Currents 072: Ben Goertzel on Viable Paths to True AGI

Jim talks with Ben Goertzel about the ideas in his recent essay “Three Viable Paths to True AGI.” They discuss the meaning of artificial general intelligence, Steve Wozniak’s basic AGI test, whether common tasks actually require AGI, a conversation with Joscha Bach, why deep neural nets are unsuited for human-level AGI, the challenge of extrapolating world-models, why imaginative improvisation might not be interesting to corporations, the 3 approaches that might have merit (cognition-level, brain-level, and chemistry-level), the OpenCog system Ben is working on, whether it’s a case of “good old-fashioned AI,” where evolution fits into the approach, why deep neural nets aren’t brain simulations & attempts to make them more realistic, a hypothesis about how to improve generalization, neural nets for music & the psychological landscape of AGI research, algorithmic chemistry & the origins of life problem, why AGI deserves more resources than it’s getting, why we may need better parallel architectures, how & how much society should invest in new approaches, the possibility of a cultural shift toward AGI viability, and much more.