Decision-Based AI Governance

Govern the decisions AI makes
— not the capabilities it has.

Einstein warned that the atom changed everything save our modes of thinking. AI is that test for our generation.

Yong Tao —
Stanford-trained decision analyst and researcher
Fellow, Society of Decision Professionals (SDP)
Venture investor
Author of Steering AI for Humanity, a completed book on decision-based AI governance.

Subscribe to the newsletter See the shift

The Shift

A new mode of thinking

The unleashed power of the atom has changed everything save our modes of thinking and we thus drift toward unparalleled catastrophe.
Albert Einstein — telegram to prominent Americans, 24 May 1946 (New York Times, 25 May 1946)

Eighty years on, the unleashed power of the algorithm confronts us with the same lag between what we can build and how we think about governing it. The reflex has been to chase capabilities — to legislate against this model size, that benchmark, the newest frontier system. But capabilities move faster than any statute can, and a law written for today's frontier is obsolete before the ink dries.

There is a more durable place to stand. Every AI system, however capable, earns its consequences at the moment it makes a decision. Govern the decision — who authorizes it, whether a human can override it, whether it stays inside the law — and you have a framework that does not need to be rewritten each time the technology leaps ahead. This is the shift: from governing what AI can do to governing what AI decides — a decision-based, capability-agnostic paradigm, and a minimum viable legislation for the age of powerful artificial intelligence.

That is the argument of my work, developed with the late Professor Ronald A. Howard beginning in 2022 — before ChatGPT reached the public — and it is the through-line of everything here: the book, the papers, and the newsletter.

The Five Pillars and
six AI Mandates

The framework rests on the Five Pillars — a completeness argument for what any adequate AI governance regime must cover — operationalized through Six AI Mandates (AIMs 1–6) that function as minimum viable legislation.

I

Serving User Values

II

Protecting Society's Values

AIM 1 · ELR (Ethics, Laws, and Regulations) Compliance
III

Preserving Human Control

AIM 2 · Resource AuthorizationAIM 3 · Human Override
IV

Limiting AI Rights

AIM 4 · No AI PersonhoodAIM 5 · AI Owns No ResourcesAIM 6 · AI Holds No Offices
V

Mitigating Other Risks

Together, the pillars define the whole surface a governance regime must cover — no pillar is optional.

The Tightening Hoop

The framework's guiding image comes from Journey to the West. The monk Tang Seng cannot match the raw power of his disciple, the Monkey King — so he governs not the monkey's strength but the monkey's choices. A golden hoop circles the monkey's head, and at the monk's chant it tightens — but only when a decision crosses the line. That is decision-based governance in a single image: not a cage on capability, but a constraint triggered by the decision itself.

The Book

Steering AI for Humanity

The Decision-Based Minimum Viable Legislation

A capability-agnostic blueprint for governing the decisions AI systems make — before they make them for us. Across nine chapters, the book builds the Five Pillars, the six AI Mandates, and a workable path from principle to statute, drawing on decision analysis, the ethics-laws-and-regulations (ELR) tradition, and an unlikely guide from classical Chinese literature.

A completed manuscript.

The Newsletter

New thinking for governing AI decisions.

A biweekly essay on decision-based AI governance — legislative drafting, the Five Pillars in practice, the geopolitics of strategic AI decisions, and dispatches as the book moves toward launch. Written for legislators, builders, and anyone who suspects we are governing the wrong thing.

Some issues are co-written. If you are a researcher working on adjacent questions, joint essays are welcome here — reach out to co-author a piece.

Yong Tao

About

Yong Tao holds a PhD in Decision and Risk Analysis from Stanford, where he studied under the late Professor Ronald A. Howard, the founder of the field. He is a Fellow of the Society of Decision Professionals and an early-stage healthcare venture investor in Silicon Valley.

His work argues for governing AI decisions rather than AI capabilities — a decision-based framework expressed through the Five Pillars and six AI Mandates as minimum viable legislation. It is a lens that draws as readily on Bayesian decision analysis as on Journey to the West, whose monk-and-monkey allegory anchors the book's central metaphor.

Selected writing & research