Far too expansive of a topic to cover in one notebook, but I’m going to try and start keeping notes on this! This notebook is going to be specifically on different lenses to view evolution.
List of related notebooks:
as a minimal criterion
Although evolution is often seen as a force of convergence, you might also see it as a diffusion operator. Rather than humans being the “peak” of evolution, we’re just one niche in a world with millions(?) of niches. You don’t need to be the most fit to reproduce, simply fit enough. This gives a lot of leeway for what is able to reproduce.
Stephen Gould’s Full House covers this and on the computational side, Kenneth Stanley has done a lot of work in this area too.
as a novelty creator
Another way to view evolution is through creating new niches and novel creatures. How exactly such a simple process can seem to infinitely create new things is still an open problem. My hunch is that it isn’t that evolution is so deep that makes it amazing, but that the environment it is in is so complex. Any simple enough algorithm in a complex enough substrate can produce wonders. (See AI for more on the Bitter Lesson)! This is also why I see so much promise with neural networks and reinforcement learning. If you can create something that can always scale with more data/learning, then the primary challenge is to give it a more complex environment!
One idea here is that coevolution is a key thing missing from our current DL models. They learn on “static” distributions, but this limits the complexity. You can always grow it by collecting more data, but like evolution, we’d like something that can naturally complexify. Something to study is why GANs fail and how we might make them work better?
as an optimization tool
Even though the minimal criterion suggests that evolution creates different niches that can coexist, this idea of “survival of the fittest” has been adapted to create tools that can search for optimal solutions given a problem. Generally they go under the name of genetic algorithms or evolutionary strategies and vary on how biologically inspired they are.
as a constrainer
Humans have a mutation rate of about 10^{-8} but since we have ~ 3*10^9 base pairs, that’s also a lot of mutations (roughly 3,000 per person). How is it that with all those base pair changes, nothing terrible seems to happen to most of us?
I think a key thing here is that with a diverse enough environment, evolution is incentivized to create evolvable agents: agents that can change quickly. It’s kind of amazing that Hox gene mutations can create wacky but perfectly viable body plans and doesn’t just kill the development of an organism. In that case, evolution tightly constrains what is able to evolve while at the same time creating a more tractable search space.
General notes
From the MODES toolbox, this idea of a “persistence filter” of individuals only mattering if they survive n generations. Key point since people seem to forget that most of evolution is noise, and the only way something is a signal is if it survives.