Toward machines that can daydream

摘要

This paper provides a new insight into the possibility of building a plausible computational model of human mind. We take a fresh look at some ideas propounded more than a century ago by William James and Sigmund Freud, which have been re-considered recently by Peter Naur and the ATR Brain-Building Group, respectively. Naur proposes his synapse-state theory of human mind (SST), while the research at ATR resulted in the machine psychodynamic (MPsiD) paradigm. We argue that SST and MPsiD propose complementary ideas about implementation of mental functionalities, including those related to the quest for consciousness. The 20 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> -century AI gave machine the ability to learn. The great challenge for the 21 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> -century AI is to make a robot actually want to learn. MPsiD proposes a solution based on the notion of pleasure defined as a measurable quantity to be used as a general reinforcer. SST proposes a neuroscience-inspired architecture, where the key blocks are item-nodes, attention-node, and specious-present excitation. MPsiD potentially supplements SST with a pleasure node and related pleasure principle.