Taegon Kim, Ph.D.
April 21(Thu) - April 21(Thu), 2022
12PM
ZOOM (ID: 728-142-6028)
Neuro@noon Seminar
Date: 12:00 PM, Thursday, April 21st
Speaker: Taegon Kim, Ph.D.
(Brain Science Institute (BSI) in Korea Institute of Science and Technology (KIST))
Title: A neuroscientist's perspectives on bottom-up approaches toward biologically plausible artificial intelligence
Abstract: It seems natural and has been considered as a proper strategy that mimicking structure or functional mechanism of a biological system is chosen to overcome saturated performance and efficiency of an artefact. Modern connectionist’s implementation of artificial intelligence is still far from the ultimate general intelligence and is facing the limited scalability arisen from concomitant hardware issues. When Turing’s bottom-up approach to build a ‘thinking machine’ was criticized as the ‘sure’ but ‘too slow and impracticable’ way by himself, he must have realized that those hurdles adjure us to learn more carefully from a biological neural system, the brain. Although a simple copy-and-paste strategy is hindered by the extreme level of complexity of a brain, the pipeline that bridges between concurrent gradual progresses of neuroscience and the AI engineering may let us deal with both challenges. A possible example pipeline can be analogous to an artificial neural network based on the simplified model that well captured the essential nature of computation in a certain neural circuit from a state-of-the-art neuroscientific discovery although building and stabilizing such easy-looking pipeline would be very difficult. In this presentation, current gap between the ideal and the status at which we are, and our attempt to build such pipeline will be introduced. Also some honest concerns from a neuroscientist living in the era of rapid rising of AI will be casually shared. In addition, by choosing the system well characterized yet bearing inexhaustible novelty, the cerebellum, the possibility that a neuromorphic system can be used as the proof-of-concept of neuroscience will be shown and eventually one of the bottom-up approach of BNN-inspired AI will be suggested.