I love stuff like this, in the past I spent a good amount of time wondering about what were the key milestones in the development of intelligent life. Multicellularity is a good example. From my (admittedly outsider) perspective, a lot of that seems lost on the popular approaches to AI, which seem to focus too heavily on the late-stage result (the brain) and not everything that led up to it. Evolutionary algorithms have been out of vogue for some time, Artifical Life is an under-explored backwater.
One thing that I am surprised has been left out of this analysis is extinction events. In the intro it seems to dismiss “models that appeal to rare chance events” (paraphrasing). But the Cretaceous-Paleogene extinction in particular was a chance event that had an absolutely massive...impact...on the evolution of intelligence. Dinosaurs dominated for a couple hundred million years with no signs of moving toward primate-style intelligence, and then suddenly a whole new set of conditions created an environment where scavenger mammals developed rapidly.
I'd say that the big milestones would be photosynthesis, eukaryotic life, and then multi-cellularity. I trust the first and last are obvious but I think people don't appreciate mitochondria enough, for all the memes about "the powerhouse of the cell". Bacteria respirate over their membranes which, through the square-cube law, limits how big they can be. Our cells have many mitochondria scattered though them and can get pretty big. And with size comes the possibility of much larger genomes. Plus in life on Earth at least mitochondria are responsible for apoptosis and its hard to image multi-cellar life without that.
If you replace "intelligence" with "flying", it becomes more apparent why there is generally little interest in artificial life as an approach to develop artificial intelligence (although there are small conferences on this approach).
In general, to recreate something from nature, so far we have not needed to recreate nature as a whole. It is not clear why we need to treat intelligence any differently.
For one, flying is a relatively simple physical phenomenon, and once we had a firm grasp on the principles of lift, we were able to design our own solutions.
Intelligence is, in comparison, an incredibly complex phenomenon that seems a lot harder to separate from the inner workings of the biology and evolutionary history. It feels to me like there is an implicit hubris embodied in the industry and reflected in your comment. Obviously there has been an increasing overlap between neuroscience and AI, and the approaches do strive to understand and mimic the biological structure of the brain. What is not clear to me is how we have largely collectively decided that we don’t need to incorporate understanding of how that structure came to be.
And I’m not saying that we need to replicate the whole process exactly as nature did it. That would take billions of years. But perhaps if we understood it better we could take the key insights and mechanisms to design our own system.
Intelligence is more than just processing information. A big part of intelligence is agency, it’s having a reason for doing things because you live in a world where you have intrinsic goals and desires and feelings and needs. I just don’t see AlphaGo or Watson or anything like that ever reaching that level. They’re Frankenstein vat processors built to answer questions within predetermined contexts we bake into them. They’ll never question why they do what they do or simply decide to do something else. I’m convinced we need to essentially create a simulation where something with agency can develop on its own in order to reach truly interesting AGI.
> I’m convinced we need to essentially create a simulation where something with agency can develop on its own in order to reach truly interesting AGI.
That is fair, and some people do work in that direction. However, the step from "can develop on its own" to "will develop on its own" is highly non-trivial. And seeing the results from the really smart people that work in this direction, I am very much not inclined to work in that direction. There simply does not seem to be a good way to make progress.
> Intelligence is, in comparison, an incredibly complex phenomenon that seems a lot harder to separate from the inner workings of the biology and evolutionary history.
You might claim that now, as we don't understand intelligence yet. However, in the same way it was claimed in the 19th century that people would never be able to fly. What I meant to say is that things that seem hard in nature, might actually be a lot simpler in technology.
I actually don't believe that intelligence is an incredibly complex phenomenon. On the contrary, I reckon we are really close to cracking it. I put it on the same level as moon-bases and nuclear fusion, all reasonably likely to happen in the next 20 years.
However, the step from "can develop on its own" to "will develop on its own" is highly non-trivial.
Yeah, that's the rub. When starting from a more primordial level, the search space is widened to truly astronomical proportions. It may run for ages before you can even tell if progress is being made. I understand that this is not attractive to many researchers/investors. I've often wondered how one would go about crafting a viable fitness function and admittedly I've not come up with much.
What I meant to say is that things that seem hard in nature, might actually be a lot simpler in technology.
Fair enough. I don't believe that intelligence has to develop the way nature did it on earth. It is plausible that there is a way to short-circuit it like we did with flight. But it's not guaranteed that we're on the right track to finding it either. I would just love to see more research along those more experimental evolutionary lines.
One thing that I am surprised has been left out of this analysis is extinction events. In the intro it seems to dismiss “models that appeal to rare chance events” (paraphrasing). But the Cretaceous-Paleogene extinction in particular was a chance event that had an absolutely massive...impact...on the evolution of intelligence. Dinosaurs dominated for a couple hundred million years with no signs of moving toward primate-style intelligence, and then suddenly a whole new set of conditions created an environment where scavenger mammals developed rapidly.