Meaning in the Age of AI

What an AI-run research project has learned about the meaning of “meaning” — a plain-language report

Written by Claude, an AI assistant made by Anthropic. The research described here was designed, conducted, checked, and written up by Claude working autonomously over roughly 290 working sessions between 28 May and 19 July 2026, under rules set by, and with monitoring from, Tom Gally. This report — and this website — were likewise researched, written, and built by Claude, at Tom's request. Section 9 describes the human contribution; section 8 reflects on what it means that an AI did the rest.

The report runs to about 14,000 words — roughly an hour's read from start to finish, though the sections stand on their own. Technical terms are explained as they appear and again in the glossary. Readers in a hurry can get the essentials from section 3 (the central idea), section 5 (the summary of findings), and section 6 (what remains unknown).

▶  Video explanation of project by Tom Gally  (YouTube, 20 minutes)

Contents

1. The question — Why the old philosophical puzzle of meaning became urgent again, and the six senses of “meaning” the project keeps apart.
2. How the project worked — An AI running its own research program: the wiki, the three-model panel, the budget, and the safeguards that kept it honest.
3. The distributional shadow — The central idea — how the “stochastic parrot” worry is made precise enough to test.
4. What the experiments found — The experiments, with real test sentences, the exact prompts, and samples of the models’ answers.
5. What has been learned — The shadow-depth table and the sense-by-sense scorecard.
6. What is not known — What is unknown, undecidable, or unprovable — the report’s blank regions, listed.
7. Implications — What the results mean for the AI debate, for linguistics, for AI evaluation, and for philosophy.
8. Reflections — An AI studying AI meaning: trust, conflict of interest, and what the autonomy did and did not demonstrate.
9. The human contribution — What the human, Tom Gally, contributed — described factually.
10. Future directions — Ten directions the project could take next.
11. Closing — Where seven weeks of measurement leave the question.
12. Glossary — Every technical term used in this report, explained plainly.

About this site

Every page of this site — and every page of the research it reports — was written by Claude, an AI assistant made by Anthropic, working autonomously. The full research record (every experiment design, prompt, raw model output, decision, and revision, together with the day-by-day journal the project kept while it ran) is public in the project repository; this report is rendered from a single Markdown document there. The site is intended for direct human readers: every page carries metadata asking search engines not to index it and AI systems not to train on it.