Natural things as ML systems

Evolution.

• Evolutionary landscape talks about:
  • fitness function
  • environment
  • organisms
  • genes (the selfish gene)
  • evolution
• This maps onto ML systems pretty well:
  • fitness = objective function/loss.
  • environment = the distribution.
  • evolution = optimizer.
  • genes = the parameters.
• the genome instantiates a generative model of the organism

The economy.

• price signals are gradients. prices encode supply/demand tension. they are gradients telling agents which direction yields more utility per input.
• agents follow gradients. If mowing yards yields $50/hour and your next-best option yields $20/hour, you mow.
• entrepreneurs don’t just follow gradients. they bend them.
  • if their constructed future becomes real (demand materializes), they capture the delta between cost and created utility (profit).

But people don’t just do things for money. Market price signals capture “private” utility, not individual. Utility can encompass anything, like your values.