AGI

Every star in the sky probably has planets and life is probably emerging on these planets. But I think the commentorial space associated with these planets is so different. Our causal cones are never going to overlap or not easily. And this is the thing that makes me sad about alien life, why we have to create alien life in the lab as quickly as possible because I don't know if we are going to be able to build architectures that will intersect with alien intelligence architectures. Intersect, you don't mean in time or space-

So your big assembly theory paper was published in Nature. Congratulations. Thanks.

All right, before we get too far, let's talk about the assembly equation. Okay. How should we do this? Let me just even read that part of the paper. We define assembly as the total amount of selection necessary to produce an ensemble of observed objects quantified using equation one. The equation basically has A on one side, which is the assembly of the ensemble, and then a sum from one to N, where N is the total number of unique objects. And then there is a few variables in there that include the assembly index, the copy number which we'll talk about. That's an interesting, I don't remember you talking about that. That's an interesting addition and I think a powerful one. It has to do with what, that you can create pretty complex objects randomly, and in order to know that they're not random, that there's a factory involved, you need to see a bunch of them. That's the intuition there. It's an interesting intuition and then some normalization. What else is and-

Okay. What about, if I show up to a new planet, we'll go to Mars or some other planet from a different solar system, how do we use assembly index there to discover alien life? Very simply, actually. Let's say we'll go to Mars with a mass spectrometer, with a sufficiently high resolution, so what you have to be able to do, so a good thing about mass spec is that you can select the molecule from the mass, and then if it's high enough resolution, you can be more and more sure that you're just seeing identical copies. You can count them. And then you fragment them and you count the number of fragments, and look at the molecular weight. And the higher the molecular weight and the higher the number of the fragments, the higher the assembly index.

I didn't know which one was higher. We didn't really do any detail there. Because now we are doing that. Because one of the things we've done, it's a secret, but I can tell you. I think it's- Nobody's listening.

Signatures of chemistry are fascinating. So you've been saying a lot of chemistry examples for assembly theory. What if we zoom out and look at a bigger scale of an object, like really complex objects, like humans or living organisms that are made up of millions or billions of other organisms, how do you try to apply assembly theory to that? At the moment, we should be able to do this to morphology in cells. So we're looking at cell surfaces, and really, I'm to trying to extend further. It's just that we work so hard to get this paper out and people to start discussing the ideas, but it's kind of funny, because I think the penny is falling on this. So yeah-

Okay. Another piece of criticism or by way of question is how is assembly theory or maybe assembly index different from Kolmogorov complexity? So for people who don't know, a Kolmogorov complexity of an object is the length of a shortest computer program that produces the object as output. Yeah, there seems to be a disconnect between the computational approach. So Kolmogorov measure requires a Turing machine, requires a computer, and that's one thing. And the other thing is assembly theory is supposed to trace the process by which life evolution emerged, right? There's a main thing there. There are lots of other layers.

Maybe just a little insider interesting information. What were the editors of Nature, what the reviews and so on, how difficult was that process because this is a pretty big paper. So when we originally sent the paper, we sent the paper and the editor said that... This is quite a long process. We sent the paper and the editor gave us some feedback and said, "I don't think it's that interesting." Or "It's hard. It's hard concept." And the editor gave us some feedback and Sarah and I took a year to rewrite the paper.

I'm sure I make mistakes on that.I argue lots with Sara and she's shocked. I've argued with Joscha, Joscha Bach, in the past and he is like, "You're just making that up." And I'm like, "No, not quite. But kind of." But I had a big argument with Sara about time and she's like, "No, time doesn't exist." I'm like, "No, no, time does exist." And as she realized that her conception of assembly theory and my conception of assembly theory was the same thing, necessitated us to abandon the fact that time is eternal, to actually really fundamentally question how the universe produces combinatorial novelty. So time is fundamental for assembly theory? I'm just trying to figure out where you and Sara converged.

Now, I guess this is just purely a guess. I have no data other than hope. Maybe not hope, maybe... No, I have some data. That every star in the sky probably has planets and life is probably emerging on these planets. But the amount of contingency that is associated with life, is I think the combinatorial space associated with these planets is so different. Our causal cones are never going to overlap or not easily. And this is the thing that makes me sad about alien life. It's why we have to create alien life in the lab as quickly as possible because I don't know if we are going to be able to be able to build architectures that will intersect with alien intelligence architectures. Intersect, you don't mean in time or space?

Well, okay. So, given that you're disrespecting the power of the initial conditions, let me ask you about, how do you explain that cellular automata are able to produce such incredible complexity given just basic rules and basic initial conditions? I think that this falls into the Brouwer-Hilbert trap. So, how do you get cellular automata to produce complexity? You have a computer, you generate a display, and you map the change of that in time. There are some CAs that repeat like functions.

When it comes to nuclear weapons, I grew up in the '70s and '80s where there was nuclear doom and a lot of adults really had existential threat, almost as bad as now with AI doom. They were really worried. There were some great ... Well, not great. There were some horrific documentaries. I think there was one called Threads that was generated in the UK, which, it was terrible. It was so scary. And I think that the correct thing to do is obviously get rid of nuclear weapons, but let's think about unintended consequences. We've got rid of ... This is going to be such a non sequitur. We got rid of all the sulfur particles in the atmosphere, right? All the soot. And what's happened in the last couple of years is global warming has accelerated because we've cleaned up the atmosphere too much. So ...

You've mentioned that, what'd you call it ... Chem Machina? Yeah.

Yeah. You mentioned GPT for electron density. So a GPT like system for generating molecules that can bind to host automatically. I mean that's interesting. I's really interesting. Applying this same kind of transform mechanism. I mean, my team, I try and do things that are non obvious but non obvious in certain areas. And one of the things I was always asking about in chemistry, people like to represent molecules as graphs and it's quite difficult. It's really hard if you're doing AI and chemistry, you really want to basically have good representations. You can generate new molecules are interesting. And I was thinking, well molecules aren't really graphs and they're not continuously differentiable. Could I do something that was continuously differentiable? I was like, well, molecules are actually made up of electron density. So I got thinking and say, well, okay, could there be a way where we could just basically take a database of readily solved electron densities for millions of molecules? So we took the electron density for millions of molecules and just train the model to learn what electron density is.

This is going to be a fun continued conversation on Twitter that I look forward to. Since you've had also from another place some debates that were inspired by the assembly theory paper, let me ask you about God. Is there any room for notions of God in assembly theory? Of God. Yeah. I don't know what God is a... I mean, so God exists in our mind created by selection. So the human beings have created the concept of God in the same way that human beings have created the concept of super intelligence.