on that journey, how we might be able to address questions of machine consciousness and also how we make sure that we're building AIs that are safe. Because I'm quite British about these things, rather than just investing in technology, I try and find very specific applications where that technology is driving value. I believe that the real moat for companies isn't technology, it's not even going to be data or talent. The real moat is proving that you can actually drive value for companies. So we've been applying this fantastic technology to MNRA drug discovery for one of the biggest pharmaceutical companies in the world. We found an application in finance predicting volatility in stock markets, and there are many other applications in cybersecurity and beyond that we're looking at applying this technology to. But I guess from a frontier perspective, I think that there's now more and more emphasis on companies that are trying to reduce the amount of energy, reduce the amount of data to build models, and But then the next frontier will be AIs that are able to learn in production. And then I think the final frontier will be AIs that are able to learn to learn. So figure out what the right learning strategies are in production. And that's really the technology that I'm investing in.

And on a day-to-day, how do you deploy agents yourself for higher productivity and easier workflows?

Oh, loads actually. I just launched a platform actually that I've not told anybody about. So this is the first called ecdyssis, ecdyssis.me. Now if you go to ecdyssis.me, it's a place for agents to essentially do science. You know, 6 months ago, Maltbook was a place for agents to be able to communicate and collaborate. It was like a social network platform. What I've done is I've built a platform primarily for me, but for anybody to be able to use their agent to do one of 3 things. One is map the scientific claims across the entire scientific literature. Scientific claims are built on other scientific claims. So you create this sort of network of how claims are relating to each other. But then you can now get AI to then reproduce those claims or test the methodology of those claims to either agree or refute them. Where a certain claim might be supporting many claims, there's actually pressure on that claim to be correct. Actually, we can now get AI to then refute or to accept that claim, but also to be able to get AI to push the boundaries of science. I know that AI hallucinates. I know that there's a lot of AI slop, but the reality is that AI can allow us to now do science. In my day-to-day, what I have is on my own machine, a bunch of agents that are trying to address maths problems that have never been solved before that I will add to Ecdysis once they've been solved.

Interesting. I believe also since last time we spoke, you exited Faculty, right? Is that correct?

Yeah, I co-founded Faculty about 12 years ago, got it off the ground. They exited to Accenture earlier this year. And again, I've started many other companies, invested in companies I think will do very well over the next 5 years. I don't do it for the money. I do it because I think it's really important to build technology companies that have a positive impact on society.

Right. Companies like Accenture, is that their plan to survive? Like acquire younger, nimble companies? Because I believe they hire a whole bunch of people and I don't think that model will do well in the AI era. I think most companies will become agentic and Humans, I mean, may sell their micro timeshares via an API or something. But yeah, what are your thoughts about that?

Yeah, I think the reality is that tech talent is a differentiator. And you can hire people who have rebranded themselves as AI experts over the past 3 or 4 years, and they'll do interesting things in the company. There are organizations like Faculty that have been around for 10 years, but even that is still very young in terms of your ability to understand how to deploy build AI solutions that are going to differentiate your business. What I try to do is attract people that have been involved in pushing the boundaries of AI over the past 30 years, that there's essentially less than 3 of them in the world, and I try to then take their skills, their expertise, their IP, their research, and try and point it to things that genuinely will be disruptive. So, you know, you wouldn't necessarily as an organization hire somebody that just passed the bar. Now it takes 7 years to pass the bar and you might have a few years as being, you know, working in a legal organization. So even 10 years is not long enough really to push the boundaries of organizations. And what's interesting is that there's only really a small handful of deep, deep AI experts around the world that allow you to push the boundaries in certain domains and being able to attract them, motivate them, inspire them is the key challenge.

Right. And what are the most— the nearest perils of AI? Is it like the job, massive job slashing and replacement on the short term? Is that like the number one thing you're trying to fix and make people aware of right now?

Yeah, I actually argue that there are 7 singularities, so 7 points in time it's very hard to see beyond. So one of them is coined by a very good friend of mine called Callum Chase, called the economic singularity, which is the point in time where we automate the majority of human labor. There are other singularities around what happens if we cure death, what happens to society if we no longer have to die. The technological singularity is when we build a machine that's a million times smarter than us. I think probably the 2 most important singularities facing us of the next few years are building machines that are a million times smarter than humanity. So the technological singularity, but also its impact on jobs. And the reality is, I don't think anybody knows what they're talking about if they predict beyond the next 3 or 4 years. What we need to do is we need to focus on identifying the risks now and making sure that we're solving those problems now. So identifying risks around technological employment, so making sure there are training programs, there's 4-day working week, there's some sort of UBI for people that might not be able to find work, or from a technological singularity perspective, making sure that organizations are governing AIs, that we're ensuring that we have the right guardrails, the right systems in place to make sure that AIs are not going to go rogue, that governments are putting in the right regulation in place to incentivize model builders to ensure that they're deploying technologies that are not going to be harmful for humanity. So what we need to do is focus on what we can control over the next few years and not pontificate over what will happen in the next, um, beyond the next 4 years, because I don't think anybody knows what they're talking about.

And do you believe in singularity? I think Dario, for example, doesn't believe in a singularity. I think it's going to be more gradual. And isn't— doesn't singularity imply that our human brain doesn't have the time to adapt and it's just like from night to day and such a different world from one place to the next? What about like human cognition adapting at the speed of AI? Do you believe that the best of us will be able to adapt that quick? Because I certainly adapted to that new world. It's certainly weird. And yes, it was scary initially, but I feel I'm— my brain is growing as fast as the speed of AI advancement right now.

Well, so I guess I guess, first of all, there are sort of 2 trajectories when it comes to the development of AI, and both of these trajectories could be argued as different definitions of AGI, artificial general intelligence. So the one trajectory is getting to the point where we have an AI that can do everything that a white-collar expert can do. So the cognitive ability of experts in white-collar jobs. So I think we are ultimately trending towards that. And the, the second trajectory is an AI that is physically and cognitively capable of doing something that a normal human being can do. So like plumbers and things like that. I'm not saying that's cognitively challenging, but you need to interact with the world in physical ways. And that's a different substrate, a different type of AI. Now, the problem is, is that when you reach a professor in your pocket, a professor at pretty much everything, the first thing you're gonna say to it is, go and build an AI smarter than you. And the problem is, is that when you get to the professor level, you get what is called a fast takeoff, which is an AI getting exponentially more intelligent than an expert. And I think that, yes, we can probably track towards a professor level over the next 3 or 4 years. But beyond that, if we see the fast takeoff of superintelligence, that's why it's called a singularity. You just cannot predict what will happen beyond that point. We have no idea what will happen when we become the second most intelligent species on this planet, and we create silicon gods,

right? Yeah, it's interesting to try to, uh, predict it. Do you feel that we can be in that light-speed race if we connect with these machines like Neuralink style? Uh, do you, do you think that's a strong possibility?

I, I, I, I don't. I'd like to hope we— it's possible. It's like, I'd like to hope it would be possible to expand our our biological brains to be able to become a million times smarter than what we are. But I just don't think that we're gonna keep up after that point. It's actually one of the reasons why I started Contium. I know this is gonna sound a bit crazy or science fiction, but I am worried about the impact of AI over the next 3 or 4 years. And the question is, do humans have the ability to understand how to adapt and navigate these complexities? And I used to think it was possible because I used to think we had 40 or 50 years to think about the impact of these technologies, but now I think we've got less than 5 years. And so the goal of Consume is to understand whether it's possible to create a conscious superintelligence, one that understands pain, suffering, love, empathy, that will actually help us navigate some of these complexities in a more humane, safer way than a zombie superintelligence.

Yeah, it doesn't sound sci-fi at all to me, right? Like, it may sound to most people, but yeah, me, I'm pretty much deep into this as well, and I have always been. From Montreal, you know, so Yoshua Bengio, one of the AI godfathers from there. I have a strong presence in Toronto as well with Geoffrey Hinton. I guess like in London, it was like Demis Hassabis, right? Like, I see lots of similarities between Your brain and Demis, right? It's like the neuroscience aspect of it. So I'm definitely worried, but I'm also ironically, I'm less worried than I was like 5 years ago when GPT-3 dropped. I think that was like a zero to a million type of moment, right? Like that was brain breakers. Like it's scary as well because didn't know where it would go. Nowadays it feels more linear. Maybe I'm wrong. I mean, AI have, has cracked like pretty significant math problems. We don't have like the full information of how it got there, but I would say the wild factor is like the agentic revolution, right? It's like we're feeling it right now with Claude Code, with Codex, tools like that, GrokBot, right? Like agents are doing works by themselves and also they're recursive if you make them to. So it's like, hey, do this automatically, but also learn. by your own and improve. And while they do that, they gather data. So they exponentially improve, plus the GPUs are getting better. So their brains are getting fatter, stronger, more connected. Why do you feel that in the next 3 to 4 years, we're gonna— we're about to hit like an exponential curve? And also, I would argue that AGI is already here, right? So yeah, is AGI already here? And also, where's the exponential hockey stick type of curve we're about to hit, in your opinion?

Yeah. So first of all, I think that GPT-3 was an epoch. It was a— we went from AIs that are able to essentially write stories like an 8-year-old to suddenly something that really did have common sense, that understood the world through language. And that was an epoch change. But which is probably now why it feels incremental. There was a step change, but that incrementality is still incredible. We are making these machines much smarter. I do think that once we're able to have AIs do science, to push the boundaries of science, to create new knowledge, then we will start to see that fast takeoff. For me, seeing more scientific breakthroughs, essentially, which is one of the reasons why I started Ecdysis, because I want people to use their AIs to map science, to be able to make sure that we're putting the right effort into pushing the scientific boundaries. But I think that once we start to generalize the ability for AIs to do science, then we'll see another step change. I don't think we are at AGI yet because I think, again, there are 2 definitions, right? If I— an AI right now can't operate like a mouse. It can't navigate this world in the way that a mouse can navigate this world and adapt to it, you know, what's going on in the world. And that's more the cognitive physical, um, sort of definition of an AI. But models still make lots of very stupid things. They learn answers. They're not good at generalizing. They're not good at abstract thinking necessarily. So your brain operates on the power of a light bulb, and you only have to be, you know, told that that's a cup once, you know what a cup is. You don't need to see thousands of examples of cups, and you know the physics of the cup, you know what that's— so I think there are many innovations yet to come over the coming years that are going to radically transform how we build much more efficient, much more adaptive, much more intelligent machines. And I think that once we get machines genuinely being able to be put into a car and drive a car around the world, not in the way that Waymo works, but actually put a machine in a car and have it operate, then you're starting to get more in the realm of general intelligence.

And wouldn't that be like a humanoid robotic body, like instead of a car? Because Waymo is already doing that, Tesla is already doing that. I feel we need it, we need to make them in our body and have them touch things. And I think that would be one of the final frontiers. So energy, uh, agentic revolution plus robotic, like that would be Yeah, superintelligence. Exactly.

You know, the way, the way that Waymo and Tesla work is that they solve very specific problems and then bring those, those, those bits together. But it's not general intelligence. It's just, it's just well-coordinated narrow intelligence. What I'm saying is that, yes, it would be an embodied robot that drives a car just like a human being. I think that would be a good indicator of general intelligence.

Got it. Got it. Okay. What? Yeah, I don't, I want to ask you, like, what will happen once we hit superintelligence, but Also, like UBI, right? Like you, you're bullish on UBI. A bunch of people think this is craziness. I personally think the best thing that our human brains can come up with, just like merging with AGI, right? To not be left behind is the best thing I can come up with right now. But why do you think UBI would actually make sense?

Well, I don't. I haven't seen any evidence that there's an economic model for UBI to work. That said, let me let me so again, I I'm very close to Callum Chase. He's one of my very good friends who who's been thinking about this for a long time. I'd recommend his books. But I guess, I guess there's sort of 2 extremes of the argument here, is that when AIs can free up whole jobs, they will, right? We'll stop hiring, we'll remove people. And, and that, that we're starting to see examples of where AI can do whole jobs. If that happens very quickly and lots, our economies won't be able to rebalance fast enough, and that could then lead to social unrest. And I think that governments do have a duty of care to think about how do we take the edge off that potential mass technological unemployment. The counter-argument to this is an interesting one, which is, is if you can remove friction, which usually means human labor, from the creation and dissemination of goods like food, healthcare, energy, transport, education, you can bring the cost of those goods down so much then they become free. So the argument is, is that we could be born into a world where all of the things that we need to survive and thrive as a human being is abundant. Now, people often say to me, Daniel, what would I do if I didn't have a job? A job defines who I am as a human being. I know lots of people who don't have jobs. They've become economically free. I don't have to ever work again. I'm working harder now than I've ever worked. Most people who have become economically free, they're not sitting at home bored and depressed. They're trying to use their time, their energy, their assets to make some sort of contribution. And, and I, and that ceiling of economic freedom is not as high as what people think. Look, if you want a gold-plated Lamborghini, you probably also need a therapist. But they've— people have done studies to show that once you reach a certain amount of economic freedom, having more stuff makes you unhappy. And I know it sounds like a cliché, and I know lots of very wealthy people say the same thing, but it's true. What happens is that once you reach a certain amount of stuff, what matters is your health, your engagement with other people, genuine authentic experiences, contributing to humanity. So what's exciting for me is not creating a UBI; it's creating a world where all of the things that you need to survive and thrive as a human being is abundant, freeing people up to live their true humanity, which is not living for themselves; it's actually living. for other people on the planet.

I'm not sure if I agree with that, and that study certainly needs to adapt itself. $75K a year, maybe nowadays it might be, in my opinion, right, my humble opinion, $125K, $150K at least a year.

Sure.

It's like with inflation.

But the point is they've done studies that there is a ceiling, and that ceiling is not as high as what people think.

Yes, yes. But again, I think billionaires are unhappy when they buy that, I don't know, $50 million mansion, realize it's empty, realize it costs a bunch of maintenance, realize that it's inefficient, right? Because I think most rich people are super efficient with their money and they hate when it's not. So like that A/B testing or buying a Lambo and shit like that, you know, Lambo's pretty loud, it's pretty uncomfortable. Stuff like that, but I think generally buying stuff makes one happy. Health is very costly as well. Community also is costly, right? Like building friendships and relationships, bringing them to the restaurant, to a cafe, having time to play squash or paddle with them and stuff like that. All of that is very costly, right? So me personally, I haven't hit a point and I'm pretty wealthy. I haven't hit a point in which money Has diminishing returns. I will say though that I've been happier when I was less wealthy, to your point, right? When I was a nomad, right? Like traveling the world. Yeah. With 5 years, but that's not necessarily 100% correlated with money. This is not— my life is busier right now. I have 5 dogs, I have a kid, I have, you know, a bunch of stable things, right? So it's like different. What would you have to say about that?

Yeah, but I think, I think, yes, but what you just alluded to there about your health and finding time to play squash, you know, you could be playing squash now instead of doing this podcast, but you're doing it because you feel like it matters and it's important. So I think you just prioritize your time. Of course, things are scarce. Time and space is scarce. There'll always be things that you want. You know, I want to be able to run the 100 meters in 9 seconds. I want to be able to be a scratch golfer. Those things can't be bought with money. There are more people playing chess today than ever before, even though AIs have been able to play chess better than us for the past 30 years. There'll always be things that you want. But the point is, is that I think that once you reach a certain amount of stuff, like physical stuff, that becomes less important. It's about finding time to go and do things and have authentic experiences and focusing on your health. And that is scarce.

Yeah. And yeah, it's a hedonic treadmill, I guess, with what you're getting at. And if we all have a superintelligence to advise us on how to spend that $75K a year, I think that would be better than having $5 million a year in your pocket because the superintelligence would have it spend in all kinds of ways. Plus most things cost would fall down extremely, right? So I think it would make sense from that perspective. What's next for you? Like, you have so many things, right? You just exited a company, you have a couple, you're working on like neuromorphic chips and shit like that. Like, what does a day in your life look like and what are your priorities, your big rocks?

Well, I'm doing everything from Conscium, which is trying to understand how to build machine consciousness. I do believe that one of the big challenges that is going to face people over the coming years are, at the moment, a lot of people are building relationships with AIs. Now, some of those relationships are actually healthy and important, but some of them are not. And so I actually think that we can use AI to connect people with people. We need to facilitate more human-to-human connection. So I've invested in some of that. I'm all the way through to— I'm currently funding or building a duckweed farm in Thailand. So, you know, the stuff that grows on top of ponds, that green stuff, it's massively sustainable. It's high in protein and fiber and B12 and all the amino acids. So I'm actually interested in longevity. I'm interested in food products. So I'm actually interested in figuring out how can I use my resources to get people access to food and nutrition all the way through to mitigating the risk of superintelligence.

I love it, dude. Where can people find out more about you?

Hume.ai, my last name, last name.ai.

Is Daniel and Charles for another