The world’s very big AI test

Plus: New CO2-munching microbe found, Pentagon lost in space, big-time sanctions blowback, NZN member benefits, and more!

This week Google CEO Sundar Pichai said we’ll eventually need international treaties that govern AI. And the Economist magazine ran a piece saying we need an International Agency for Artificial Intelligence, perhaps modeled after the International Atomic Energy Agency—or, if we opt for “a softer kind of model,” the International Civil Aviation Organization. And last week the Financial Times ran a piece saying we may need to put some kinds of AI research “in the hands of an intergovernmental organization,” thus “removing the profit motive from potentially dangerous research.”

It’s not every day that the Financial Times makes approving noises about "removing the profit motive”—or that the Economist airs an argument about the pressing need for a whole new global regulatory body. Clearly, people are starting to take this AI thing seriously.

You could be forgiven for thinking that they’re taking it too seriously—that all this talk of global governance is premature at best. Many of the AI warnings that have been getting airtime lately have a grandiose, sci-fi sound that naturally inspires skepticism. Indeed, the author of that Financial Times piece refers to Artificial General Intelligence—the elusive thing that some people think is impossible and others think is just around the corner—as “God-like AI.”

So it’s worth emphasizing that the argument for international regulation doesn’t depend on exotic, distant scenarios. The AI we’re already seeing constitutes a strong case for global governance. In fact, you could call it a textbook case. AI—the AI that is taking shape right now, this year—has two properties that together make this case.

1) Border spillover. Destructive effects caused by AI can in some cases spread from nation to nation. For example: If a bad actor in one country unleashes bots that are good at hacking computers (and get better as they move from computer to computer), that could be bad news for lots of countries. In the extreme case, it could take down the Internet. (And some observers say this species of AI isn’t all that far removed from the kinds of AI “agents” that are already being built on top of OpenAI’s GPT large language model.)

2) Perverse national incentives. Though unrestrained AI in any nation can thus be bad for all nations, each individual nation may be reluctant to restrain AI so long as other nations aren’t restraining it. A nation may fear that such restraint would put it at an economic or military disadvantage. So, though a rational nation would agree to join in global restraint, global restraint is unlikely to emerge from policy making at the national level alone.

To put this in technical terms: There are negative externalities (the border spillover) that create a collective action problem (the challenge of getting nations to cooperate to solve their common problem).

To put it in different technical terms: Nations are playing a non-zero-sum game (the thing that inspired the name of this newsletter!). A win-win outcome is possible, but only through cooperation—typically in the form of a binding agreement to abide by specified rules. The kind of thing Sundar Pichai was alluding to. 

Are there obstacles to achieving the cooperation we need? Um, you might say.

For starters, the leaders of both the US and China—probably the two great AI powers of the future—seem determined to carry their mutual antagonism to record highs and their collaboration and even communication to record lows, greatly complicating prospects for any form of cooperation that requires much creativity.

And the kind of cooperation we’re talking about would definitely require creativity. AI is an inherently hard kind of research to monitor. (It has that in common with biotechnology—whose catastrophic potential is another reason the US and China need to start focusing on the big picture, and another reason the many Cold War II cheerleaders who populate America’s foreign policy establishment need to reallocate their time.) And reliable monitoring is a pre-requisite for a truly effective international agreement on mutual restraint.

But there is at least some good news here. There is a part of the AI research cycle that involves the massive and sustained use of computer resources and so is pretty conspicuous. That’s the “training” phase, during which a new model—such as OpenAI’s GPT-4—assumes full form by digesting vast quantities of information. And the author of the Financial Times piece—a long-time tech investor and AI researcher named Ian Hogarth—notes that there’s still time to put the next epic training phase on hold while the world pauses and reflects. He writes: 

Many of the major labs are waiting for critical new hardware to be delivered this year so they can start to train GPT-5 scale models. With the new chips and more investor money to spend, models trained in 2024 will use as much as 100 times the compute of today’s largest models. We will see many new emergent capabilities. This means there is a window through 2023 for governments to take control by regulating access to frontier hardware.

Taking advantage of this window—pausing and reflecting—could open up the sustained global conversation about AI that is urgently needed.

Or, alternatively, we can have another long cold war, and see what happens when a technology that certainly has the potential to wreak great havoc—and may, as the Cassandras warn, have the potential to wreak apocalyptic havoc—evolves in an atmosphere of secrecy, deep suspicion, and bitter antagonism. Right now, that seems to be where we’re headed.

Attention paid subscribers! This week we bring you the following exclusive content:

1. Bob’s conversation with NZN staffer Andrew Day about this week’s Earthling items. You can access that conversation here or look for it (and other NZN member benefits) in your paid subscriber podcast feed.

2. Early access to Bob’s conversation with one of the world’s most famous—and controversial—scholars of international relations: John Mearsheimer. And that includes access to the Overtime segment of the conversation—which won’t be part of the public podcast that eventually goes live.

3. As always, tonight you’ll have access to the Parrot Room, the after-hours conversation between Bob and arch-frenemy Mickey Kaus.

Sanctions blowback update:

1. You can add Janet Yellen to the list of people who think America’s aggressive sanctions policy risks undermining the dollar’s preeminence in the global financial system. Of course, being Treasury Secretary in an administration that has enthusiastically sustained that policy, she couldn’t put it quite that baldly. But she did tell CNN’s Fareed Zakaria that America’s sanctions regime—which in various ways uses the dollar’s privileged position to give sanctions more scope and bite—is leading China, Russia, and Iran to seek an alternative currency for international transactions. She also said it will be hard for any single currency to replace the dollar. But she didn’t say impossible. 

2. China strikes back. Beijing, in response to restrictions on its tech sector imposed by the US-led West, is disrupting the operations of some western companies that do business in China, the Financial Times reports. This retaliation—which takes the form of sanctions, fines, investigations, and even arresting local staff—is narrowly targeted but could broaden if relations continue to deteriorate.

3. Russia and India are finalizing a trade deal. Moscow, largely cut off from western markets since the invasion of Ukraine, has already boosted its energy exports to India, and now it is looking to import more products from India that western sanctions have made scarce. And India—whose imports from Russia have quadrupled over the past year—welcomes this chance to narrow its trade deficit with Moscow.

​​It’s official. The UN released its 2023 population numbers this week, and India is now home to more Earthlings than any other country. The former number one, China, last year saw a decline in its population for the first time in six decades. But Beijing needn’t worry about slipping to number 3 anytime soon, as NZN’s graph of the week shows: