Every Definition of Intelligence Is Wrong. Here's Why — Michael Bennett

Value of the Information & Strength of the Argument

The value of the information is high for those interested in theoretical foundations of AI and philosophy of mind. Bennett provides a coherent argument for why current definitions of intelligence are inadequate, offering a synthesis of ideas from AIXI, active inference, and biological systems. His argumentation is logically structured, though some points are speculative and lack empirical support. He effectively critiques the simplicity bias in AI and emphasizes the importance of embodiment and causality.

Scientific Rigor, Source Quality, Title Accuracy

Bennett references several academic works, including his own papers and those of Legg, Hutter, Chollet, and Friston. The sources are credible and relevant. The title is somewhat clickbait but accurately reflects the content’s critical stance on definitions of intelligence. The discussion is rigorous in its use of formal concepts, though it remains at a high level without deep technical detail.

The title is somewhat provocative and matches the content, as Bennett critiques various definitions of intelligence and proposes his own.

8/10

The discussion is grounded in formal definitions and references to academic papers, but it is primarily an expert opinion interview with philosophical speculation, lacking empirical validation.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:54Discussion of definitions of intelligence, including Pei Wang's definition.
• 07:06Comparison of AIXI and active inference formalisms.
• 10:45Exploration of causality, abstraction, and embodiment in intelligence.
• 25:51Introduction of computational dualism and mortal computation.
• 31:30Discussion of modern AI, AGI progress, and benchmarks.
• 35:00Hybrid AI approaches and the importance of self-organization.
• 39:35Consciousness and the hard problem, Bennett's perspective.
• 53:20Mention of the Diverse Intelligences Summer Institute (DISI).
• 54:17Living systems and self-organization, biological inspiration.
• 1:04:24Closing thoughts and summary.

• What the F*** is Artificial Intelligence — Bennett's paper critiquing definitions of AI.
• Are Biological Systems More Intelligent Than Artificial Intelligence? — Bennett's paper on biological vs. artificial intelligence.
• How To Build Conscious Machines — Bennett's PhD thesis on building conscious machines.
• Diverse Intelligences Summer Institute — Where the interview was filmed.
• Michael Timothy Bennett's website — Personal website with more information.
• Transcript of the interview — Full transcript of the conversation.

• Universal Intelligence: A Definition of Machine Intelligence — Legg and Hutter's definition, which Bennett critiques.
• On the Measure of Intelligence — Chollet's paper introducing ARC benchmark, discussed in the interview.
• The Book of Why — Pearl's work on causality, which Bennett references.

• The Bitter Lesson — Sutton's argument that scale and search are more effective than human-designed inductive biases, contrasting with Bennett's emphasis on biologically inspired approaches.