Pacing
There was a lot of talk in September 2026 about pausing AI development.
Everything that I had read about previous efforts at a pause suggested that it wasn’t going to happen. The principal actors in the game don’t trust each other to respect a pause, so there’s no point.
In the middle of a sidebar discussion on the subject, Muhammad Zane referenced a distinction between a pause and a pace. He shared a piece that Douglas et al (2026) put together, and it does an excellent job at explaining pacing. You should read the entire paper. 66 pages. Easy read.
I have a few thoughts.
There’s no silver bullet
Simple solutions are popular because they’re simple. They’re easy to explain. They’re easy to say.
If it only took me a few seconds to write on the back of a cocktail napkin, I don’t see why I should take more than a few seconds to implement it.
Pause.
Simple to say. Fits on the back of a napkin.
Extremely hard to execute.
Opportunity, Costs
You may have learned the word titration in high school chemistry.
You had some liquid suspended in a container over another with some liquid in it. There was a super sensitive faucet hanging over it. Your job was to let the tiniest amount of liquid, a single drop, form. Then you watched that single drop fall into the beaker.And you waited to see a reaction. Eventually the clear liquid in the beaker would change colour, sometimes suddenly, all at once. Sometimes, if you were lucky, you’d merely watch ribbons form. And then you measured the final amount in the container and got the calculation for how much solution, how many mol, it took to cause a change. That process of very gradual change, letting a drop fall and watching for a reaction is called titration.
Titration is a way of pacing change.
Deploy in a drip. Observe. Drip. Observe.
Remember how long titration took?
It cost a lot of time.
And remember that you had a chemistry teacher that was paying attention. There was an adult in change. You weren’t mixing dangerous combinations.
Take a look around society. Does it seem like there’s an adult in charge?
There’s a lot of risk.
Are we mixing dangerous chemicals or are we mixing something that’ll make glitter precipitate out of the solution?
Who knows for sure?
Now imagine that there’s a competition in that chemistry class. The first student to calculate the tipping point gets an A. Everybody else gets an F. I’m not saying that it’s a good design. Winner-take-all dynamics make people scared. And scared people do scary things.
So, in this scenario, we got all sorts of costs in coordinating the behaviour.
Coordination can be expensive.
Who decides on the paperwork? Who does the paperwork? Who reads the paperwork? Who files the paperwork? Who books the meeting to talk about the paperwork? What decision does the paperwork inform? Who makes the decision? How can the decision be appealed? When does the coordination end? Who decides when it ends? How does it end?
Expensive!
No wonder we spent so much time and effort folding self-interest as a self-regulating function into our institutional designs!
Speaking of self-interest: won’t somebody think of the shareholders!?
All those who expended capital on rival goods: chips, data centres, talent, they expect a return. Some very powerful people were promised a godhead they could enslave before other powerful people could create a godhead and enslave it. A lot of money is riding on that bet. TESCREAL (Transhumanism, Extropianism, Singularitarianism, (modern) Cosmism, Rationalism, Effective Altruism, and Longtermism) outcomes aren’t going to fund themselves!
Any pacing is going to show up in the balance sheet.
And that would be an outrage.
Backcasting The Pacing
This quote from Douglas et al (2026) is lovely:
“In developing a potential intervention it is useful to start with the intended goal and then back out the details, but pragmatically it makes sense to plan around actors using power to pursue their own goals.”
Sure, let’s accept and extend. Let’s assume that paradigms are emergent, that they perhaps precipitate out of a solution of goals. Goals are held by individuals and people. The usual euphemism for power is energy, but sure, let’s talk about power at it is. As of today, at the top level, China and the US, aren’t going to stop competing. There’s some misunderstanding between the two sides. There doesn’t seem to be any reason to believe that they’re going to change their understanding.
The goal of American AI discourse, if it is to be believed, is focused on dominance and supremacy. What intervention could change their mind?
The goal of Chinese AI discourse, if it is to be believed, is focused on industry and development. What intervention could change their mind?
Well it isn’t going to be said by a frontier lab CEO! Even if they were sincere in their desire for a Leviathan to pace the frontier, it isn’t likely in the White House’s interest to unleash one. Even Capital, itself, unilaterally, is unable to call a truce. They can’t just collapse the investment into a single cartel, and together share in those sweet, sweet, Schumpeterian rents. Not without incurring the wrath of anti-trust regulators. The regulators got one good tooth left, it isn’t all gummers with them. Antitrust? Why, cooperative capitalism is all they’re really trying to exercise. No, they’re in a laqueus knot (Or the inconveniently named for this context: Chinese Finger Trap) of their own making.
China, for its part, is playing the situation well. Fine, if you won’t give us the chips, we’ll just distill your models. Tell you what, we’ll make models that are 90% as good as yours, but for 10% of the price? How about that? 90/10. Hey emerging markets, how do you like that? And you Americans can go off and try to enslave God. Have a good time. We don’t have the time or the patience to understand what a TESCREAL is anyway.
What could be offered that would be better?
“The world will pace progress one way or another. Absent better tools, it might do so haphazardly, through improvised reactions overfit to prior expectations, in ways that fail to actually address risks, through institutions that outlast their use. Our hope is that, with proper research, pacing can become progressively more deliberate, targeted, proportionate, decisive, and legitimate.” Douglas et al (2026)
Discussions of pause will give way to pacing. The news cycle will switch over. We’re overdue for a court room drama that grips the country.
That’ll give us some time for some science, and to address some of the real uncertainty about how to build institutions and levers that are constructive. The authors list some 82 potential interventions. That’s a lot to sort out.
I’m not entirely convinced that this entire situation was designed. We find ourselves in some dynamics that we could, maybe, think our way out of it.
It’s worth the effort in reading the paper though.
Check it out.
[1] Douglas, R., Dillon, C., Moore, N., Leech, G., Bonde, M. K., Krishnan, R., Perez, N., Young, N., Slade Byrd, C., Casper, S., Kulveit, J., Duvenaud, D., & Avin, S. (2026). Pacing the frontier: A framework and research agenda. Pacing.tech. https://pacing.tech