A working paper, published in the open. Criticism, corrections and clarifications are welcome.

Send yours →
AIGovSpec
Menu

Print edition · the slides

AI Governance From First Principles

Every slide, one per landscape page. Print it from your browser, or save it as a PDF. Speaker notes are on the presenting page.

Download the PDF ↓Present the slides →

  1. 1 / 15 · AI Governance From First Principles

    The essay, in fifteen slides

    AI Governance From First Principles

    One definition—and the questions it teaches us to ask.

    AIGovSpec

  2. 2 / 15 · Those things matter. But a list of instruments does not explain the problem they are meant to solve.

    When someone asks what AI governance is, we answer with a list

    laws · policies · standards · audits · risk frameworks · oversight bodies

    Those things matter. But a list of instruments does not explain the problem they are meant to solve.

    It also makes the field sound distant from anyone who does not write laws, build models, or attend policy meetings.

    AI Governance From First Principles

    2

  3. 3 / 15 · Practice came long before agreement on a definition

    Part one · How we describe AI governance today

    Practice came long before agreement on a definition

    2017–18

    Naming a field

    Layers of standards, ethics and law. A field of study.

    2019–21

    Principles, oversight, rights

    OECD, IEEE, UNESCO, the African Commission.

    2022–23

    Inside the organisation

    Rules, processes and tools; ISO standards; NIST.

    2024

    The word was everywhere

    The EU AI Act, the Council of Europe, the UN.

    2025–26

    Dialogue and agents

    Kenya, India, a UN Global Dialogue, agentic AI.

    38

    documents read, 2017 to 2026

    13

    offer a definition

    19

    describe an approach

    6

    build on the term

    From AIGovSpec’s History: how the world has thought about governing AI

    3

  4. 4 / 15 · Four patterns stand out

    Reading across the record

    Four patterns stand out

    01 Practice came first

    Institutions, laws and dialogues were built while the meaning was still being worked out.

    02 Definitions name the toolkit

    Most pair instruments (rules, processes, practices) with a purpose: what governance is made of.

    03 Adaptation is shared

    Many call for agile, living governance. When to act is raised far less often.

    04 Development and choice

    The African Union, Latin America and the UN frame governance around inclusion and policy choice.

    From AIGovSpec’s History: what the record teaches

    4

  5. 5 / 15 · Everything in our definition owes something to this record. What we add is a way of holding it together.

    Part two · Why a new definition

    Everything in our definition owes something to this record. What we add is a way of holding it together.

    What we take

    A field of inquiry · Dafoe

    Practice someone has to run · Mäntymäki, Kenya

    Benefits broadly shared · UN, African Union, Montevideo

    Governance that keeps learning · NIST, OECD, Singapore

    What we did not find in one sentence

    1 · All four things governance acts on

    2 · Those, a purpose and timing, held together

    AI Governance From First Principles

    5

  6. 6 / 15 · Our definition, from first principles

    Our definition, from first principles

    AI governance is the field of inquiry and practice concerned with how societies should shape the incentives, institutions, infrastructures and safeguards around AI to share its benefits broadly, prevent and address harm, and maintain the capacity to change course as today’s choices become tomorrow’s conditions.

  7. 7 / 15 · Breaking down the definition into smaller parts

    Part three · The sentence, part by part

    Breaking down the definition into smaller parts

    Who decides

    1 Inquiry and practice
    2 Societies
    3 Should shape

    Where it acts

    4 Incentives
    5 Institutions
    6 Infrastructures
    7 Safeguards
    ↻ Power, across all four

    What it delivers

    8 Share its benefits broadly
    9 Prevent and address harm

    Over time

    10 As today’s choices become tomorrow’s conditions
    11 Maintain the capacity to change course

    One thread runs through every part: a health ministry deciding whether to buy an AI tool that helps triage patients.

    AI Governance From First Principles

    7

  8. 8 / 15 · Governance is something we study, and something we do.

    Parts 1–3 · Who decides

    Governance is something we study, and something we do. 

    1

    Inquiry and practice

    learning what works, and making it count

    What do we need to learn—and how will that knowledge change what someone does?

    2

    Societies

    who gets to decide?

    Who has authority, who is affected, and how can those affected influence or challenge the decision?

    3

    Should shape

    which possibilities are we making easier?

    Which possibilities are we making easier to pursue—and for whom?

    Question to carry, under each part

    8

  9. 9 / 15 · Shaping happens in particular places. Here are the four objects.

    Parts 4–7 · Where governance acts

    Shaping happens in particular places. 
    Here are the four objects.

    4

    Incentives

    what does the environment reward?

    What makes that promise hold under pressure?

    5

    Institutions

    who can act?

    Who is responsible? And what makes them capable of exercising that responsibility?

    6

    Infrastructures

    what does this depend on, and who controls it?

    What are we becoming dependent on—and what alternatives will remain?

    7

    Safeguards

    what makes protection real?

    What prevents, detects, limits, or remedies failure—and what happens when it fails?

    Question to carry, under each part

    9

  10. 10 / 15 · Underneath all four lies one thing: power. It runs through all of them.

    ↻ Power · who sets the terms?

    Underneath all four lies one thing: power.
    It runs through all of them.

    Whoever controls incentives, institutions, infrastructures or safeguards sets terms for everyone else. And governance does not only constrain power. It creates it: every auditor, regulator and review board is a new holder of authority.

    Question to carry

    What power does this mechanism create—and who will hold it accountable?

    AI Governance From First Principles

    10

  11. 11 / 15 · Our definition names both sides of the ledger: the gains, and the harm.

    Parts 8–9 · What governance should deliver

    Our definition names both sides of the ledger: 
    the gains, and the harm.

    8

    Share its benefits broadly

    who gains?

    Who gains, who bears the risks and costs, and what would count as a fair outcome?

    9

    Prevent and address harm

    two verbs, two jobs

    What harm could this cause, who will notice it, and what happens for the people already affected?

    Real example

    17.7% → 46.5%

    A US hospital algorithm used past spending to predict need. Correcting that bias would raise the share of Black patients flagged for extra care from 17.7% to 46.5%.

    Obermeyer et al., Science, 2019

    Question to carry, under each part

    11

  12. 12 / 15 · Prevention and remedy both depend on when we act. And acting early means acting with incomplete knowledge.

    Parts 10–11 · Over time

    Prevention and remedy both depend on when we act. And acting early means acting with incomplete knowledge.

    10

    As today’s choices become tomorrow’s conditions

    why timing matters

    What are we committing ourselves to, and how difficult will it become to change course?

    11

    Maintain the capacity to change course

    governance must be able to learn

    How will we discover that this arrangement is failing—and how will we change it?

    The timing problem

    “When change is easy, the need for it cannot be foreseen; when the need for change is apparent, change has become expensive, difficult and time-consuming.”

    David Collingridge, 1980

    Question to carry, under each part

    12

  13. 13 / 15 · A guide to carry into the next conversation

    Put the parts back together

    A guide to carry into the next conversation

    Look atAsk
    Inquiry and practiceWhat must we learn, and how will it change action?
    Authority and participationWho decides, who is affected, and who can challenge?
    IncentivesWhat behavior is rewarded under pressure?
    InstitutionsWho can act, and do they have the capacity?
    InfrastructuresWhat dependencies are created, and what alternatives remain?
    SafeguardsWhat protects people, and what follows when it fails?
    Power and distributionWho gains, who bears costs, and who holds new authority?
    TimingWhat becomes harder to reverse if we wait?
    Learning and revisionWhat evidence triggers review, and who can change course?

    Begin with the real-world outcome a proposal is meant to shape. Then examine the arrangements behind it.

    13

  14. 14 / 15 · What this changes in practice

    Return to the ministry

    What this changes in practice

    Before

    “Is the model accurate enough?”

    One question, about the tool.

    After

    Perhaps we should begin with...

    • A limited pilot, keeping existing clinical capacity
    • Evidence of performance across patient groups
    • Access to records, and rights to export data
    • Independent scrutiny
    • Conditions for expansion or suspension
    • Ability to change tool or provider.

    Illustrative choices, not a universal prescription.

    14

  15. 15 / 15 · When someone asks what AI governance is, begin here

    When someone asks what AI governance is, begin here

    How do societies shape the conditions around AI—and who has the power to influence those choices?

    Govern while choices are still open.

    And build governance that can still change its mind.

    Read the essay, the history and the parts at aigovspec.com · A CENTPOL product · Built by Samuel Abinsinguza

    15

  16. Colophon

    AIGovSpec · aigovspec.com/slides

    AI Governance From First Principles

    Dataset 0.7.0, dated 2026-10-27.

    Research preview: independent expert review is pending.

    Licensed CC BY 4.0. Share and adapt with credit to AIGovSpec, and note your changes.

    Essay and slides by Samuel Abinsinguza · A CentPol product