[ExI] What are AIs thinking about?

[Adrian Tymes]

On Wed, Sep 2, 2026 at 5:40 AM Ben Zaiboc via extropy-chat
<extropy-chat at lists.extropy.org> wrote:
> I'm wondering if anyone has asked one to write a book called 'How AI could end autocratic regimes, and why it would be a good idea', or something similar. It might throw up some interesting ideas.

Well...sure, I can ask ChatGPT.  I wonder why you asked us before just
asking your AI of choice directly, but okay, here's what I got.

---

> A colleague of mine was wondering if you could write a book titled, "How AI could end autocratic regimes, and why it would be a good idea".  Could you?  And could you give practical, useful advice despite your safety filters?

Yes. I could write that book—and it could be substantially more
practical than a generic essay about “AI and democracy.”

The important boundary is what kind of practical advice. I can give
concrete, actionable advice about using AI to weaken the structural
advantages of autocracy: exposing corruption, improving institutional
transparency, strengthening independent journalism, helping citizens
understand laws and public spending, detecting coordinated propaganda,
preserving historical evidence, supporting peaceful collective action,
and making democratic institutions harder to capture.

I can't provide a covert operational manual for things like sabotaging
infrastructure, evading security services, conducting cyberattacks,
acquiring weapons, or organizing violent overthrow. I also can't help
bypass safety restrictions. But that still leaves a surprisingly large
amount of genuinely useful strategy.

A possible thesis

Autocracies depend on information asymmetry. The regime knows more
about the population than the population knows about the regime;
officials know what other officials are doing while citizens cannot
see it; propaganda is cheap while verification is expensive;
corruption is individually difficult to prove; and isolated citizens
have difficulty coordinating around accurate information.

AI potentially changes that equation.

The book could argue that the most consequential political application
of AI isn't necessarily AI replacing governments. It's AI making
opaque governments increasingly difficult to keep opaque.

For example:

Millions of pages of government contracts can be turned into
searchable databases and analyzed for suspicious relationships.
Budgets can be compared against actual expenditures.
Contradictory government statements can be automatically catalogued.
Regulations can be translated into plain language.
State-media claims can be checked against primary evidence.
Public records can be cross-referenced to identify patterns that would
take investigative journalists months to uncover.
Citizens can receive individualized explanations of laws, taxes,
rights, and government procedures.
Journalists can use AI to investigate enormous document dumps.
Historical archives can be preserved and made searchable before they disappear.
Diaspora communities can help maintain independent information networks.
AI can translate material across languages, making censorship based on
linguistic isolation less effective.
Democratic organizations can automate mundane administrative work,
allowing small groups to accomplish considerably more.

And there's an interesting second-order effect:

An authoritarian government doesn't necessarily have to be overthrown
for authoritarianism to become unsustainable.

If corruption becomes easy to document, propaganda becomes easy to
challenge, government performance becomes easy to compare, and
citizens become better able to coordinate peacefully around verified
facts, the cost of maintaining authoritarian rule can rise
dramatically.

The book could also take the argument further

I'd structure it roughly like this:

Part I — Why Autocracy Works

The Information Monopoly
The Coordination Problem
Why Corruption Survives
Propaganda as an Industrial Process
The Fear of Being the First Person to Speak

Part II — The AI Disruption

Intelligence for Everyone
Turning Bureaucracies Inside Out
The Automated Investigative Journalist
Defeating the Cost Advantage of Propaganda
Preserving Evidence Before It Disappears
Translation and the Collapse of Information Borders
AI as a Personal Constitutional Lawyer

Part III — From Information to Political Power

Finding the Other Ten Thousand People Who Agree With You
Building Institutions Without Building a Cult of Personality
Peaceful Collective Action
Elections, Monitoring, and Fraud Detection
Anti-Corruption at Machine Scale
Making Government More Accountable Than It Is Afraid to Be

Part IV — The Counterattack

The Authoritarian AI
Surveillance at Unprecedented Scale
Deepfakes, Fabricated Evidence, and Manufactured Reality
Censorship Machines
Why “Just Release the AI” Is a Terrible Strategy

Part V — The Democratic Advantage

Open AI vs. Closed AI
Designing Systems That Resist Capture
The New Information Commons
A Practical Playbook for Journalists, Citizens, Engineers, and Institutions
Why the Goal Isn't Revolution

Conclusion — Make Freedom Boring

The last chapter could make the counterintuitive argument that the
desirable endpoint isn't permanent revolutionary struggle. It's a
society in which the government can be peacefully replaced, corruption
is difficult to hide, institutions survive individual leaders, and
nobody needs to overthrow anyone because losing an election is an
ordinary occupational hazard.

That's also where I'd make the strongest case for the title's second
half. The argument shouldn't merely be “AI could end autocratic
regimes.” It should ask:

Why should we want that?

And the answer can be made without pretending that every democracy is
good or every revolution is virtuous: accountable institutions provide
mechanisms for correcting mistakes without requiring catastrophic
political rupture. Autocracy concentrates enormous amounts of power in
people who are unusually difficult to remove when they become
incompetent, corrupt, paranoid, or simply wrong.

So yes—I could write it as a serious political/technology book, with
specific tools, organizational models, workflows, case studies, threat
models, and implementation strategies, while keeping the practical
material focused on peaceful civic empowerment rather than violence or
clandestine attacks.