059. Dominic Cummings: Whitehall's war against the future

Tom Ough (0:07)

What's going on, Midshipman? They have no jurisdiction over orbital spacepubs.

Calum Drysdale (0:11)

I've got a stormtrooper on the line now, sir. He says the Online Safety Act now encompasses the vast infinitude of space. It's over.

Tom Ough (0:18)

There's nothing for it, Midshipman. It's time to deploy our deadliest weapon.

Calum Drysdale (0:22)

Sir, you can't mean the Cummings interview.

Tom Ough (0:26)

I mean it, Mr. Drysdale. Initiate the broadcast.

Calum Drysdale (0:32)

Who now has anything to say about the deindustrialization of this country?

Tom Ough (1:29)

Tom, a key fact of the world, as we see it right now, is that the rules-based international order is crumbling. So there are many examples of this.

Tariffs, the Iran War, perhaps even the furor over the Chego silence. What do you see in this respect?

Dominic Cummings (1:49)

Yeah, I mean, if you go back to the institutions built in the rubble of 1945, and as the Cold War was beginning, you have obviously UN, NATO, WTO, EEC as it was, EU, et cetera, et cetera, WHO, IMF, all these sorts of things. And they all seem really kind of played out and exhausted, right? They're in many ways just clearly incompetent. Their legitimacy, trust in them is imploded.

COVID obviously undermined a lot of things.

Brexit happens partly because the pro-EEC slash EU forces lost the argument over a couple of decades about that. So yeah, you have what you often see in history, right? You see this sort of pattern in European history of every kind of like 50, 75 years or so. You have the international order of its time. The institutions and the ideas kind of gradually drift out of alignment with reality, and then you have crisis. So you can, I think it's interesting to think about the mid-19th century. So if you go back to 1815, you have the end of one huge cycle, the French Revolution, the Napoleonic Wars. 1815, you have the Treaty of Viena, and a set of new ideas about how the world will be organized, clear kind of great power concert, Metnick, Britain has a role in the system. You fast forward to the 1840s and you can see people all over Europe start to write letters, Metnick and Bismarck and Gerlach and all these people were saying, the old system's running out of rope, the young people are going crazy, there's all these new technologies, then you've got the telegraph and railway coming through, et cetera, et cetera. You can sense that things are going to come into a crunch and then you have the 1840 revolutions, regime change and then a new world arises out of it.

And same thing essentially is happening now, right? It's a similar sort of thing. You've got 1815, 1945, set of new ideas and institutions, that generation has died out, gone. And the ideas and institutions that they left behind clearly are clearly failing and something new is being born. But for the moment, unclear exactly what.

Tom Ough (4:36)

I'm going to quote something which you have put many times at the end of your blog posts, which is, technical changes, such as genetic engineering and machine intelligence, are bringing revolution. And then there's a Bismarck quote, it would be better to undertake it than undergo it. So what I want to know is, what is the world that you think we could be heading into maybe 30, 40 years from now? Are we talking a billion Chinese with 150 Ikea? Are we talking American superintendents organizing a vast space economy? Well, what actually is at stake?

Dominic Cummings (5:09)

Even the people at the actual frontier in the AI labs will tell you that it's very hard to predict how the models are going to evolve in some ways, right? The various of those guys put straight lines on graphs a decade or so ago. And in all sorts of amazingly interesting ways, the straight lines on graphs have turned out to be true.

Much derided by most of them.

Tom Ough (5:35)

Ash and Brenner's hockey stick.

Dominic Cummings (5:37)

At the time, professors here, there and everywhere laughed at these guys, at the idea that scaling laws would be true, but in all sorts of ways, they have been. But within that, they're very spiky, right? And as Dario and Sam and Demis will all say and say publicly, each time you do it, they spike up on that, but they're weirdly rubbish and blah. So there's no theory of it at the moment. It's just an empirical thing. And you find out with, as I've got to add. But I guess my, so one of my observations on it would be, I've watched this since 2020, and it is kind of amazing that the most optimistic, if that's the right word, or in any event, the people who predicted the most success for machine learning and the most dramatic changes and were most optimistic about the straight lines on graphs and the three labs, who were completely divided by the entire mainstream, have actually turned out to be the most accurate predictors of the future. That's extremely striking, extremely striking. And I think that at least for the next two years, so they release these models, but they have line of sight internally to at least, I would guess, roughly speaking, to like a year or two ahead of what they can see privately before things are released. And as far as I can see, they're very confident that the straight lines on graphs will continue for at least another couple of years.