What was it that pushed you towards artificial intelligence?
I’d like to tell you my story, how I got close to it. It comes out of a conversation I had with a friend - hi Valerio - where we were talking about our first moment with artificial intelligence.
For me it was more than 20 years ago. I remember I had to choose a university, and I’d found this degree path in psychology and neuroscience, cognitive psychology, where there were neural networks. I didn’t know what a neural network was, but for psychological hang-ups I’d rather gloss over, I’ve always had a tendency to want to give schemas to behaviour - mine and other people’s. Maybe out of a need to understand, to control. Out of a form of fear.
The fact is that as a kid I started looking at the first diagrams of artificial neurons on the internet, back in the days of the 56k modem. I looked at those diagrams and I was hooked, because digging a little deeper there were equations underneath - schemas par excellence. And yet there was no schema tied to the symbol: the dog, the cat, the house. There was instead a schema of the internal workings of a thing that each of us has in our little head.
Sub-symbolic
That suggested what I later discovered is called the sub-symbolic approach: the fact that these neurons encode things, and encode them in a distributed way.
If I say “dog”, you understand what a dog is - but for you a dog might be something completely different from what it is for me. For you it’s maybe a poodle, for me a German shepherd, for my friend a giant black beast. All these circuits encode, build and knead together an idea of the world that we construct over the course of a lifetime. We start with innate things, and then we add, and add, and add.
Actually, more than adding, we select. We lose neural connections over the course of life. That’s another discussion - learning happens by selection, not by instruction. To learn means to remove, not to add, from many points of view.
That stuff struck me like lightning, because they were schemas, but schemas on a plane that wasn’t necessarily the plane of language. They were schemas on the plane of survival, of selection, of organisms elbowing each other for space. So it was immersed in biology, immersed in timescales far longer than a single life, a single individual.
And these groups of neurons, put together, have fantastic properties. What you could call memory, seen locally, was an interaction of these nerve cells that managed to reconstruct and recognize patterns. My head opened up. From that moment on I never let go of these things.
I remember the professor who introduced me to neural networks - hello, Alessandro Londei, in my opinion one of the greatest minds I’ve ever met. As a physicist he opened me up to this world of neural networks when I started attending university. For me it was love at first sight. Deep, real love.
What’s really underneath
I try to remember these things often when I look at today’s world of artificial intelligence, which is loud, made of money, made of infrastructure, made of politics. It has become something else entirely. So I wanted to dedicate today’s video to remembering what is really underneath.
I spent years pitching this machine learning and deep learning and neural network stuff to small-time bullshitting entrepreneurs and hearing back: “Ah, but what’s the use case?” The use case you can see now, when it’s too late, buddy. These language models, these giant things, the role of scale. And it’s all sub-symbolic. There is no symbol.
In all of this, the great discovery of my life - the one for which I think it was worth being born and having lived even just like this, so that if I ended tomorrow I’d still be happy - is that I understood how the mind works, more or less.
And first of all I’ve almost abandoned the very concept of “mind” in favour of “brain”. The concept of cogito ergo sum, of feeling, of consciousness, that too is neural. I’ve brought it all onto a very, very material plane. Materialist, even. And it’s no less interesting for that - which is another thing again.
Against the dualists
I’m going a bit off the cuff here; every so often I do a video like this with nothing written down.
One thing I don’t understand about the dualists, the ones who keep insisting on the human primacy of intelligence - or rather, on the human primate of intelligence - is this: admitting that we are this, that we are matter, we are brain, we are selection, we are organisms (assembly theory is fantastic), takes nothing away.
If you look at how a simple sodium-potassium pump inside the membrane of a neuron works, you see stuff that makes you say: how the hell is it possible that this exists? Across all these scales of magnitude, over tens of thousands and millions of years, this stuff was conserved, it made its way through. And there is nothing symbolic about it.
What we slap on as a symbol, thinking we have this crystalline idea of the world, has nothing to do with what actually is. And what is it? Activations of groups of neurons that encode a continuously updated model of reality and of the interaction with the environment, so you can keep living, propagating, and elbowing your way into existence.
Swallowing that took years. But once you’ve swallowed it, you appreciate it. There is an infinite beauty in it.
What else to add for today? I’ll stop here. If you want to know more in detail about these themes, let me know - maybe we’ll do a bit of philosophy. Because it’s exactly in this case, exactly when you go deep, that vibe godo.
