Gary Marcus, Professor Emeritus at NYU and co-founder of Robust.ai, discusses his recent book Rebooting AI: Building Artificial Intelligence We Can Trust.

On this episode of Technology & Prose, Gary Marcus joins host Nikita Aggarwal to talk about the “AI chasm”— the gap between the goals and reality of artificial intelligence (1:40) GPT-3 and the limits of deep learning (4:38) how to build AI systems that have knowledge, reasoning and understanding (9:14) the politics of AI research (and more on the limits of GPT-3) (16:00) innate knowledge in machines and the need for a hybrid approach to building AI (33:25) the Montreal AI Debates, the System 1/ System 2 analogy and lessons from developmental psychology (37:04) and Gary's new venture, Robust.ai (43:15) .

References

Marcus, Deep Learning: a Critical Appraisal

Goyal and Bengio, Inductive Biases for Deep Learning of Higher-Level Cognition

LeCun, Hinton, Bengio, Deep Learning

Kahneman, Thinking Fast and Slow

Montreal AI debates https://montrealartificialintelligence.com/aidebate/ ; https://montrealartificialintelligence.com/aidebate2.html

Marcus, Guitar Zero: The Science of Learning to be Musical

GPT-3

• Open AI’s GPT-3 model
• Marcus and Davis on GPT-3 in MIT Tech Review
• Guardian article ‘written’ by GPT-3
• Yann LeCun on GPT-3

DeepMind

• Alphafold
• Mastering the game of Go in the absence of human knowledge

Recorded on January 28th 2021