The Future of (Artificial) Intelligence
summary
Shows a probabilistic programming system called Bayesian Logic (BLOG) that generates large, dynamic graphical models and runs MCMC over data from ~150 seismic stations to detect underground blasts. The NETVISA model reduces prior automated detection error by ~2–3× and a later model achieves a factor of 10 improvement over combined human experts, using raw seismograms; also demos a laundry-folding robot Brett headed for a museum. Proposes value alignment via inverse reinforcement learning and uncertainty to yield provably beneficial shutdown-safe agents.
AI-written summary · about
chapters (27)
- 0:00Introduction
- 0:25Progress in AI
- 1:35probabilistic programming
- 8:02machine learning
- 8:46machine translation
- 10:08Cumulative discovery
- 11:08We dont just fail
- 11:38What does that mean
- 12:12A step change in our civilization
- 12:39Killer robots
- 13:45The end of the human race
- 15:17Alan Turing
- 16:15Norbert Wiener
- 16:52Value misalignment
- 17:39AI drives
- 18:55AI chess match
- 19:22Human objective is mismatched
- 19:51Selfdefense
- 21:08Humanlevel AI
- 21:48History Lesson
- 23:57Center for Human Compatible AI
- 26:03Inverse reinforcement learning
- 26:51Active process of cooperation
- 29:20Uncertainty in objectives
- 31:35Difficulties with humans
- 32:10Can we change AI
- 32:56Question
related talks
01h 03mAn Overview of Probabilistic Programming
56mIron Man or Ultron: Is AI here to help us or hurt us?
18mNext-Generation AI Agents: Orchestrating Technological Evolution
30mIntroduction to AI Safety. Let's be ready before it's too late!
17mUnveiling the AI Engima: Navigating the Promises & Perils of Artificial Intelligence
21mLearning from History: The Past as a Compass for Systems Engineers in the Age of AI
43mAI Systems: Foundations for Artificial Minds
31mFrom Single AI to Multi AI Systems