[cjbgkagh]

Think of how complex LLMs are and yet their emergent behavior is based on a few very simple rules and an easy to understand loss function. Due to the emergent behavior things can have the appearance of being far more complicated than they actually are. It is hyper-dimensional and granted humans are not good at that but computers are and some humans are capable of using computers as a tool in that way.

  >> The Fourth Law of Behavior Genetics, as proposed by Chabris, Turkheimer, and others, states that a typical human behavioral trait is associated with many genetic variants, each accounting for a very small percentage of the behavioral variability.

I think a big part of this assumption is the use of Linear Regression in GWAS studies, the problem is that SNPs have a multicollinearity problem where the inputs are not independent from each other. That the results of these studies reflect that assumption should be of no surprise. I think using better math this stuff is more easy to detect.

[cannonpr]

So… I hope your version of trying to understand life is a fruitful path, however my understanding of the underlying biomechanics, messy, deeply “physics” based and 3d, leaves me with confused more often than not. The behaviour is emergent but not out of simplicity.

Biomolecular condensates: https://pmc.ncbi.nlm.nih.gov/articles/PMC9974629/

[cjbgkagh]

In a similar analog to machine learning there are fundamentalists approaches and there are empiricists approaches. Most medical researchers need to be able to defend their work and it’s easier to defend fundamentalist ideas - but they spend so much time arguing over minutia that they often miss the forest from the trees. Not to say that such work is unimportant but the academic environment is rather suboptimal in driving research direction. It’s hard to navigate uncertainty when deviations from what is formally known are heavily punished.

Take for example that we can read DNA and therefore have a list of every bioactive peptide that the body produces, we also have WGS so we can identify people with changes to these peptides. By using algorithms on these populations we can understand quite a lot about what these peptides do without understanding the mechanism by which they do it.