Blueprints for a Universal Reasoning Machine
summary
Proposes a working blueprint for a proximate “universal reasoning machine” that composes probabilistic inference, program synthesis and causal counterfactuals. Uses probabilistic programming (Stan/Pyro/Omega), a compiler-like parametric inverse to invert forward simulators, and trained neural stack implementations to realize algebraic data-structure laws; demonstrates program-from-video synthesis (Autumn) and counterfactuals on a car-crash and glucose case studies, with empirical results showing plausible posterior distributions and working inversion proofs by construction.
AI-written summary · about
chapters (12)
- 0:00Start
- 1:07Building Blocks of Universal Reasoning
- 2:40Mental Rotation
- 4:25Simulation Models
- 6:00Bayesian Inference
- 8:00Marr's Levels of Analysis
- 9:42Neural Stack
- 10:39Abstract Structures
- 19:22Parametric Inversion of Compositions
- 22:10Changing Program Meaning
- 25:34Causal Probabilistic Programming
- 28:02Counterfactual Generative Models
related talks
59mA Welcome to Causal Probabilistic Programming
01h 03mA Personal Viewpoint on Probabilistic Programming
01h 10mNeurosymbolic Program Synthesis: Bridging Perception and Reasoning in Real-World(…)
07mProbabilistic Reasoning and Learning for Trustworthy AI
01h 03mAn Overview of Probabilistic Programming
41mFrom AI Software Engineers to AI Knowledge Workers
17mProgrammatic Strategy Synthesis: Resolving Nondeterminism in Probabilistic Progr...
43mThe Power of Probabilistic Programming