 Dave, you have your, is your presence here? Yeah, I think it's here. Well, this is a perfect way to follow up on the previous speaker, because I want to say the same thing. I'm thinking from the software point of view, I'm thinking from the digital point of view I could get a graph about real biology. And it seems to me obvious that if we're going to have true evolution, open-ended evolution, we're going to have major evolutionary transitions of Sapphorn and John Maynard Smith, or else whatever we have is not open-ended evolution. And so I will recapitulate the argument that was just made, that if we observe open-ended evolution over time, the individuals are going to get bigger and slower that we are considering. That's what it means. And if we are in a finite system eventually, we will reach a population size of one that lives forever, and evolution will stop by definition. So if we actually want philosophical, theoretical, true open-ended evolution, it has to be an intimate system in the end. It has to be an indefinitely large system in the end. Now that may be completely irrelevant, because all of the complexity that you could ever want and all the things that could keep you watching the simulation until you drop could happen in a completely finite system. It's all a matter of time, it's head, short. Space time bottom. So the first point is there's a lot of stuff we do where we take for granted what the creature is in the Packard and the Delac, which we take for granted what the component is that we are attempting to measure evolutionary activity of. And if we do that, we are dead. If we are stuck with the component that we have, and we cannot say now an exponential number of those guys, or even a linear number of those guys out here, come together to form a transformer that marches down the road, whatever it is, then we're stuck at the level that we started at. And we are going to observe at best climbing some kind of hill, perhaps very slowly, perhaps moving through some high-piped space bypasses and ravines before we saw it. And maybe that, in fact, is as much open-ended evolution as we could hope to observe until we redefine what the components are. I must go right this way. So what I would like to propose is if we're going to have a satisfying, truly notion of open-ended evolution, and I'm not sure it's even available. I'm not sure it's just moving in advance in things that we induce from looking at very limited data. But if we want, we should be looking at it in a way that the components, the things that we think are evolving, are observables in the data and not priors that we define and put upon the system. Okay, that's step one. So if we cannot say, okay, consider a creature K, consider a population N of creatures K, if we can't do that, then what do we do? You have to hard-code something. Well, what we do is we start with an artificial chemistry or an artificial physics. We start one level down. And we say, yeah, we're willing to keep this, this level, absolutely fixed. We're willing to postulate this and say everything has to happen above that. But it has to happen above that and above that and above that. So we're going to find our components, we're going to find our creatures where we present things inside some kind of artificial physics chemistry. A kind of artificial physics or chemistry should it be one of the great joys and sins of artificial life is that we get to make up the laws of physics anew for every damn paper. And that's why people from the outside, by the way, think it's all crap. But sure, you can make it happen. You can make it happen. You can make it happen. You can't make it happen. You can't make it happen. You can't make it happen. You can't make it happen. You can't make it happen. It does not happen. Of course, we know that life had to be that in the limit because what did happen is not very satisfying. So the proposal I made last year at A-Life, the proposal that has become my purpose in life, so it's part of every talk I've given, is that satisfying models should be indefinitely scalable. It's not, it's unacceptable to say, I have a 100 by 100 array. You can say I have a rule, an update rule, that applies to anything. I'm going to do experiments on 100 by 100 and also 200 by 200 array or whatever it is. But nothing changes about the framing of the model that couldn't be scaled up to arbitrarily large. And we all nod our heads and say, oh yeah, my model came back. But it turns out there are things that are not quite so obvious that don't scale. And the classic one for A-Life is a simplest clock. The classic one for A-Life is imagining that deterministic, synchronous models. They definitely scale to be large. They cannot. So Conway's Game of Life. All traditional, serious, simple, how the window can't happen. This argument has some bite. This argument says certain models are admissible certain needs are not. So there's an increasing pile of papers about indefinite scalability. And basically, if you're willing to adopt a probabilistic cellular atomic where the probabilistic is reasonably real so that you can't just assume away, I'll just do triple modular redundancy and assume away the probability, then you're probably okay. Because other automatism is basically local. So they don't have a global clock. They don't have any global synchronization among us. So they're mostly good. But you have to give up on determinism. You have to give up on 100% reliability. If you're ready to do that, you're ready to build models that in some, at least some way, simple way, could be transformable into other non-deterministic and definite-scale models. There is a hope that we could start having, actually, instead of a brand-new laws of physics in every paper, we can say, I am using an indefinite-scale model. I am using an indefinite-scale model. And in fact, we could think of a polynomial transform between his indefinite-scale model and his indefinite-scale model and actually start to build up some progress. That's it. All right. How do I make this thing go? That's hard. Let me try. Here's my fantasy. I've talked to a couple of people about this. I got here because I wasn't sure whether this has already been done or proved impossible. And I don't know that either is true. But it seems to me this is what we ought to be able to do. Suppose you said, here is a digital cube of billion by a billion by a billion. One of the x and y is going to be a two-dimensional world, and y is going to be time. So we have this cube of bits. And we can take that as our model and we can impose it upon reality any way we like. Finite volume of space-time granularity. So we could make individual bits go all the way down to quarks or take those bosons or whatever we want. Or we could make those individual bits represent cells or people or planets. But we have this finite realm of space and time. And the question is, is evolution happening inside that? And if so, is evolution open-ended? Evolution happening inside that? And then we could say, we'll move our box around. We'll zoom in. We'll zoom out. We'll put it over here. We'll put it over there. And we'll lock. And the way we do it is, okay, put your box someplace. Put it at the size of cells. Make some glob model of everything below that. And just live with it. Great. It's your choice. Great. Look at the first slice of tunnel. Do a spatial auto-correlation shift that thing against itself and look for repeating patterns. Look for nearly perfect repeating patterns. If you look for really perfect repeating patterns, you'll find crystals. Spatially absolute repeating patterns. You want nearly repeating. They're really different but similar. And those are the ones that are candidates for components. Those are the ones that are candidates for creatures. And then you take those, and you look over successive slices of time, and you connect them together based on adjacency to make a computer-to-life box. This guy at this slice is probably this guy at this slice. Probably this guy at this slice. Now there's two of them that are probably related. You take the lifelines and you project them across time to get a fitting space. And now you have a line moving around corresponding to each lifeline. And you see how far it gets in the phase space compared to a random walk in that same phase space. And that's the neutral model, the shadow. And if it moves, you have evolution. If you can take then that model and zoom out to a larger scale and see those things happening again, then you have open-ended evolution. It's all on us, but it should all be post-hoc. And that, I propose to you, is if we haven't done it already and I wish someone would say we had a research proposal for the grand challenge open-ended.