,
8 tweets,
2 min read
Read on Twitter
It's difficult enough to derive emergent properties from inanimate components. Now imagine how difficult it would be to derive emergent properties from animate (or intentional) components. medium.com/intuitionmachi…
It is easier to discover models that demonstrate how (generative) than finding models that explain why (explanatory). This is deep learning approaches that can build generative models move us an incremental step forward in understanding complex systems.
Having an ensemble of generative models can extricate the invariances of these models and thus lead to explanatory models. One cannot explain what is different if one cannot show what remains the same.
Nature's evolutionary process never needs to know "the why", it only needs to discover how. It discovers how reusing the adjacent possible. Its designs for individuals are compatible with the environment, but it is not optimal. Optimality requires understanding why.
Evolution creates Rube Goldberg like machinery that contains many unnecessary parts. However, evolution cannot optimize away something that is unnecessary. In fact, it doesn't, what's unnecessary now may be what is necessary for the future.
What's necessary is also not encoded in an organisms DNA. Organisms are dependent on their environment. We are surrounded by our own personal microbiome that is a necessary ingredient for our survival. This microbiome evolves synergistically with our evolution.
@threadreaderapp unroll
