Computational rationality: A converging paradigm for intelligence in brains, minds, and machines

摘要: After growing up together, and mostly apart in the second half of 20th century, fields artificial intelligence (AI), cognitive science, neuroscience are reconverging on a shared view computational foundations that promotes valuable cross-disciplinary exchanges questions, methods, results. We chart advances over past several decades address challenges perception action under uncertainty through lens computation. Advances include development representations inferential procedures for large-scale probabilistic inference machinery enabling reflection decisions about tradeoffs effort, precision, timeliness computations. These tools deployed toward goal rationality: identifying with highest expected utility, while taking into consideration costs computation complex real-world problems which most relevant calculations can only be approximated. highlight key concepts examples show potential interchange between computer neuroscience.