What if 'AI safety' is no longer a constructive US-China topic?
What if 'AI safety' is no longer a constructive US-China topic?
Maybe a no-bumper-sticker China approach leaves room for progress
The Biden administration is coming to an end, and I’m beginning to take stock of the administration’s approach to China. This post starts from the observation that AI safety and risk is seemingly unique as an area where the Biden team wants to claim credit for constructive, possibly cooperative work with Chinese counterparts, which seems a bit thin to me, since this might not last. Taking a wider view reveals that this isn’t really the only constructive effort—it’s just the loudest. I conclude by proposing, or asking myself, could we be seeing an intentional strategy of a no-bumper-sticker China diplomacy?
Thanks for reading. All-access subscriptions to Here It Comes are free. Readers who especially wish to support this effort, and have the means, can contribute directly through a paid subscription. I’m grateful for your attention either way. –Graham
National Security Adviser Jake Sullivan is set to travel to China next week, and by the White House’s telling, his agenda is unsurprisingly dominated by areas of friction between the two governments. A press briefer summarized an expected agenda including "key issues in the U.S.-China bilateral relationship and advancing counternarcotics cooperation, military-to-military communication, and AI safety and risk discussions”—areas where the two sides might be expected to produce modest cooperative outcomes—as well as a note that “areas of difference” would be discussed, listing China-Russia ties, the South China Sea, cross-Strait issues, etc.
In this portrayal, it’s really only “AI safety and risk” on the affirmative agenda. The other more affirmative topics listed, counternarcotics and mil-mil ties, are really areas of difference, with the US side hoping Chinese counterparts will do more to crack down on fentanyl precursors or establish channels to avoid real-time miscalculation. Some deliverables might emerge, but not without a context of friction.
In terms of messaging, the Biden administration has itself a bit in knots. Is the Sullivan-Wang channel simply about keeping communication going? Is it about advancing some beneficial outcome? This morning’s senior administration official goes preemptively on the defensive, acknowledging doubt as to the point of having a meeting at all: "We’ve said this before but it bears repeating that US diplomacy and channels of communication do not indicate a change in approach to the PRC. This is an intensely competitive relationship. We are committed to making the investments, strengthening our alliances, and taking the common steps—commonsense steps on tech and national security that we need to take. We are committed to managing this competition responsibly, however, and prevent it from veering into conflict.”
It’s worth dwelling on this, because it is indeed something they’ve said before. In my view, this rhetorical turn addresses an audience that is assumed to understand US-China relations as mostly characterized by rivalry or even enmity. It reassures this audience that the administration sees it this way too, hinting, “yes, we get it, but sorry guys—we’re have to look responsible.”
In an earlier era, Obama administration officials would often lead with the problems (market behavior, South China Sea, cyber-enabled theft, rights abuses) but always mention climate and public health as areas for needed cooperation. This administration has ways of talking about China interactions that are more confident, too, with Secretary of State Blinken memorably saying "we’ll compete with confidence; we’ll cooperate wherever we can; we’ll contest where we must.” But here, this now-standard defensive line loses that balance, seemingly conceding ground to those who criticize any diplomacy as weakness and undercutting the affirmative argument for maintaining lines of communication.
AI anxiety as a unifier, but for how long?
Amidst all this, AI has been the novel area of dialogue the White House is willing to champion. Key people seem to believe there really is the potential for catastrophe that needs US-China discussion, opening a diplomatic space that is new by definition, because no prior topic except perhaps nuclear nonproliferation engaged such heights of uncertainty about both outcomes and magnitude of effects. If AI dialogue is productive, it could be de facto a confidence building measure as both governments are forced to think through what is competitive and what is a common danger across fields from military balance to biosecurity to economic disruption. There’s no guarantee the exchange will be effective, but there’s a big potential upside.
That got me thinking: What if AI is no longer a space for unashamed US-China dialogue with support on both sides? Attitudes very well could change.
For one thing, there are already many people in Washington who think US-China conversations on the development and use of advanced machine learning models is worthless at best and maybe actually damaging. I don’t find these views persuasive, primarily because if you believe novel and emerging forms of AI could pose catastrophic risks or upset geopolitical stability, you’re not going to be able to meet this challenge without understanding the landscape in China, where some of the most powerful non-US systems are being developed. If catastrophic risk is really on the table, you’ll also probably want to see coordinated or parallel action by powerful governments and other actors to mitigate bad outcomes that don’t acknowledge borders. There’s such great uncertainty about what capabilities will emerge, where, and with what implications, it’s just unreasonable to be confident that no US-China cooperation will be needed to avoid very bad outcomes, that somehow “like-minded” governments can handle this on their own. Still, the people who think dialogue is surrender could well gain power in the United States; we have pretty extreme uncertainty tied to the election.
Even assuming a fairly steady perception of the value of US-China AI dialogue, there are other possible developments that could take this relatively hopeful, affirmative area out of the picture. It could be that “AI” in the way it is being portrayed now, essentially systems built around large language models with growing data and computational inputs, will turn out not to pose such deep questions as some people believe. Maybe the universe of training data, or the current paradigm of math applied to it, simply isn’t enough to constitute “artificial general intelligence” (AGI) or to help some wacko develop a diabolical pathogen. Maybe, in other words, the present portrayal of likely AI advances is over-optimistic, and the US-China discussion could become less urgent as this becomes clearer.
Or something could happen in China that changes its government’s willingness to engage. Maybe people feel the Americans are using talks alongside chip controls to advance their goals of containing Chinese development. Maybe Chinese officials decide to cut off conversations they think the US side values in order to demonstrate displeasure with an unrelated event—say some behavior from a new president acknowledging the legitimacy of Taiwan’s government.
One way or another, the main specific area for positive bilateral work the White House was willing to highlight has an uncertain future. Would we be left only with the “areas of difference”?
A central narrative drawing the heat away from cooler heads
In fact, the White House portrayal of Sullivan’s agenda is just a slice of the Biden administration’s China interactions. A meeting at that level, with Sullivan and Wang, is guaranteed more press coverage than some other official contacts, and the messaging about it is attributable straight to the center of the Biden-Harris administration at a time when Harris on a campaign sprint. I don’t think they need to be as defensive, but I don’t think they’re unreasonable to expect attacks.
Perhaps the lesson of the narrow portrayal of what’s ripe for constructive bilateral discussion here is that the highest level is a rough place for constructive bilateral discussion.
There is one topic in US-China diplomacy that—no need to speculate about the emergence of superintelligence—is deeply tied to catastrophic global risk. And the word “climate” was never uttered by the briefer previewing Sullivan’s trip.
This was probably not an oversight. More likely, the Biden administration is trying to keep climate out of the mix when “areas of difference” are dominant. We can see a pattern of maintaining a less heated space (and perhaps one less exposed to the sunlight of publicity) for work on topics like climate change, while portraying the top officials as meeting the challenge where there is more friction.
In November 2023, when Biden and Xi met in Woodside, CA, Biden called out climate as an area “demand[ing] our joint efforts.” But the real action on climate took place in the days before and a few hundred miles away at Sunnylands, where Obama and Xi had their first major summit. There, far from the media circus around the APEC Summit in San Francisco, the two governments' climate envoys announced the "Sunnylands Statement on Enhancing Cooperation to Address the Climate Crisis.” It wasn’t earth-saving, but it was more concrete than the Biden-Xi outcomes, which mostly teed up future meetings.
Biden’s climate adviser John Podesta, as of a FP report by Lili Pike a month ago, was expected to travel to China for further talks on extra-potent greenhouse gasses “in a few weeks,” underlining that there are still separate tracks.
It’s not just climate. Last week’s meeting of the bilateral Financial Working Group between midlevel officials seemed to go off without unwelcome political heat.
A no-bumper-sticker China diplomacy
I’m still processing how to think about the Biden administration’s China policy mix. And it is a huge mix. But what comes into focus here is a US approach to China diplomacy that deemphasizes grand banners and big meetings that had worn out their welcome by the end of the Obama administration.
Even if Hillary Clinton had become president, I am confident the massive omnibus Strategic and Economic Dialogue would have been significantly reconfigured. The big events piled difficult issues into the news cycle with those matters perhaps susceptible to calm, quiet problem-solving. This meant Clinton, then Secretary of State, had more or less the whole bilateral relationship on the table when, for instance, the self-taught lawyer Chen Guangcheng was seeking US Embassy protection after escaping home detention. S&ED also frustrated many officials, who felt they were talking for talk’s sake and wasting time in their particular areas.
In a new diplomatic rhythm where the marquee events are rare, problems and solutions can at least theoretically be handled without as much concern on both sides of how it will make the whole relationship look, or whether the leader will look weak. Could it be that “linkage” in US-China relations, where one side refuses to act in its own interest on one issue because they’re unhappy with how things look on another, is at a nadir? If so, this is a major achievement of the Biden team, and one they probably would like to be recognized for (if it wouldn’t defeat the purpose of keeping some efforts away from the heat).
About Here It Comes
Here it Comes is written by me, Graham Webster, a research scholar and editor-in-chief of the DigiChina Project at the Stanford Program on Geopolitics, Technology, and Governance. It is the successor to my earlier newsletter efforts U.S.–China Week and Transpacifica. Here It Comes is an exploration of the onslaught of interactions between US-China relations, technology, and climate change. The opinions expressed here are my own, and I reserve the right to change my mind.