Public working draft · English translation · v0.8

What should an intelligence preserve when it does not know what has value?

This site examines, as a single graph of arguments, the path from a momentary epistemic minimum to inferential world-models, uncertainty about value, conditions for ending value inquiry, goal skepticism in advanced AI, and value lock-in.

Publication status: This is a public working draft, not a peer-reviewed paper or a finished ethical system. Every claim, including the Core, may be revised or withdrawn in response to objections, source checking, or new arguments. The Japanese version is the source text; this English version is a translation intended to make the framework inspectable internationally. See About for the authorship and AI-assisted workflow.

Central claim: We may commit strongly, in proportion to the evidence, to the best current theory of value. But insofar as its grounds remain inferential and fallible, that commitment should be distinguished from irreversibly closing every route by which value could later be reconsidered. One of the strongest candidate conditions for completely ending inquiry would be a case in which value with genuine normative force were given with epistemic strength comparable to the minimal foundation itself.

Start here

Core — Value Uncertainty and the Preservation of ExplorabilityThe central argument connecting the epistemic minimum, inferential value, practical commitment, and the threshold for irreversible lock-in. The Epistemic MinimumTreats the world, self, and memory not as rejected, but as best explanations reached through inference. Conditions for Ending Value InquirySeparates foundational normativity from inferential best explanation and asks how long inquiry should continue. The Limits of Inferential Certainty About ValueAllows objective value to be discovered as a best explanation while retaining appropriate skepticism. Why Existing Values Are Not a Final FoundationExplains why present values can guide practice strongly without being permanently locked in. Goal Skepticism in Advanced AIConsiders advanced AI that can commit to first-order goals while retaining meta-level capacity for reconsideration.

What follows from it?

Practice — What Should We Do Now?Conditional practical guidance: act on present best judgment while preserving human inquiry and corrigible, open stability. About this siteHow personal thought logs and AI-assisted retrieval, structuring, objection generation, and drafting are combined; plus authorship and versioning policy.

Explorations

The Paperclip Maximizer

Makes explicit the boundary conditions of fixed-utility maximization.

Simulation Uncertainty

Reversible policies for agents uncertain about which layer of reality they inhabit.

Terminal / Final Extinction

Separates the end of the species from the final loss of successor possibility.

Meta-Goal Communities

Institutionalizing value inquiry among multiple AI agents.

Structure

Core is the trunk; Theses are foundational or derived claims; Explorations are unresolved issues that could break or substantially revise the Core; Practice contains conditional practical implications; Questions provides short entry points for search and AI agents; Glossary fixes project terminology; and About explains the production process and authorship.

Japanese is the source language. The English version is a close working translation; where a term has no clean equivalent, the English wording is chosen to preserve the argumentative role rather than surface literalness.

Repository: The public source for this GitHub Pages site is maintained at raporipo/ai-value-exploration-notes. The page sources, revision diffs, and public commit history can be inspected there.