The Trajectory Lee Spector - The Next Phase of Evolution Is Artificial (Worthy Successor, Episode 27)
6 snips
Apr 17, 2026 Lee Spector, an Amherst professor known for evolutionary computation and genetic programming, discusses evolving executable programs and autoconstructive evolution. He contrasts evolutionary methods with LLMs, explores how evolution can produce radical novelty and new neural architectures, and reflects on recognizing alien intelligences and what a flourishing far future might require.
AI Snips
Chapters
Books
Transcript
Episode notes
Use Simple Evolutionary Abstractions To Discover Powerful Mechanisms
- Simple Darwinian abstractions (variation and selection) let researchers explore how far evolutionary ideas go without fully mirroring biology.
- Spector argues fidelity to biology isn't required; powerful abstractions can still reveal useful mechanisms.
Evolution Excels At Producing Radically Novel Designs
- Evolutionary methods shine when you want radically novel designs unlike anything in existing data or training corpora.
- Spector contrasts LLM recapitulation of training data with evolution's direct search for genuinely new solutions.
Programs That Make Their Own Offspring
- Auto-constructive evolution has programs construct their own offspring rather than relying on human-designed mutation operators.
- Spector recounts early experiments where programs both perform tasks and generate children, letting reproductive methods evolve too.
