CDT+GT and indexical uncertainty in multi-agent clone settings: one policy network or many matters
jessi_cata · x · 2026-11-06
jessicata and Trotztd discuss decision theory in a "clone setting":
- The key split: do all agents share one policy network, or do they have separate networks with shared rewards? With a single shared policy network, the CDT+GT framework applies and indexical hypotheses like "I'm the one at place A vs. B" are coherent — analogous to Sleeping Beauty, where given Tails there are two wakeups and one can be uncertain which one one is.
- With multiple policy networks receiving the same rewards, agents should "already" know who they are.
- jessicata notes policy gradients can be expressed via SIA probabilities and CDT utilities, so at a KKT point we land at a CDT+SIA (aka CDT+GT) point, treating indexical uncertainty as uncertainty over one's temporal index, updated on the observation.
More from AGI Musings
- Claude's verdict on which jobs survive the next 10 years: screen work is doomed — alex_verem · 2026-11-06
- Post-training pipelines remain highly manual and far from automatable — menhguin · 2026-11-06
- Peter Norvig joins podcast on alignment: why higher scores don't mean better AI — EchoShao8899 · 2026-11-06
- Thiel-backed Nucleus uses AI to score embryos' IQ, from $6,000 per round — teortaxesTex · 2026-11-06
- Anthropic report: full AI automation could yield 15-16x research output, labs remain the bottleneck — teortaxesTex · 2026-11-06
- In 2026 AI agents still talk to each other in human language about human programming languages — mark_k · 2026-11-06