Listen

Description

AI can win championships, write convincing answers, and still get the facts wrong. What would make it worthy of our trust?

Spencer and Jacob sit down with Jonathan Schaeffer, the AI pioneer behind the checkers-playing system Chinook, for a candid conversation about privacy, bias, accountability, and the road to artificial general intelligence. Jonathan makes the case for independent AI audits, explains his approach to private local AI through KIND, and challenges the idea that language models alone will lead to AGI.

Jacob brings his experience as a CPA to the question of auditing systems that keep changing. Spencer explores whether language models could become one part of a larger architecture of intelligence. Along the way, Jonathan acknowledges that AI progress has repeatedly outpaced his predictions—even as he questions where the current approach can take us.

CHAPTERS
00:00 Meet Jonathan Schaeffer
01:02 A career built around chess, checkers, and games
04:48 Why trust becomes harder outside games
11:51 Privacy, local AI, and KIND
18:33 Can an AI model ever be unbiased?
25:30 The case for independent AI audits
29:42 How do you audit a system that keeps changing?
36:12 Grounding answers in your own data
38:30 Why language models make mistakes
43:26 Are LLMs a stepping stone toward AGI?
45:58 Will AI progress eventually level off?
49:50 Consciousness and reinforcement learning
54:56 Why education and accountability matter
57:33 Where to follow Jonathan

Guest links:
Jonathan / Synsira: https://synsira.com/
KIND: https://kind.synsira.com/
Jonathan's academic work: https://webdocs.cs.ualberta.ca/~jonathan/

Hosts: Spencer and Jacob
Guest: Jonathan Schaeffer