Legible optimality in biology is interesting in part because it sheds light on explainability in AIs. I’d expect AIs to be explainable ar about the same rate you can answer the question, “what is this organism uniquely good at optimizing”?
Conversation
Replying to
Except that evolution isn't optimization. Phenotypes settle into "good enough" local maxima. AI has fewer bounds checks and risks runaway optimization.
1
Replying to
Yeah that’s exactly what I’m trying to probe via examples where biology does optimize more than it usually does. See previous tweets and my recent post ribbonfarm.com/2018/04/24/sur
Replying to
this is critical, and poorly understood. Evolution perfectly represents the ‘No Free Lunch’ theorem in machine learning.
1


