AI Debate 2: Night of a thousand AI scholars

Abstraction: Sixteen scholars debate moving AI beyond deep learning

Key points:

• Gary Marcus moderated 16 scholars (Dec 2020) including Kahneman, Fei-Fei Li, Judea Pearl, Rich Sutton, and Yejin Choi; 3,500 pre-registered
• Convergence noted among Bengio, LeCun, and Marcus: deep learning alone (e.g., GPT-3) is insufficient — causality and reasoning must be added
• Rich Sutton declared reinforcement learning "the first computational theory of intelligence" and argued AI needs an agreed-upon computational theory
• Pearl proposed causal reasoning as "goldmine" for "deep understanding" distinguishing three levels: "What is?", "What if?", and "If only?"
• Celeste Kidd and Margaret Mitchell criticized "bias laundering" in ML pipelines; Kidd called out Google's firing of Timnit Gebru
• Ryan Calo argued AI ethics principles are "largely meaningless" without legal enforcement; Kahneman was pessimistic about correcting human biases

Connections: Gary Marcus · Fei Fei Li · Judea Pearl · Daniel Kahneman · Rich Sutton · Deep Learning · Causal Reasoning · Neurosymbolic AI · AI Safety · Algorithmic Bias

Source: https://www.zdnet.com/article/ai-debate4-2-night-of-a-thousand-ai-scholars/