Title: Byoung-Tak Zhang, Seoul National University
Start date: Jan 19 2018
Start time: 11:15 am
End time: 12:15 pm
Location: Max Planck House Lecture Hall

Since its inception in 1950s, artificial intelligence (AI) has explored two main paradigms. One is the knowledge-based, rationalist, and “symbolic AI” and the other is the data-driven, empiricist, and “connectionist AI”. The former has focused on modeling the logical, analytical, conscious, slow-thinking mode of the mind (System 2 a la Kahneman) while the latter on the intuitive, emotional, unconscious, fast-thinking mode (System 1). However, human minds have both. We propose to study an alternative paradigm of “cognitive AI” that unifies the two previous paradigms and extends them to pursue human-level AI. Relying on the discovery of fundamental importance of embodied cognition, i.e. neocognitivism, the cognitive AI paradigm emphasizes the grounding of intelligence on actions through sensorimotor dynamics of interaction with the environment. This constructivist model of mind is not new and, in fact, has a long history (though fragmented). In this talk we give a systematic view on this tradition and present our own efforts toward cognitive AI. Specifically, we describe projects for building cognitive agents that learn everyday life in a multisensory world. A success story of this approach will be reported in which the humanoid robot AUPAIR successfully completed the missions in home environments and won the RoboCup@Home 2017 competition.

More Information