The Human Code
The Human Code" podcast unravels the intricate blend of technology, leadership, and personal growth, featuring insights from visionary leaders and innovators shaping the future. Host Don Finley dives deep into the human stories behind technological advancements, inspiring listeners at the crossroads of humanity and tech.
The Human Code
Exploring the Future of AI through Quantum Lenses with Dario Villani
Quantum Cognition: Redefining AI's Future with Dr. Dario Villani
In this episode of The Human Code, host Don Finley interviews Dr. Dario Villani, co-founder of Qognitive, Inc and CEO of Duality Group. Dr. Villani discusses his journey from a theoretical physicist to a leader in AI and quantum cognition. The conversation explores the limitations of traditional AI models, particularly large language models, in achieving general artificial intelligence. Dr. Villani asserts that true AI requires embracing human cognitive fallacies and leverages quantum cognition for contextual reasoning and abstraction. Highlighting the inefficiencies of the current AI development path, he stresses the potential of quantum cognition to overcome these limitations, presenting it as a promising avenue for future AI advancements. The episode concludes with advice for entrepreneurs to find their unique voice in the tech industry.
00:00 Introduction to The Human Code
00:50 Guest Introduction: Dr. Dario Villani
01:51 Dario's Journey and Career
03:59 Understanding Quantum Cognition
08:06 Challenges and Insights in AI
21:06 Applications and Future of Quantum Cognition
30:11 Advice for Entrepreneurs
32:42 Conclusion and Sponsor Message
Sponsored by FINdustries
Hosted by Don Finley
Welcome to The Human Code, the podcast where technology meets humanity, and the future is shaped by the leaders and innovators of today. I'm your host, Don Finley, inviting you on a journey through the fascinating world of tech, leadership, and personal growth. Here, we delve into the stories of visionary minds, Who are not only driving technological advancement, but also embodying the personal journeys and insights that inspire us all. Each episode, we explore the intersections where human ingenuity meets the cutting edge of technology, unpacking the experiences, challenges, and triumphs that define our era. So, whether you are a tech enthusiast, an inspiring entrepreneur, or simply curious about the human narratives behind the digital revolution, you're in the right place. Welcome to The Human Code. In today's episode, we are honored to welcome Dr. Dario Villani co-founder of cognitive Inc and CEO of duality group. With a remarkable career, spanning finance, theoretical physics and artificial intelligence. Darrio brings a fresh perspective on the future of AI through quantum cognition. Today, Daria, when I will share how quantum cognition is revolutionizing our understanding of human learning and how it differs from traditional AI models. Y large language models have fundamental limitations in reaching general artificial intelligence and what it will take to go beyond. The critical role, that insight, abstraction and contacts play in. Creating truly intelligent systems modeled after the intricacies of human cognition. This conversation will challenge the way you think about AI and where the future of technology is headed. Don't miss this fascinating discussion with one of the thought leaders pushing the boundaries of artificial intelligence. I'm with Dario Villani and the man's got a wide variety here. I gotta say, Dario, it's a pleasure to have you on the show. And the first question that I'm going to drop to you is what got you interested in the intersection between humanity and technology?
Dario Villani:Thanks for having me. I'm Italian. in Italy, there is a bias that if you're really smart and intellectual, it's all about quoting something in Latin or in Greek or in philosophy. It's the tradition of natural sciences. So, you you're struggling between trying to be a philosopher or being able to write poetry is writing well. And being smart with the girls or in school with physics and mathematics. So it's natural given where I'm from.
Don Finley:Interesting. So you just have a natural kind of like curiosity about this intersection. And then, your background it looks like you've run a number of businesses, and are the chief executive, many of them, it looks like you're proficient in startups as well. where does that come into play when we talk about the, this intersection?
Dario Villani:Yeah, very briefly, I came to the U. S. in 97. So it's been 27 years I'm in the U. S. I came as a post doc in theoretical physics. I left, then eventually went to banks. I started my own hedge fund together with my business partner. And, most recently I started a company in quantum cognition, it's called Cognitive, where we try to push a little bit forward how human learning actually happens. And, I would say a lot of the journey, when you come from science, you don't want to be pegged as the quant, especially when you have an accent. You dream of being the macro trader, drinking scotch, being like the main character of Wall Street. And on a whim deciding rates are going to go up or down. But then, life is full circle. You go back and you start thinking there is a lot of untapped value in the skill you had. And now you have also a sense of what macro trading is and what risk reward is. So we started a hedge fund that was quantitative and driven by machine learning. Microsoft And then the adventure of cognitive, it is quantum cognition that I think is spectacular.
Don Finley:So let's dive into the quantum cognition piece of this because, one, a quantum comes up in any number of conversations around whether it's physics, whether it's around quantum computing, other spaces that I've also heard it just used in the dialect of it's the next level kind of space or, looking at that you mentioned that quantum. Cognition was also focused on human learning as well. what is the goal? What is the problem that you're solving with quantum cognition?
Dario Villani:there is a whole field. It's a field we didn't invent. That is called, essentially quantum cognition. Where a bunch actually sits in psychology department. Where the thesis is that some of the cognitive fallacies of humans can be explained essentially postulating, assuming, or testing the fact that the way humans learn is quantum in nature. And it's related to things being non commutative, like positional momentum of the electron. Our, breakthrough is to be able to represent datasets as a quantum state. And you have gigantic economy of representation. And, just to give you the context, there is a gigantic wave out there of people that believe that you start from a cogent, coherent statistical framework It can be machine learning, all the traditional statistical methods, neural networks, and so on. And you add enough data or models, you can bootstrap yourself to what people call general artificial intelligence. So it's a matter of having enough computer power, enough ability to scrape data, enough ability to add models, and eventually you're going to get to human like behavior with the ability to abstract, the ability to have context dependent reasoning, and all of that. We, have actually the opposite view, you need to start from a system that has all the cognitive fallacies of a human. there is a limit. In which the human becomes a statistical system. So there is a beautiful sentence actually in a lecture that Turing gave, I think in 47. And he said, if you expect your machine to be infallible, it cannot be intelligent. So think of the analogies in physics. If you start from a Newtonian system, it doesn't matter many springs and balls and whatever you add. The classical system is never going to behave Like with a wave uncertainty If you start from a quantum system, there is a limit, it's called the semi classical limit, in which the quantum system becomes classical. A little bit like in general, in theory of relativity, at low enough speed, relativity becomes Newtonian. There is no limit in which Newton physics becomes relativity theory. So, you need to embrace the cognitive fallacies of humans. And, we actually worked out the limit in which we get back the statistics. Instead of starting from a statistical system and thinking that adding a lot of data models you can get human like intelligence. So, it's very different. there is a bunch, we are not unique in this, there are a lot of other people that believe that AGI, just fluency of large language model doesn't mean intelligence, doesn't mean abstraction. But there is something special that belongs to quantum. in quantum probability, in classical probability, the uncertainty of the whole is always bigger than the uncertainty of the parts. The quantum framework is the only one where you can have uncertainty without entropy. That is the key to abstraction. So I always tell people that if you do a large language model, You train it long enough, 1 plus 1 is 2, but if you don't train it enough, 1 plus 1, you can say 3, I love you, or the capital of Italy is Rome. there is nothing crisper than 1 plus 1 equals 2, where that 1, what it refers to, you have no idea what it is. So it's the maximum uncertainty about the object you're discussing, but also it's the Crispus, statement you can make That's only possible with the quantum framework.
Don Finley:Oh, that's incredibly interesting. and we're in a hype cycle right now for LLMs. it's driving so much activity, but I was also reading some news just the other day where people were saying that, I can't remember the guy's name who said this, but he was basically making the point that Our sole focus on thinking that LLMs will drive us to AGI is misguided, for one, and that possibly that just because we've been so hyped on it that we've set AGI back by five years. And I think one of the challenges that we have with LLMs today is they're good at essentially reciting information that it either already knows probabilistically in some sort of like domain space, but it doesn't solve new problems. it's, not really advancing outside of its own, understanding.
Dario Villani:I believe it's a huge, technology breakthrough, but I have to say, this is I wanna be a little, out there. You said at the beginning, and I want to go back to that, what makes you at the intersection of, humanity and technology or physics? a lot of the traditional scientists, think of Gödel or, Einstein. they also studied, like the CEO of Palantir. They studied Goethe. They studied Spinoza. They asked themselves deep questions of what intelligence is, what abstraction is. I'm always a lead, and that's the tradition, at least in Italy or in Germany and so on. I'm always skeptical of pure technologists that they've never read the A book of Gator Spinoza and so on opining on intelligence and abstraction just because they get 80% success rate on some benchmark they come up with. And, the depth of the work of Goodle is very much connected to our breakthrough. It's, the countable, safe and all but without going into detail. There is something deep that I think it's missing when people say, Oh, I got 85 percent. If I get 89, I'm approaching it. And even the concept of what general artificial intelligence is, what is general anyway? we rather solve specific problems well. And, of course a lot of people tell me, can you do better than LLM? And I was, do you really believe that if you put all the data up until 1905 of every book on Ether and Newtonian physics and Maxwell and you put it to an LLM, you would have come up with General Theory of Relativity? I don't think so.
Don Finley:I think you're hitting on a key point of where that,
Dario Villani:they are raising a lot of money, so one thing I've learned from finance, because finance is a business of people that you deem to be much less smart than you become a billionaire and you grow a lot of resentment about it. So my take is not to fight people or try to show the golden proof that AGI is unreachable because I don't even know what they mean by AGI. We rather do our thing well, build tools to solve real problems and time will tell who is right.
Don Finley:I agree with you. the definition of AGI is going to morph over time. Look at what's happened with the Turing test, How many times does a new technology come out that passes the Turing test and yet doesn't resemble anything close to what, AGI is? Or, we'll do the thing that we've done with, other human races over time, we'll qualify that they are either part of this category or not part of this category. We'll look at animal cognition as well, we've separated the idea of consciousness out. So that it is either, part of humanity or not. Yeah.
Dario Villani:the one thing of quantum that is beautiful, and that's why it's very much like humans, apart this technical thing of entropy and uncertainty, is that generally when you encode data, you have an underlying distribution function and the features are numbers. In our case, we don't encode data like that, and features are not numbers, they are operators. So my age is an operator, like the position and momentum of the electron are operators. There is an expected value, and if operators don't commute, you can't measure them at the same time. So, what that allows, it's if my age is an operator. There is a context, imagine a social context, a cocktail party, where my expected value in that context is 80 years old, because I believe like a toddler sometimes in social settings. But if you look at my health, maybe my expected value is 60. So it really lend itself to a very natural way of doing context dependent reasoning, context dependent assessment of what the situation is. If data are encoded with numbers, you don't have that flexibility, the same way you wouldn't with the positional momentum hope you just don't walk away thinking you just sound like an 80 year old, but
Don Finley:No, I think who is it that said, if you ever understand quantum physics, then you really don't understand it.
Dario Villani:Oh, that's very much true.
Don Finley:yeah, it always seems to surprise me, but at the same time, like you're saying that your age is an operator, just spin or momentum. Position, momentum Yeah. and
Dario Villani:it's an operator, it's, it's a matrix in a Hilbert space. You can compute expected value depending on the state. And, of course, depending on the state, the ground state, or whatever it is, You can have different numbers. Instead, That's what allows the economy of representation. one example I always like to give is that we beat the curse of dimensionality because our complexity doesn't scale exponentially with the number of features, but scales linearly. So if you want to assess the probability of a hostile task in a year Generally people give features like sex, age, alcohol consumption, beer, body fat, pressure, and all of that, And very quickly you get to hundreds of millions of beans. if you have sharp categories like it happens in statistical learning. In our case, instead, it's not like that. So you embed all the information, you jam it in a much smaller space and you can, do the same analysis. You don't need billions of data points you're never going to get. And also, the issue is even if you had, like when you do financial markets. If you add data going back to the Mesopotamian time, that state is not relevant. Like the ability of a rast attack is not relevant of people of 5, 000 years ago. That different diet, different life expectancy, blah, blah, blah. So it's important to be able to deal with gigantic number of features with a limited amount of data, very much like humans learn. once I teach you 1 plus 1 equals 2 and y is 2, you don't need to train for, a billion times to be able to say that 17 plus 12 is 29. You have,
Don Finley:okay, so that is one of the problems today of, training LLMs is the cost associated with it. the repetitive nature of the data in order to get that one plus one equals two component. And then additionally, the inference cost, is continuing to climb, so even if we reach human levels of intelligence, we'll still probably have, will be off by an order of magnitude of maybe two, three on the actual like cost to create that. So the amount of calories consumed or watts consumed, are you saying that with quantum cognition, we would also be able to, effectively train on, you just don't need it. It's just like human.
Dario Villani:you have insight. You have an ability to abstract. And the insight is connecting the dots. You the issue of the LLM, as I said before, if you train it halfway, you don't converge. It can tell you anything. One plus one is, Salerno is 30 miles out of Naples, you can get anything. And there is no real insight. That abstraction is really the ability to essentially integrating out all the possible instances of what that means, but having a crisper statement that is only possible with quantum probability theory where you have maximum with zero entropy. That's what abstraction actually is.
Don Finley:okay.
Dario Villani:but, we'll see. We are very confident. Listen, we have done great things. We are doing great things. We started just seven, eight months ago, the time scale of which this company go up or down, it's very fast.
Don Finley:Yes.
Dario Villani:We have a beautiful joint development agreement with IBM. We have incredible sponsors and investors and, Ribbit was the lead, and Jim Palotto from Raptor, and there have been, Morgan Stanley. we have great people supporting us, and we think we are on to something special. the one thing I want to say, there are a few people who actually are in our camp. They don't believe AGI is reachable. They don't think the statistical learning, it's the way human works, like pencils or
Don Finley:Yeah.
Dario Villani:but, they stop short of providing a solution. I think, we believe that. There is a whole train of people doing large language model. But, I'm sure they can do amazing stuff. But, that ability to abstract and for reasons that are connected to Gödel's theorem, it's just not in the cards. It's not just adding another 10 zillion parameters and 500, 000 models you're gonna get there.
Don Finley:I think We will find the limits of LLMs,
Dario Villani:yeah, But also that's the big thing of quantum theory. even the simplest of objects, like the hydrogen atom, the only way you get the energy spectrum being discrete or quantized is the Schrodinger equation. You can't get it otherwise. So it's a different, like you can do incredible things like, with Newtonian physics, you can go to the moon. not that it's useless it's superconductivity is not Newtonian physics
Don Finley:like Newtonian physics isn't going to give us GPS, Like we would lose the time tracking of the satellites because of the effects of general relativity. If we. failed to account for that,
Dario Villani:You know that ability to absorb the limit humans Have with all the cognitive. It's like in finance Like people say I want more returns. you need to have also more volatility, You So intelligence, the other side of fallibility and cognitive fallacies. That's why I said at the beginning, Ture, you, there is no free lunch. You can't have a cogent, coherent statistical system and just adding stuff on top of each other in a Bayesian way or whatever other statistical way you get. Insight. Abstraction. No, we have insight because we are also evolutionary. We are very flawed. We have cognitive balances.
Don Finley:and one, my favorite mathematician is Gödel. So like absolutely endear that man. and I don't think we could ever get to an omniscient AI. Like the omnipresent AI because of the limitations of Godel, but I think you're also offering up that we may not be able to get to AGI because of Godel as well. but you're offering the solution as like the quantum cognition piece of
Dario Villani:Yeah, encoding data is a quantum state.
Don Finley:Yeah. And I think that's a really,
Dario Villani:I think it's an incredible, actually, insight. of course, when we talk to people for the first time, some people say, Oh, and, it's very separate from quantum computing. Like, once you encode data as a vector in a Hilbert space and represent operators as matrices, you can implement it on classical computers. But of course, you can also, it's natural to do it for quantum hardware. And that's interesting because the two big things in the world now in technology, one is quantum hardware, it's a national security issue and so on, and the other one is inference and machine learning. And people, whenever you ask them what they're going to do with quantum hardware, they either mention cryptography or quantum chemistry. we're saying actually there is a new way of doing inference. And it's natural for very large number of qubits, you can have real advantage there. And for small number of qubits, you can have a completely different representation, and it doesn't suffer from the cost of dimensionality. So
Don Finley:are some of the use cases that you're looking at today for this? Because yeah, you're in a hot market and any bit of reasoning or agentic kind of like thought processes that could be used in AI will really make your investors very happy. I could imagine, but as well as your clients too.
Dario Villani:problems with huge number of features, a limited amount of data, unfortunately it's genetic. Healthcare, and we're doing that with some companies. Finance, measure of similarity. recommendation engines. Banks. So we have dialogue with all of them. And one of the biggest thing we have discovered is like the assessment of dimensionality of data set. Like whenever you have Imagine a hundred features and you have data lives in this hundred dimensional space. Everybody knows the data actually lives on a manifold of a smaller dimension. now what's the dimension of that manifold? What's the geodesic? What's the minimum distance? It's a measure of similarity, how similar are two bonds? How similar are two stocks? How similar are two consumers? And quantum is the only way to make an assessment of that dimensionality that is robust to noise. So other techniques exist, they give some decent results, but if you put noise, they create shadow dimension. So Imagine Earth. Somebody from out of space comes and wants to understand the shape of the Earth. And of course, they start probing, they get Everest, they get some depth of the ocean. Think of it as a noisy sphere, So the fact that there is these points are not exactly on a perfect sphere. Most technique dimensionality to be 3. So if you think the dimensionality of the Earth is 3. Basically, the distance between the North Pole and the South Pole is the diameter of the Earth. you instead ignore the noise and you say it's a sphere, well, the distance between the North Pole and the South Pole is the semi circumference of the Earth. It's a very different measure of similarity. And that's very powerful for recommendation engines. It's fun. Which patient is very similar to another patient? Generally, we bucket by age. I'm 53. So every patient that is between 50 and 55, that has this blood pressure between 120 and 140, and has this body fat, blah, blah, blah. That's similar. That's not context dependent. And it's possible that given my blood panel and a lot of other things, the closest to me could be somebody who is 72 years old and exercise all their life. And you can use that data to inform your decision about how to cure that person. So that's the use cases we are working on and we have very advanced discussion. I have to tell you though, we started seven months ago. We're getting very close to the first contract. And everybody's so anxious. Everybody asks me, Did you get the first copy yet? And then there is the LLM guy, that they were like, instead they go, they spend billions. They were like, we have no intention even to engage into a discussion about revenues. But, whatever. Like We want to run it properly and, be profitable and make people happy, our investors.
Don Finley:I think that's the joy of capitalism in the United States is everybody can have a different view of what's necessary in order to go and the market's going to help figure it out. I'm of the mindset that you're never too early to get your first customer because you're going to learn a ton from understanding how people outside your organization actually, utilize your
Dario Villani:our first customer has been our hedge fund.
Don Finley:Yeah.
Dario Villani:And it was so helpful because Even just pinging the API, measuring the usage, and all of that, give us tremendous amount. And this company we are working with regarding circular tumor cells for classification of cancer. you understand that, we come from the world of finance where everybody has terabytes of data, and the first discussion we had with them, they were like, oh, we have. 62 patients and 200 features and we need to get to this level of specificity. It's a daunting, the first coming from any other field, vision or, it's so scarce the amount of data you have to do and you have to try to make sense of it. And the data is nasty, sometimes not synchronous. So to me, I always tell people it's like being an architect. If you never build a building, building makes you also a better architect.
Don Finley:Absolutely. So there's one thing that you just highlighted is sparse data sets, we all tend to run into, both sparse data sets, but also incomplete or you added 10 new features last month and half your data has at half of it doesn't, part of being,
Dario Villani:us,
Don Finley:yeah,
Dario Villani:for us it's very, also very, the other issue is the inhomogeneous data.
Don Finley:yeah.
Dario Villani:if you have mortgages, you have zip code, you have age, sex, male, female, you have income, so you have, variables that are, that Integers, some that are binary, some that are continuous. For us, it's very easy because every feature is an operator, the same dimensionality in the Hilbert space we work with. So we have no issue whatsoever to deal with that. And with sparsity we deal with it extremely well. really not an issue whatsoever. But generally, people, instead of dealing with the curse of dimensionality, what they do, there is a lot of domain knowledge. They say, all right, there is a hundred features, but these are really the three, four that are really good. And I said before, the ancestry of the whole is always bigger than the ancestry of the part in classical studies. So people have to really be sold in adding a new feature. Because it's going to also have the uncertainty, unless it's very relevant, being parsimonious. It's really important. But in a world in which there's gigantic proliferation of data set, if you can tackle all of them at once, without having to increase, the uncertainty, and having to do all that domain knowledge that in certain area you just don't have, It's a gigantic advantage. The way if a kid has played tennis and has played, ping pong, can learn paddleboard and be able to play a little bit in five minutes. It doesn't need to train for five million hours on five thousand GPUs. Yeah,
Don Finley:now, where, the ability to transfer, learn, or also to discover new skill sets, that can then be applied is something that our AI systems today don't do well. And you're saying that quantum cognition is a path forward on that, or at least a viable solution.
Dario Villani:it's definitely a viable solution. And it's, in my opinion, the only way you can get real abstraction in the real sense of the word.
Don Finley:Nice.
Dario Villani:And again, I want to give you that example so that I'm clear. When you say one plus one equals two, what are you training on? One orange plus one orange equals two oranges? But What if one orange is big and the other one is small? What if one is rotten and the other is not? One bottle plus what? What if one bottle is a liter and the other one is two liters? The crispest of the statement is that one refers to the most generic, where you integrate over all the instances possible what that one can refer to. And so it's the maximum uncertainty. But with zero entropy. So there is, my business partner is a genius, I'm just like more the flamboyant Italian. He always tells me, Think of it, if I tell you, imagine the Mona Lisa. You have a clear picture in your head. What if I ask you, What color is the pixel where is the neck? You have no idea. you have maximum uncertainty on each one detail, but the aggregate, it's very clear in your head. And that's what quantum is amazing at, that's what abstraction is, that's what insight is.
Don Finley:Ah, Daria. you've provided so much quality information for us and knowledge and I appreciate your wisdom because in order for us to truly advance upon our, current position, we have to have different thoughts than what got us here. And we are at a verge of, really possibly hitting the limitations of what is in the hype right now and will require like your intelligence and your team's foresight into, furthering where AI can get to. in the course of like this changing world that we're in, what advice would you give to other entrepreneurs out there around either the adaptation of this technology or, just in general from your life experience?
Dario Villani:generally I come from physics. It's, I think the hardest thing that, What you can do is to get into the path of somebody else. I tell my kids if Bruce Springsteen tried to do music the way Mozart did music, probably not have sold one, album. But he has his own voice, he wrote his own music that sings to the heart of everybody. It changed my life when I was 12 in Italy. instead of going to feed like LLM or in String Theory and adding after the people have published 2, 000 people, you try to do the fifth order correction to something. find your own voice. it's hard, but it's very rewarding instead of walking the shoes of somebody else because you are, you are pushing somebody else and It's very hard to give a contribution after the leaders have already passed, have built a moat, have built a lot of defense mechanisms, where you are there trying to, to instead of 500 trillion, you do trillion, instead of 82. 1 precision or accuracy, you get 22. 3. Nobody cares. So the difference between E2. 1 and E2. 3 is smaller than the ancestor you have on the whole thing, so nobody cares.
Don Finley:So basically, let's,
Dario Villani:voice and be brave.
Don Finley:yeah, Find your voice be brave. find your authenticity, be you
Dario Villani:you have a good idea, just pursue it. I was watching last night Steve Ballmer laughing. at the concept of the iPhone. Nobody's ever gonna buy an iPhone. laughing. And, the first time we talked about quantum cognition to some people were like, Oh, is it some sort of PC? Why wouldn't you do with LLM? That's a good thing. some skepticism makes you stronger. You have to answer to those questions, but then you have a real edge. You're doing somebody nobody else and time will tell if you're right or not.
Don Finley:Dario, thank you. Such great wisdom. Really appreciate you making the time to talk to us today.
Dario Villani:Of course. All the best. Thank you for having me. Very grateful that you gave me a chance to chat about our thing. All the best.
Don Finley:Thank you for tuning into The Human Code, sponsored by FINdustries, where we harness AI to elevate your business. By improving operational efficiency and accelerating growth, we turn opportunities into reality. Let FINdustries be your guide to AI mastery, making success inevitable. Explore how at FINdustries. co.
