The thing is, living beings don’t learn by brute-force trial and error like mathematical optimization models you mentioned. Besides enormous energy spent, an in

The thing is, living beings don’t learn by brute-force trial and error like mathematical optimization models you mentioned. Besides enormous energy spent, an individual organism will just be eaten by predator on another iteration of its ‘error minimization’ loop. the idea of learning via reinforcement, that came from Skinner behaviorist experiments has been long discredited in cognitive psychology. (I highly recommend Wayne Wickelgren’ work on learning and memory if you’re interested, it’s brilliant and concise http://www.columbia.edu/~nvg1/Wickelgren/ http://www.columbia.edu/~nvg1/Wickelgren/ ) Biological plausibility might not be needed for recognizing check signatures or images of traffic lights, where backprop is working just fine, but I believe true cognition would require such energy expenditures that brute-force trial and error will never be feasible. Moreover such error correction imposes artificial constraints that limit the amount of information that can be learned, kind of like those mechanical calculators of the 17th century with gears and wheels and crude mechanical actuators.