Scaling Theory #20 – Melanie Mitchell: The Science of Artificial Thinking
Jul 7, 2025
Melanie Mitchell, a Professor at the Santa Fe Institute and author, dives into the complexities of intelligence in her conversation. She challenges misconceptions about AI's capabilities, critiquing traditional measures like the Turing Test. The discussion highlights 'jagged intelligence' and how biological insights can inspire more adaptable AI systems. Melanie also emphasizes the importance of metacognition in AI and explores the concept of emergence in complexity science, advocating for a more decentralized approach to developing truly intelligent machines.
AI Snips
Chapters
Books
Transcript
Episode notes
Mentorship by John Holland
- Melanie Mitchell shared a personal experience studying under John Holland, a pioneer in complexity science.
- She described him as a generous, enthusiastic advisor who deeply influenced her career.
Jagged Intelligence in AI
- Modern AI systems excel at language use but lack true understanding and consistent cognition.
- This leads to "jagged intelligence," showing strengths in some areas and surprising weaknesses in others.
Limits of the Turing Test
- The Turing Test misleads us by equating human-like language fluency with true thinking.
- It primarily reflects our tendency to project intelligence onto conversational entities.
