Gauguin, Descartes, Bayes: A Diurnal Golem's Brain
summary
Builds a small probabilistic program in Church that samples and conditions over program-space to infer its own source; uses a grammar-based generator, eval, infer, and equality conditioning over postfixes of past guesses. The constructed gauguine repeatedly guesses, and after roughly ten iterations the posterior collapses to a fixed-point program that outputs its own source. Claim: endogenous bootstrapping of self-knowledge is possible; demonstrated practically with working code on GitHub.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:27Identity crisis
- 6:51Reflection as knowledge
- 12:57Objections to self-knowledge
- 15:04Minimal model of self-knowledge
- 17:31First probabilistic program
- 20:39Inferring definitions
- 25:13Writing Goine program
- 29:11Running Goine program
- 33:23Counterarguments
- 36:12Program and humanity
- 38:40Conclusion
related talks
43mThe Power of Probabilistic Programming
01h 03mA Personal Viewpoint on Probabilistic Programming
24mExact Bayesian Inference for Loopy Probabilistic Programs Using Generating Functions
46mBlueprints for a Universal Reasoning Machine
14mExact Probabilistic Inference Using Generating Functions
36mProbabilistic Programs Which Make (Common) Sense
16mA Domain-Specific Probabilistic Programming Language for Reasoning about Reasoning(…)
38mProbabilistic programming and meta-programming in Clojure