Introduction

So my name is Tal. It's great to be here this evening.

I spend my time between Atlanta, Georgia and Tel Aviv, Israel. And I'm about actually to open something here in Paris as well.

About the speaker

And I'll share with you a little bit about, based on Stefano's request, a little bit about A, what we do, and B, what's the, from our perspective, where the AI is going to lead us.

So I call it distributed AI for whatever reason, but a little bit about myself. I spent most of my life in a place called Georgia Tech in Atlanta, Georgia, by which I was a student, and then I was a professor as well for some time, teaching computer science, mostly in data and AI.

but also entrepreneurship. While I was a faculty there, I did a couple of startups. I built an accelerator, and I worked with many of my students and faculty, built probably 20 startups and helped finance them as well as an angel investor.

Professional background

About almost 10 years ago, I found an operation that's called Drive, which is like a commercialization hub. It's a fancy name for a place that facilitates collaboration between corporations and startups. It's a global operation.

I have startups from the US, from Europe, and from Tel Aviv, from India lately as well. Our partners are... from Honda in Japan to Total here, John Deere United States, and many others.

Drive operation and goals

And the goals of the partnership is basically to find innovation, to commercialize innovation, which is a very difficult job for them and in general.

And my personal goal is to help them make impact. So hopefully, 100 startups by the end of this year probably help 100 startups to be successful.

means that they can actually scale.

We have many IPOs, unicorns.

But I focus on investment in early stage, so pre-seed and seed. I have two funds, $130 million, $147 million, that we invest first checks.

And we bring our partners to further invest in later stage. That's a little bit about us.

The evolution of AI

So AI.

One of the first questions that I ask entrepreneurs why they want to present ideas is why now? It's always relevant questions. Why now? Just because why? It's not to annoy, it's to basically to raise the concept.

And so why are we talking about AI now? Did you think about that?

So I, in the 90s, when Stefano was a baby, I was a frustrated young researcher and asking myself why this AI doesn't take off. Why we have a rule-based system, expert system, and it's good, it was working, but nobody was doing much with that.

Even as far as, let's say, 10 years ago, in Georgia Tech, we wanted to solve this natural language problem that many people want to solve at that time. And we had two schools of thought. One was that, hey, we need to understand the grammar, structure of a sentence, the role of every word type in order to predict what will be the next. And then we have another school of thought that I was not there, yeah, admittingly, that to say, hey, we're just going to have to count on the power of statistics, and let's just treat it as unknown objects, and just the power of statistics will give us the hint what's going to be the next character, which eventually won, okay?

So that's the first slide conceptually that I put in front of you.

Historical context

The invention of the chip, or the silicon chip, was like somewhere late 50s in the US, Bell Labs mostly. And you know this gentleman guy, the British guy, that started the state machine and whatnot?

And the data that provided us from the internet revolution, the alignment of those things created this big thing that we see now. This renaissance of the AI, or at least big wave. Let's see how long it's going to carry us.

But that's the thing that's in the works for 50, 60 years. So it's a big thing.

And we were frustrated for many years with machine learning and even deep learning. So we have the concept of neural network for many years now.

Recent advances

But all these things that happened to us lately with the transformers and the gaming brought us the GPUs.

We're going to talk a little bit about agentic AI, and then physical AI, actually, what's interesting to me.

So the last two boxes I'm working on, I didn't even put the AGI there, because I think it might be a little outreach. I mean, I think it's going to happen, but it's going to happen in stages fairly quickly.

Analogy and implications

So the way our brain works is funny, because we mostly kind of analogy basis.

There was a paper, I think it was somewhere in 95, And basically, when fMRI start to be more an issue, that start give us a possibility to look what's happening in the brain.

And the funny thing is that they find out that in the frontal cortex, somewhere in this area, where we have abstract thinking, when we do things that are more abstract, an analogy, that area is becoming more active.

So we are linearly thinking, because we wake up every day. We have a schedule. We think what's going to happen the last Monday is going to happen today. It's very tough for us to think out of this box.

And they try to figure out how to be creative and how to activate this analogy mechanism.

And in the paper, they had a case study by which they showed the structure of an atom. and try to explain very thoroughly. And then the analogy was how the solar system is working. And they saw with fMRI that this frontal cortex is very active.

And similarly, they further explored that. And they realized, whenever this analogy, this frontal coaxial work, so I'm going to try to give you an analogy to get us out of this box linear thinking. So I'm going to use the internet as an example.

So I know many of you Maybe most of you were not alive when the internet was invented early on, like the foundation of the internet, probably in the 70s, maybe earlier. But it's become more real when the browser came out just about in the early 90s.

Somebody heard about Mosaic. So Mosaic was the first browser, official browser. And the internet was such an early phase by which every day that new websites were coming up, they published a list of new websites in this mosaic.

So just to give you a perspective, it was like a handful per day. But the big winners in the beginning were the entities that put out the infrastructure, the pipeline, the TCP IP, the understanding of the protocols, That was Cisco.

It took about maybe 10, 15 years. By the way, before this Facebook or Google, there was other search engine like Yahoo or Alta Vista and other archaeological names that you can dig into the history of the internet that allows the browsing.

And then somebody invented the concept called Hotmail. And I remember that. Because that was a huge exit for that time. I think it was Microsoft that bought Hotmail for a few hundred million dollars. It's amazing. And they didn't understand why Microsoft is buying something that nobody knows how to quantify for hundreds of millions of dollars.

So basically, nobody made money from the internet. Not too many people make money from AI right now, yet.

And then, boom. third generation, distributed computing. So we had distributed databases. We had distributed computing as a concept from computer science way back, locking, unlocking, understanding how to leverage different resources.

But the implementation of that, so when I did my first startup as an entrepreneur, it was early 2000, I raised about a million dollars, like the first money I raised. Basically, most of the money went on buying those servers. I had boxes sitting near engineering to develop the software. We didn't have this concept, the Amazon Web Services, so that's user evolution.

So that's actually utilizing the network in a way that nobody thought before. Salesforce was one of the first guys to do that. And then, obviously, Amazon Web Services.

And that's led to another revolution, different business models. So this type of evolution was totally unpredictable. I mean, who could have guessed somewhere in the late 90s you're going to have

like concept like Airbnb or Uber, build new business model, build on top of whatever network is. Some people predicted this social network phenomena. Because they say, oh, the network is going to basically make people connect from side to side.

But to make the leap of what type of application is going to come out of that, let alone what the social ramification of that, no chance. Nobody had a clue about that.

So the purpose of this slide is to set up the next slide. Very similarly, the infrastructure play is the first one to benefit. Very dominant. The most successful company is Nvidia right now.

It's more successful than what Cisco was, but it's very, very much similar. And then obviously we have the network effect. And then instead of distributed computing, I call it distributed intelligence. What exactly does it mean?

A little bit easier to look at this as a continuation of the agent concept. But coming up with what's going to happen after that, very difficult. Very, very difficult. You can have guesses.

And we'll talk a little bit about more, maybe how can it look like or what should it look like. But you understand where we are. We are probably in this point in this revolution.

By the way, if you have any questions or something, you can bother me here.

Applications of AI

Let's see what distributed intelligence may mean.

What I didn't tell you is that most of my work now is I'm not doing research much, a little bit. What I do is help entrepreneurs. I help entrepreneurs build startups.

So what I like to do is I find really, really energetic, not necessarily young. Some of my entrepreneurs are older than 60. But energetic and that they want to have a big impact. And I help them by working with them.

Personally, I have an organization of about 30 people, 25, 30 people. Stefano met some of them across the universe.

And basically, I have a long list ideal solutions and requirements from the industry. So all my partners provide me with things that they would like to have and like to be. And I help match the talent of entrepreneurs with what the market wants on a very, very high level.

And then if there is a good match, basically I help them connect. If there's an exceptional match, I may invest in them.

Focus areas of investment

So from that, I focus on energy and mobility.

manufacturing, supply chain, logistics, all these sensors are something that I can help startups with.

So if you look at energetic mobility, you know that autonomous vehicles are already here. So autonomous exists.

So about a few weeks ago, I was in San Francisco, I drove a Waymo, and then I was in Atlanta. There's a company called May Mobility. Both of those rides without driver.

Robotaxi is the most advanced form of autonomous that we have right now. There are two main leaders, China and the US.

In the US, there are about 2,100 rides per week by Waymo. They have over a few hundred vehicles that can do that.

In China, they have five operators, over 2,000 autonomous type of vehicle, Robotaxi. So it's significant. It's happening. It's the beginning, but it's spreading.

I think Hyundai is building the first official, didn't declare it yet, first official Robotaxi manufacturing plant. So pure Robotaxi platforms.

Agentic mobility

So in the future, with agentic mobility, you will have a situation by which those vehicles are going to be able to maintain themselves, negotiate a path in smart city, pay tolls automatically. All this, it's land itself. You can understand how it may happen. It's not that difficult to imagine.

You can have dynamic pricing. You can have all the facilities that are going to make those fleets extremely efficient. and creative.

So for example, understand what type of loads they need to deliver and match the right platform with the right load. Can exchange platforms to optimize, so to speak, load per ride. So the optimization function can be infinite, honestly.

That's amazing. Progress is embedded there.

So what we call agentic commerce is directed to mobility. It's very promising.

Physical AI in Manufacturing

manufacturing is basically another progress that is about the physical AI. So you have actuators, robots, sensors that are going to basically be agentic, so in other words, agent, and they're going to execute jobs in a way that is very hard to imagine these days.

More than that, it will be relatively easy to deploy that and to execute on that in terms of compared to the complexity of the task that you want to create.

So the wisdom and the intelligence that's going to be embedded, by the way, in a small scale, edge devices, the efficiency of the model and the execution of the... One of the first investment I have is a company called Halo, which is on the edge computing. It's a very small device. If you buy any Raspberry Pi, The Raspberry Pi software, AI software that's involved is Halo software.

And it's doing really wonderful stuff. And as a mechanical engineer, you can design all sorts of great things with that. But this is a little bit siloed.

What I'm trying to say is if you connect this with the previous slide, you'll be able to basically generate on demand the right components for maintenance or the right platform as part of that. So the ability to understand how to optimize the equation on multiple silos is going to be there.

So the whole concept of what we have today of, hey, how to think about manufacturing or how to think about mobility is going to be almost oblivious to us and to the organization. So it's very hard to imagine that because you have these silos.

But we used to have a lot of silos that we don't have right now. So the whole operation of mobility, manufacturing, supply chain, for us as humans, is going to be like one stream of activities that most of this we're not aware of. It's mostly about what we want and what we need, basically.

Healthcare and daily life integration

So just to put into context, because we are the users, I'm going to talk about health care and this. You can read the slide, but I think it's mostly about preventative and personal.

So in other words, The AI and the sensors that we have and the sensor around us, it will be embedded in our day by day life.

So now we have, let's say that we get to a certain age. We need to take vaccination or do a certain routine test. All of this is surplus. It's going to just happen because it's going to be part of what we do.

Let's assume you sit in a restaurant and you look at the menu. Each one of us can look at a different menu based on what are optimally good for our health. So basically, the seamless preventative health is going to be dominating.

And obviously, the way we conduct medicine, execute on treatment, most of it is going to be embedded in our day-by-day life in a seamless fashion. I just put dating here to kind of excite the imagination because dating, you know, for my, I have daughters, and their concept of dating is totally different than my concept of dating, okay?

We used to go and meet some people. They don't do that, you know, necessarily blind date. What's that exactly, okay?

So this is, I put this on just to also raise some questions potential problematic area, because imagine the AI knows so much about us. In fact, there is a point by which the AI is gonna numb us, okay?

So in other words, because everything is gonna be so convenient and so empathy, and the AI will know when we feel uncomfortable and will make us avoid this feeling uncomfortable. So we basically not gonna stretch ourself. There'll be less pain, but we more empathize, okay?

I don't know if I'm giving you this vision, clearly, because let's assume that I have all these devices on me, and there will be vectors that understand how cars should drive, what noise level we have, what temperature we'll have, and everything will be fit to what we need. We'll just be too comfortable sometimes.

Same thing with this dating thing. If the AI will know exactly who we are and then magically will match us with somebody that's so comfortable for us, maybe they'll be too comfortable. So just food for thought about that.

Which leads me to some other issues.

Global transformations

So when I was thinking about another analogy, is that China, in the last 50 years, and the Chinese people did amazing, okay? They came up with amazing fist of rural, local thinking to being basically dominating industries. I mean, talking about mobility, they are literally

You know, they become unbelievably successful. I spent many visits in Beijing, in China, as part of my work and research, and working with Chinese entrepreneurs. Extremely impressive, the rate of growth.

The reason I give this as analogy is because, in my humble opinion, the part of the vision is going to happen with AI, and the empowerment of individuals with those capabilities But it's going to be like what happened in China, but it's going to be quicker and more intense, because you're going to have a central ability to have millions of agents.

And basically, if you decide on a goal, any one of you can really kind of mobilize a lot of capability. The results of this is going to be depends on your, basically, aspiration. You can create a lot of competitive advantage to whatever goal you want to achieve.

And you ask yourself, OK, we are democratizing intelligence, basically. That's what happens. You don't have to be absolutely invest in human capital, because you're going to have that at your disposal.

So what happened in China can happen in China or other places, like, let's say, small island. And they want to finally have 10 million new workers, the best they can have.

So what's actually the limiting factor for you? Potentially, you can think about money, maybe, or resources. But not so much, because this can go down in price.

Maybe energy that you can activate. So you see that normal structure is going to be mostly decentralized. And we will not need many of the social structure that we have today, in theory.

That's another foot for thought.

Energy and ecological considerations

And that's the last slide I have in this section, which is we see it already now. We see it in data centers. We see the problems with the power. And we see the problem with the cooling.

We see the problem with the ecosystem, decarbonization. It's all going to mix up with AI in a way that you won't believe.

Obviously, we have a lot of food for thought, so it's a bit philosophical. To what extent do we want to live in potential life like that? It's a good question.

I'm talking about on a personal base. What are the risks?

I mean, you see what happens to the society as a result of the echo chambers and Isolationism.

We had a dream about a certain world 30 years ago, 20 years ago, and now things are flipping out on us. Is it good? Is it bad? It depends on where you're coming from, what you think.

But there's definitely a lot of thoughts and risks that we should take into account that is probably going to land upon us. And I think the bottom-up is what is the definition of self?

To what extent this AI is gonna change each one of us? That's another question I ask myself as well.

Just to give you a perspective about energy, this is how much money US spend on the Manhattan Project, the Apollo Project, and this is how much money in the world is gonna spend on global energy transition. That's a huge mega effort, okay?

So, and that's part of what I'm saying. This is part of the chain of the AI revolution.

And when we look at, so when I look as impact seeker, I always look at those, okay? I look for electrification solution, hydrogen, SMRs, solar wind energy, my preferred one, and compact fusion, even more desirable.

So as an AI futuristic investor, and I'm looking at you as a potential contributors and entrepreneurs, I look, I ask myself, what is the top two or three area by which I'm looking for now?

Future directions in AI

So going back to the, to our level.

So I'm looking for number one, security, reliability, trust, okay?

There's a huge danger in developing the stronger and stronger AI without having it somehow contained and under control, managed, okay? And on all the fronts, okay? It's such a complex, you know, when you, the inherited in AI, you really don't understand exactly what's happening inside mechanism.

You want to have this ability to understand that in order to contextually put the stamp of approval, in order to understand how you manage the boundaries of what you do. That's something that is very difficult task. And when you're going up the level of abstraction all the way to physical AI and AGI, I definitely want to be able to do that.

Security and reliability

a look how many actions like a robot do okay or or agent do okay so who's going to guarantee that an action is something that was intended or at least on the realm of the dictionary of intention so how to combine concepts from blockchain to authentication is something that I'm working really hard on that's really these days another thing is what I call cyborgish okay so understanding the domain and this is not philosophical, this is actually real, between human and machine, okay?

So it's really, I look for that type of technologies, okay? So I'll give you one example, what do I mean by that? There's research coming from McGill University, which is in Canada,

and it shows how a cognitive load on our brain influences our motion. All this connection between social, neural, sociological, and science, that's the vector you need to look for in this type of domain. And it turns out that we are very much

Subjected to, and we don't know that, our body moves differently with other different cognitive loads. So if I ask you different questions, you don't even notice, but you move differently from your facial expressions to your limbs to your body.

Human-machine interactions

it's it's very micro movement so the question how you capture those movement and to what extent you can map it into what is your the reasoning of your cognitive cognitive load i'll give you one example okay so let's say you are drinking okay so you're a bit intoxicated even a very minute amount there's already you already express different motions okay so if you can measure that it can be helpful for you to control it or do something with that okay so Or if, let's say, you have motion sickness, then you exhibit different small motoric behavior. So this goes to what I talked before.

So if you are in a car or in a boat or in an airplane, if there's a mechanism that almost predict or giving this delta of understanding that, hey, you're about to feel sick before you actually feel sick, you can improve that, basically.

Swarm intelligence

And last slide, which is a major slide on a conceptual level, is that the thinking in the future about developing AI will be about swarm. And that is going to be a key word, if not already, in AI development.

Why? Because...

Distributed intelligence is gonna be how to optimize this swarm for some goals, okay?

So when we're gonna develop, for example, when we develop a car today, it's about the car. But when nature developed the bee, it was about swarm of bees.

It has small wings. It has hormones to communicate with other bees.

It has uber goals for other besides flying to one flower. So swarm is an important concept.

And I'm looking to help develop this concept as well.

Conclusion

That's about it. That's about 20 minutes, 25 minutes.