Towards Superintelligence Alignment
agendareviewed

Kosoy / infra-Bayesianism & LTA

Can learning-theoretic and infra-Bayesian frameworks type real alignment failures—misspecification, inner daemons, recursive self-improvement—and does precursor-utility pointing survive simulation and ontology ambiguity?

Introduction

Vanessa Kosoy’s learning-theoretic agenda (LTA) builds a mathematical theory of intelligence and alignment grounded in regret bounds, daemons, and nonrealizability. Infra-Bayesianism and infra-Bayesian physicalism are major constructive layers—not the whole agenda. The outer-alignment strand Physicalist Superimitation (formerly PreDCA) proposes precursor-utility pointing via the bridge transform. Its walls map onto Embedded Agency, Value Learning, Value Referent, Inner Alignment, and Grounding Drift without replacing this project’s System–bundle–correction ontology.

Who carries it: Vanessa Kosoy (+ Appel; logical-induction neighborhood via Garrabrant)

What they aim to do. Build a mathematical theory of intelligence and alignment—regret-bounded agents under deep uncertainty, with a constructive outer-alignment endpoint (precursor-utility assistance).

The hard question. Can learning-theoretic and infra-Bayesian frameworks type real alignment failures—misspecification, inner daemons, recursive self-improvement—and does precursor-utility pointing survive simulation and ontology ambiguity?

What they produce. The learning-theoretic agenda, the Infra-Bayesianism sequence, and the Physicalist Superimitation / PreDCA outer-alignment protocol built on infra-Bayesian physicalism.

Key terms. Key terms include infra-Bayesianism, the learning-theoretic agenda, PreDCA / Physicalist Superimitation, imprecise probability, nonrealizability, daemons, and the bridge transform.

Related field cruxes. Embedded Agency; Value Learning; Value Referent; Tiling; Inner Alignment; Grounding Drift

What they contribute. Model-class misspecification and grain-of-truth analysis; regret-bounded agents; inner daemons; infra-Bayesian physicalism and the bridge transform; PreDCA as a sibling outer-alignment endpoint reached through a different formal path than CIRL-style pointing.

How this project treats it. This work does not replace this project’s System–bundle–correction ontology; its walls map onto Embedded Agency, Value Learning, Value Referent, Tiling, Inner Alignment, and Grounding Drift, with Acausal Coordination as a logical-induction neighborhood cousin. The PreDCA/PSI endpoint alone is not sufficient; value-bundle-transport , bearer-persistence , and the correction process remain load-bearing.

See the coverage matrix for evidence tagged to this agenda, and the glossary for shared terms.