sleep is also 'reshuffling' of memory representations that's continuously happening to reconcile internal inconsistencies and converge on some coherent state t

sleep is also 'reshuffling' of memory representations that's continuously happening to reconcile internal inconsistencies and converge on some coherent state that can help with real-time predictions. this is likely the actual learning mechanism as opposed to error correction via stimulus-response reinforcement loop, so popular in AI research. Prof. Wilson has a great talk on learning as 'discovery of hidden connections': https://www.youtube.com/watch?t=1147&v=Vf_m65MLdLI&feature=youtu.be https://www.youtube.com/watch?t=1147&v=Vf_m65MLdLI&feature=y... also see his works on 'hidden state inference': http://web.mit.edu/wilsonlab/html/publications.html http://web.mit.edu/wilsonlab/html/publications.html