Telocracy

Joe Edelman, co-founder of the Meaning Alignment Institute, Pax Machina (proposal no. 7), 2026-11-03. Telocracy, with appendices on specifying mandates, purpose review, and epistemic communities. An introductory note by Nora Ammann of the OpenAI Foundation, who writes in a personal capacity, precedes the essay.

Summary

Edelman argues that optimistic visions of AI, such as Machines of Loving Grace, leave three questions open. Where will people find meaning and mix across social divides once work no longer provides it? What should humans still decide, so that they keep learning and exercising judgment? And how do people keep power over institutions run increasingly by AI agents that tirelessly chase whatever metrics they are given?

His answer, drawing on the political tradition of associationalism, is a society of many institutions bound to written, revisable purposes, and of many lightweight groups that people can easily find, join, and leave. Today we make “consumer choices” among products; telocracy (from the Greek telos, purpose) would add “values choices” among the purposes we want to take part in. Edelman says it “doesn’t require powerful AI, but it helps a lot”.

The essay’s main mechanisms:

• Telocratic governance. An institution is judged by fidelity to a detailed mandate, closer to Anthropic’s constitution for Claude than to bylaws, rather than by procedure or by what stakeholders demand. AI makes this practical by applying the mandate consistently across thousands of cases. Purposes bind more tightly as outsiders come to rely on them (“living commitments”), and may change only by “legitimate deepening”, not by a faction swapping in a new purpose. Present members are custodians who “inherit something they can revise, but not appropriate.”
• Purpose-bound property. Assets such as land, brands, and compute are licensed on a purpose, and auditors check that they are used for it. In the simplest version any purpose that creates value, rather than moving it from one group to another, qualifies, which rules out holding capacity just to block a rival.
• Values-rich markets. Pools of buyers, represented by AI agents, pay suppliers jointly for outcomes in their lives. A vignette contrasts an AI companion paid by engagement with a supplier paid when widowed members see friends in person.
• Associational discovery. AI finds people with a shared aim and proposes a small, time-limited joint act that happens only if enough opt in, then lets commitment grow step by step toward a charter and shared property.
• Government by mandate. Agencies staffed by AI decide cases against their stated purpose and explain why, rather than applying rigid rules, and the public helps revise mandates through AI-assisted deliberation such as Moral Graph Elicitation.
• Epistemic vitality. A value in a mandate should be enforced only while a living community of practitioners can still judge it, so some hands-on human work must be deliberately preserved.

Edelman also argues that telocracy would steer economic growth toward recognizable human purposes, away from proxy metrics such as clicks that can crowd them out (compare Goodhart’s law). He presents the whole as a picture of a good future: many communities of purpose, humans still judging what their values should be, and power tied to purposes they understand.

Responses

Ammann’s note calls it the most ambitious and worked-out institutional proposal she has read in a long time, and names three open problems. If power flows through what people say they value, pressure will flow toward shaping and gaming those values. Someone must build, own, and vouch for the models, compute, and data that interpret the mandates. And there is no clear path from here to there.

Why it matters here

It is a rare positive vision that is concrete about institutions, and it takes a middle position on Should the good future be designed or left open: it designs the mechanisms that hold institutions to their purposes, but leaves the purposes to the people who join them. It belongs in the A good future path as a test case for the stability problem: whether purposes written down and enforced by AI can survive competition and capture.