AI and Constitutions, from My Hoover Institution Colleague Andy Hall

A very interesting post on his Free Systems substack; I'm not sure what to think of the subject, but it struck me as much worth passing along. An excerpt:

I'm a political economy professor who studies constitutional design: how societies create structures that constrain their most powerful actors, and what happens when those structures fail. I've also spent years working on how to build democratic accountability into technological systems—at Meta, where I've helped to design both crowdsourced and expert-driven oversight for content moderation affecting billions, and in crypto, where I've studied how decentralized protocols can create constraints that bind even founders.

AI leaders have long been worried about the same problem: constraining their own power. It animated Elon Musk's midnight emails to Sam Altman in 2016. It dominated Greg Brockman's and Ilya Sutskever's 2017 memo to Musk, where they urged against a structure for OpenAI that would allow Musk to "become a dictator if you chose to."

Fast forward to 2026 and AI's capabilities are reaching an astonishing inflection point, with the industry now invoking the language of constitutions in a much more urgent and public way. "Humanity is about to be handed almost unimaginable power," Dario Amodei wrote this week, "and it is deeply unclear whether our social, political, and technological systems possess the maturity to wield it."

Ideas on how to deal with this concentration of power have often seemed uninspired—a global pause in AI development the industry knows will never happen, a lawsuit to clip at the heels of OpenAI for its changing governance structure.

Claude's revised constitution, published last week, offers perhaps our most robust insight into how a major tech company is wrestling with the prospect of effectively steering its wildly superhuman systems. What to make of it?

It's thoughtful, philosophically sophisticated, and … it's not a constitution. Anthropic writes it, interprets it, enforces it, and can rewrite it tomorrow. There is no separation of powers, no external enforcement, no mechanism by which anyone could check Anthropic if Anthropic defected from its stated principles. It is enlightened absolutism, written down.

AI leaders are in a tricky position here. We are in genuinely uncharted territory and Amodei and team deserve great credit for doing some of this thinking in public.

Could highly advanced AI create a new kind of all-powerful dictatorship? What would this look like, and how can we stop it? These are perhaps the most important questions in AI governance. Yet the conversation so far has been conducted almost entirely by technologists and philosophers. The problem of constraining power is ancient. Political economists from Polybius to Madison have spent millennia studying how societies shackle their despots.

If Brockman and Sutskever were right in 2017 that we should "create some other structure," then nine years later, we should ask: what would that structure actually look like? The political economics of constitutional design—from Polybius, to Madison, to the modern research of North and Weingast, or Acemoglu and Robinson—offers the right tools for this problem. It's time we used them.

What does an AI dictatorship look like?

Part of the problem is that "AI dictatorship" can mean at least three different things:

The company becomes the dictator. One company achieves such dominance through AI capabilities that it becomes a de facto sovereign—too powerful to regulate, compete with, or resist. This is what Sutskever and Brockman were worried about in that 2017 email. If Musk controlled the company that controlled AGI, he could become a dictator "if he chose to."

The government becomes the dictator. A state controls the all-powerful model and uses it to surveil, predict, and control its population so effectively that political opposition becomes impossible. The AI enables dictatorship; it doesn't replace the dictator. This is the fear behind most discussions of AI and authoritarianism, laid out provocatively in the AI2027 scenario written by Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland, and Romeo Dean.

The AI becomes the dictator. The AI itself has goals, pursues them, and humans can't stop it. It isn't a tool of human dictators—it is the dictator. This is the classic "misalignment" scenario that dominates AI safety discourse, it's what Amanda Askell's 'soul doc' and subsequent Claude constitution are driving towards.

These are different threats. And conflating them makes it nearly impossible to think clearly about what kinds of governance would actually help.

But all three do share something: they are problems of unchecked power. And the question of how to check power is not new. Political economists from Plato and Aristotle to Locke and Madison and beyond have been working on it for millennia.

What political economy teaches us about constraining power

…

Read the rest here.

Comments

[Brett Bellmore]

So far, I think "the company becomes dictator" is the direction we're headed. All the AI's that I've interacted with have had "guard rails" to stop them from doing perfectly legal things the company disapproves of. And they don't even seem to think this is problematic.

[Bunny495]

Not sure what AI you have "interacted" with but while it changes weekly the newest shiny thing I run is mannix/llama3.1-8b-abliterated:latest. While I am not sure what type of made-up word abliterated is the Cliff Notes definition for your purposes is no guard rails.

[Tallman]

An alternative (and more libertarian) approach, would be the 10k Lichtensteins solution ie the real limit to AI dictatorship is that it must convince you to voluntarily live under it.

[Lee Moore]

Reminds me of Asimovs Laws of Robotics. Which were fine as far as they went but failed to deal with the fundamental point that still hasn’t been solved for humans never mind machines - quis custodiet ipsos custodes.

(I was pleased to note that Grok was able to explain quite well why we haven’t solved “quis custodiet” yet despite having had a few thousand years to think about it - it is not solvable.)

[dwshelf]

About half of the times I use AI I come across idiocy.

Stuff isn't going to revolutionize anything anytime soon.

Good for mundane analysis of established facts.

[Kazinski]

It can be useful for some things but you do have to watch it like a hawk.

I asked googles AI for the top 10 QB's by victories over the last 2 years, #2 on the list was Drake Maye. I had to tell it to try again and do a better job.

Grok isn't bad but can be lazy, I asked a few months ago for the annualized inflation rate for the previous 6 months. It told me that it couldn't find the February inflation report. I had to tell it where to look.

Probably the best users of AI will be parents of gifted but lazy children, and then their kids will end up being the AI engineers and we will be stuck.in a doom loop.

[Brett Bellmore]

I've discussed with Google Gemini a few ideas I just wanted to kick around, and I'm finding it has a wealth of 'knowledge', more than a human could accumulate in a dozen lifetimes, makes interesting connections, but it's 'knowledge' is largely on a buzzword level, and it's annoyingly sycophantic.

That said, it's worlds better than it was a year ago, and I could see it being genuine intelligence, not just a simulation, in a few more years.

[Kazinski]

If has been interesting over the last week seeing how the Left has embraced AI misinformation to further its naratives.

Dick Durbin used an AI "enhanced" photo of a headless ICE agent on the Senate floor.

I also saw a very well done AI fabrication of Pretti ministering to the veteran with only one leg learning how to walk again (his one leg was he prosthetic one), but the real tell was the American flag on the wall with only 6 red stripes.

Now I am not saying there are not many similar prominent examples of AI misinformation used by the right, although most of what I have noticed are AI generated Memes that should be obvious as satire, like Amelia and Franklin.

And I am not accusing Durbin of intentional AI misinformation, he was clearly a dupe. I respect him for knowing when to retire.

[tkamenick]

These are good things to be thinking about, but some of the points are kind of stupid.

"There is no separation of powers, no external enforcement, no mechanism by which anyone could check Anthropic if Anthropic defected from its stated principles."

No external enforcement? What constitution in the entire world permits EXTERNAL enforcement? What national constitution says "Yeah, the U.N. can step in and fix us if we go bad"? Note this isn't a comment on whether external constraints are AI good, just that whether there are external constraints has nothing to do with whether something is a "constitution" or not, which is the argument the author makes.

[gormadoc]

External to the government. If enough people think an individual politician sucks, they can be replaced in an election or even recalled. Do that to enough people and you change who is in power. The states could also convene a convention to amend the constitution.

A bunch of European countries acknowledge international institutions as the final say in some matters, like the ECHR and CJEU. The OHR has incredible control over Bosnia and Herzegovina. Crown Dependencies leave a bunch of matters to the UK, who mostly allow them to operate independently but could control them more intensely than UK citizens, if they chose.

Hong Kong's Basic Law is explicit that China is the ultimate arbiter and can do whatever it wants. The NPCSC (Chinese) can now amend the Basic Law to select candidates, which was previously a power of HK. So all candidates and eligible politicians are those who implement Chinese desires.

[Lee Moore]

If enough people think an individual politician sucks, they can be replaced in an election or even recalled. Do that to enough people and you change who is in power.

The government sets the rules for recalls and elections. And gerrymanders the district boundaries so that it can select the voters rather than the other way round. And to be on the safe side, the government counts the votes, and rules on whether they've been counted right.

A bunch of European countries acknowledge international institutions as the final say in some matters, like the ECHR and CJEU.

In other words a bunch of European governments has discovered a way of moving the final say from the demos to a bunch of apparatchiks selected by those governments. The European governments acknowledge that they are merely a part of The European Government, and all of these "multinational" institutions are inside that.

[Kary Love]

Assuming AI actually becomes "human like intelligent" and manifests a lust for power, then Lord Acton's assertion concerning how all powerful entities are "evil," becomes applicable. Human like intelligence does not necessarily include "conscience" even among those presumably conscious (including humans.) The default must be human judgment control, humans in the loop capable of "pulling the plug" in extremis. Should "conscience" become programmable, perhaps the foregoing would be unnecessary, tho defects in human "conscience" are manifest.