It is my great pleasure to welcome to the stage Jack Hidary, CEO of SandboxAQ. Jack, welcome to the official USA House AI Summit.
Good to see you, Olivier.
So, Jack, I think you and I agree on something: language is great, but language is not everything when it comes to human-to-human interaction, but human-machine, human-computer interaction, right?
Well, you're right, Olivier. When you look at our world, certainly language is a big part of it, and it's the one that's easiest to train a model on. Why is that? Because there's a huge data set called the World Wide Web. But our world is much richer than that. We know, for example, that the world of biology — we can see all kinds of things in the human body, in biology, in nature — and while we could describe it in words, and that's on the web, the actual equations, the actual physics of it, is not really encaptured by just those words. And so, we must go beyond language models if we want to really understand and create true AI that can help us in our work.
Guys, look at Jack. It's more than words. It's about everything that we call embodied cognition, the fact that we need to understand how our brain works. Today, we record brain waves, facial expression, eye movement, voice intonation, physiology in order to train models, but this requires a lot of sensors, of course, acquisition of the signals, and a lot of AI models that are multimodal to make sense of this data, but also a lot of compute, and there is latency, right?
Olivier, let's also go back to the basics. Where did the transformer model come from? Where did neural networks come from? From the field that you and I come from, which is neuroscience.
And so, the brain itself is a distributed set of nodes. We have billions of neurons in our brain. Any guesses how many neurons in an average human brain? I'll take guesses from the audience. I'll start the bidding at 1 billion. 1 billion neurons in the brain. A trillion? Okay, we're going to get to a trillion in a second. It's not a trillion, but close on another topic. A billion, 2 billion, do I hear 3 billion? 2 billion? Okay. All right. There's 86 to 100 billion neurons in the average brain.
All of which are connected.
Olivier is above average, so probably more. Yes. But now, the trillion is the connections, right? So, 100 trillion to 1,000 trillion connections. Now, when we open up the brain — Olivier studied this, I studied this — you open up the brain, what do you see in there? Do you see language? Yes. We have Broca's area right here to produce language. We have Wernicke's area right here to understand language. Those are only two of the many parts of the brain.
Well, can we just pause here? Neither Jack nor I are opening brains. You don't want either of us to be taking care of neurosurgery. We can collect brain data with non-invasive methods, sensors.
That's very important because this is really different from what some are doing with implants. Non-invasive is what is scaling, and this is where we need not only the compute that we all know — GPUs, TPUs — but this is where quantum is going to change the game, right?
Absolutely. Yeah. So, again, you look at the brain, it's got so much going on: processing of data, processing of nature, light, and things like that. Now, where does quantum come in? Why do we need quantum?
Well, our world is quantum. Well, those of us, you know...
Wait, can you explain at least what you mean by 'the world is quantum'? That's a big statement, right? Okay.
So, when we look at our world and we walk up and down these steps, we play billiards, and we hit the white ball, and the white ball hits the red ball, this is governed by Newtonian mechanics. Isaac Newton, super smart guy, told us about F equals MA and how that works. But actually, the fundamental layer of our universe is not that. When you have an atom, let's say you say, 'I want to have a great drug for cancer,' and you break down that molecule to atoms, and the atoms — you look at the outside electrons to see how they can interact with the body — that is not a Newtonian billiard ball situation. That, Olivier, is governed by a different set of laws, the quantum laws. So, if we're going to help humanity, if we're going to do these things like create new medicine, create new materials, better batteries to store energy for renewable and other forms of energy, then we need the quantum space. Now, the quantum computer is only one of the many quantum ways that we can enhance our understanding of the quantum world. We also have quantum sensors. Back to your point about sensing the body, sensing the world around us. There are magnetic fields all around us, and these sensors, being quantum in nature, are much more sensitive than a traditional thermostat or things like that.
Okay. So, how better? What's the competitive advantage of quantum?
Yeah, the sensors. With the sensors, you can detect magnetic fields that are much weaker. So, for example, let's say flying into Dubai or Abu Dhabi right now — if you fly there today, you're likely, if you knock on the pilot's door and they answer, what they'll tell you is that their GPS is being jammed. There's no GPS that they can really lock onto. If you go to many parts of the world right now, there's really no GPS to lock onto. And recently, about six months ago, the plane of Ursula von der Leyen, the president of the EU, was targeted with a GPS knockout by a state-sponsored actor. And so, GPS is very fragile. And if we want to navigate without GPS, one of the ways we can do it, inspired by nature, is using the magnetic field of our Earth. But the magnetic field of our Earth really is so weak that we need something like a quantum sensor to detect that field. And then we need a new kind of AI — not language models, of course, but a quantitative model that can read that magnetic field and help us navigate. And that's happening right now in planes around the world. It's not commercialized yet, but it's in development.
That was my next question, Jack, because this is beautiful. You just gave us the menu, but how do we cook it? How far are we from this reality being scaled?
I love that. The French always come back to cuisine. It's good.
Well, amongst other things, but let's stick to cuisine.
The scaling, I think, is going to be quite interesting. Right now, in the AI field, there's an obsession with scaling data centers. When it comes to scaling these other technologies, it's a different kind of ballgame. When you want to make a quantum computer, for example, there are seven different ways of making a quantum computer. Satya was just here. Microsoft is building one or two of these kinds of computers. Google's doing that. IBM's doing it. So many different companies are trying to scale. But scaling a quantum computer is difficult. And one of the reasons why it's so difficult is to make sure we can correct the errors that abound in a quantum computer. And that's one of the big goals of the next four or five years.
