Jeff Clune: Genetic Algorithms, Quality-Diversity, Curiosity

Topics covered
Popular Clips
Episode Highlights
Origins
The concept of modularity in evolution is a fascinating topic explored by . He discusses how modularity, the organization of systems into distinct modules, is prevalent in nature but its origins remain a mystery. Clune suggests that modularity might not have evolved for adaptability, but rather as a solution to the physical constraints of biological systems, such as the brain's wiring 1.
Modularity is everywhere, but we don't really know where it came from.
---
By simulating evolution with and without connection costs, Clune demonstrates that modularity emerges when there is a cost for connections, providing a new perspective on its evolutionary origins 2.
  Â
AI Applications
Modularity's principles extend beyond biology into artificial intelligence, addressing challenges like catastrophic forgetting. explains that while humans can switch between tasks without losing previous skills, AI systems struggle with this, often erasing prior knowledge when learning new tasks 3.
Animals and humans forget gradually, whereas machine learning forgets catastrophically.
---
Clune's research explores using neuromodulation to create modular AI systems that can retain learned skills by turning learning on and off in specific brain regions, effectively solving catastrophic forgetting in simple neural networks 4.
Related Episodes


Joel Lehman: Open-Endedness and Evolution through Large Models
Answers 383 questions

Kyunghyun Cho: Neural Machine Translation, Language, and Doing Good Science
Answers 383 questions

Luis Voloch: AI and Biology
Answers 383 questions

Eric Jang on Robots Learning at Google and Generalization via Language
Answers 383 questions

Yannic Kilcher on Being an AI Researcher and Educator
Answers 383 questions

Hugo Larochelle: Deep Learning as Science
Answers 383 questions

Eric Jang: AI is Good For You
Answers 383 questions

Joel Simon on AI art and Artbreeder
Answers 383 questions

François Chollet: Keras and Measures of Intelligence
Answers 383 questions

Connor Leahy on EleutherAI, Replicating GPT-2/GPT-3, AI Risk and Alignment
Answers 383 questions

Joon Park: Generative Agents and Human-Computer Interaction
Answers 383 questions

David Chalmers on AI and Consciousness
Answers 383 questions

Yann LeCun on his Start in Research and Self-Supervised Learning
Answers 383 questions

Kanjun Qiu and Josh Albrecht: Generally Intelligent
Answers 383 questions
