Effective Altruists Are Stoking Tensions With China

The Biden administration has had an open-door policy for effective altruist think tanks in the White House — who have in turn used their influence to push a hawkish anti-China agenda by casting the development of AI as the new arms race.


Anna Puglisi (R), senior fellow at Georgetown University’s Center for Security and Emerging Technology, testifies during a Senate Energy Committee hearing focusing on the administration’s role in ensuring US competitiveness in AI technologies on September 7, 2023, in Washington, DC. (Drew Angerer / Getty Images)

Effective altruism (EA) has, in just a few years, morphed from a niche academic fad to a thoroughly mainstream worldview. After years flying under the radar, the long financial reach and ideological penetration of effective altruist organizations and individuals in Washington have finally begun to attract media attention. Effective altruists attempt to number-crunch solutions to societal ills by devising ingenious ways, frequently reliant on the latest artificial intelligence (AI) technology, that states and wealthy individuals can maximize the benefits of policy interventions or charity contributions, often with a special focus on very long-term outcomes.

In the United States, most criticism of the movement’s political influence has focused its attention domestically. The argument that AI, when it is used to speculate about the future or make decisions about the present, often bakes in sexist or racist assumptions is well known. While this is of course an important issue, bigger picture concerns, such as the actual influence of EA-dominated approaches to technology-powered decision-making in Washington, have largely fallen out of view. Nowhere has the dangers of EA been more serious, however, than in the school’s escalation of economic hostility toward China.


An Open Door to the White House

As I wrote in Jacobin last May, President Joe Biden’s historic October 2022 semiconductor export bans gave every indication of being carefully directed by the Center for Security and Emerging Technology (CSET), an EA-aligned, Open Philanthropy–funded think tank based at Georgetown University. Senior White House figures ideally situated to dictate the contents of the controls had, in their capacity as CSET members and alumni, spoken and written extensively about their desire to use export regulation to hamper China’s relative progress on AI.

The development of the technology, they argue, should be understood as part of a new arms race between the United States and the People’s Republic of China (PRC). These jingoistic aspirations and strategic proposals were reflected in the text of the October export controls, which spoke of “protect[ing] U.S. national security and foreign policy interests by restricting the PRC’s access to advanced computing” and thwarting China’s “intent to become the world leader in AI by 2030.”

Both the recent presidential executive order on AI and last October’s restrictions on semiconductor exports follow a strategy developed at CSET. Its aim is to plug the gaps left by existing export controls. Implementing this strategy seems, in turn, to have been enabled by an even wider spread of CSET alumni into key staffing positions within the Biden administration, building on the think tank’s established White House presence. This increase in reach, among other developments, points the way to an even more worrying possibility: a revolving door between the White House and an emerging network of EA-funded China hawks.

The October 2022 rules were of considerably wider scope than traditional export controls. Kevin Wolf, a Commerce Department veteran and senior CSET fellow, noted last February that unlike previous traditional controls, they are “directed at the entire supply chain” of AI-suitable chips. For example, the 1996 Wassenaar Arrangement, a major US-participating export controls regulatory system intended to limit trade of “dual use” — that is, military and civilian — products like AI hardware, pointedly excludes controls on “[g]eneral commercially applied materials or components” like semiconductors or machinery involved in their manufacture.

These earlier controls still left avenues for Chinese AI developers to make use of American microchips. Most significantly, they placed no limits on Chinese use of US chips through cloud computing and infrastructure as a service (IaaS), by which foreign researchers can, essentially, rent the use of US-owned compute. As such long-distance computing involves no actual export of American hardware; it falls outside the scope of the 2022 rules. This failure to fully “decouple” US supply chains from Chinese AI applications has had real consequences: in March 2023, the Financial Times reported that companies in China were relying on cloud compute as “a good, albeit more expensive, alternative” to US imports.

Not to be deterred, CSET last spring began to develop strategies to close this loophole. One pair of documents jointly authored with the Center for a New American Security (CNAS), “Controlling Access to Compute via the Cloud: Options for U.S. Policymakers,” describes how the Commerce Department’s regulatory apparatus might be leveraged to make US export controls more far-reaching. Crucial to this strategy is the recently expanded authority of the Commerce Department over the activities of “US persons,” regardless of location.

Under this fresh congressional mandate, the department’s Bureau of Industry and Security (BIS) is entitled to restrict any activities by a US person (a broad category extending to citizens, permanent residents, and American companies) found to be in “support” of “foreign military, security, or intelligence services.” The CSET-CNAS authors reason that these regulatory powers, which are abnormally strong among peer nations, could be deployed to sanction cloud computing providers offering services to any Chinese end users plausibly connected with intelligence or military applications.

The limited scope of such sanctions should not mislead us. The frank wish of their authors is not merely to prevent American companies from collaborating with a potentially hostile military and security state, but to block as much Chinese access to semiconductors for AI development as politically possible. This high aspiration is discussed in the first paper as “Goal 1.” Restriction to military and intelligence support is, the document makes clear, a stepping stone and a compromise in light of the political and economic untenability of a full block.

Indeed, in a revealing aside, the authors remark that the “BIS currently has no means to further restrict” access in China to data centers not connected to any US persons [emphasis added], seemingly leaving open the possibility that US export regulations might one day permit the United States to block Chinese use of fully indigenous Chinese AI infrastructure. Preventing American “support” for Chinese military AI is a means to an end: crushing China’s AI sector and preventing it from competing with America’s.

Still, the authors conclude, adopting even this limited strategy would be risky. Not only might it harm US cloud service providers commercially, without requirements for these providers to collect information on end users the sanctions would be useless. To this end, they suggest imposing new “know your customer” requirements on American IaaS companies to monitor for potential Chinese military or surveillance connections. “Future work from CSET and CNAS,” they promise, “will go deeper on this issue.”

With several well-placed CSET and CNAS alumni in the White House, this future work seems to be well underway. Last October, Biden issued an executive order that stipulated that the secretary of commerce must demand reports from US IaaS providers on any foreign end users potentially making use of their services for AI training. The order also puts in motion “know your customer” rules for IaaS providers, requiring them to extensively verify the identity of any foreign clients. While nominally to address “malicious cyber-enabled activities,” these provisions could not be better suited to paving the way for such prohibitions on Chinese use of American cloud computing.

Nor is this the only White House policy last October to have seemingly taken inspiration from CSET and CNAS’s musings on patching up leaks in the semiconductor export restrictions. Last year’s two new updates to the October 2022 export rules place significant limits on “US persons” even tangentially supporting Chinese military and intelligence — not by limiting Chinese access to IaaS, but by requiring licensure for any US person in any way “supporting” production or development of semiconductors for military or intelligence use in China. Without targeting cloud services specifically, these restrictions are otherwise taken right from the new CSET-CNAS playbook.


Entryism for Altruists

The close resemblance here of CSET’s suggestions for eventual White House policy should come as no surprise. Listed as one of the coauthors of the aforementioned policy papers was CSET’s “Commerce Department fellow” Emily Weinstein, who became a senior advisor at the BIS by the time the policies would have been drafted. She is joined in the Commerce Department by fellow CSET alumni Will Hunt, senior advisor at the Office of Export Administration, and Saif Khan, now senior advisor to the secretary on critical and emerging technologies. Khan was among the most ardent proponents of semiconductor export restrictions and was the National Security Council’s advisor on technology at the time of their announcement.

Khan was succeeded in this position by another former CSET fellow, Ben Buchanan. Buchanan had, like others at CSET, written extensively about his support for semiconductor export regulations before their eventual adoption by the Biden administration. In an August 2020 white paper, he stressed that such controls will be a useful policy tool in slowing the development of Chinese AI since compute is the most important leg of the “AI triad” (compute, algorithms, data), and he made a similar, albeit more broadly pitched, argument in the same vein that month in Foreign Affairs. By the time the updates to export controls and executive order came into play, he had graduated to the role of special advisor for AI.

This proliferation of EA-backed foreign policy staffers in the White House is not just background context to the policies. The executive order on AI contains extensive provisions for installing an ever-increasing cadre of “AI experts” in advisory positions, opening up a formal staffing pipeline for like-minded effective altruists.

For example, §4.3(a)(v) directs the Department of Homeland Security to “establish an Artificial Intelligence Safety and Security Board as an advisory committee” to be filled with “AI experts from the private sector, academia, and government” — the private sector, academia, and government already being sites of extensive, well-funded, and energetic effective altruist organizing around AI. Further directives on cultivating AI “talent” abound throughout the document, perhaps most notably in §10.2(b), which establishes a task force specifically for this purpose. The policy thus seems to be as much an opportunity for EA entryism as it is an effect.

Acquiring positions within the White House is not the only way that the EA movement has succeeded in entangling itself with a bellicose foreign policy establishment. CSET’s collaboration with CNAS, a major foreign policy think tank with an extremely aggressive outlook and deep ties with recent presidential administrations, goes well beyond the odd policy paper.

Since 2021, CSET has donated hundreds of thousands of dollars to CNAS, and has given Open Philanthropy over $500,000 a year. (The organization received — and subsequently returned — a smaller but still hefty donation from the infamous FTX, the trading firm founded by the now-incarcerated Sam Bankman-Fried.) Munificent EA donors are not the only people dictating aspects of CNAS’s policy recommendations, joining their ranks are arms dealers like Raytheon and foreign governments like Taiwan. These financial ties, combined with their emerging policy alignment, suggest a merging and mutual influence of EA organizations concerned about AI on the one hand and more traditional hawkish influences in Washington on the other.

The ascendancy of this movement’s anti-China agenda is not a foregone conclusion. In particular, any possible efforts to block Chinese access to American IaaS are in an as of yet nascent stage, and the relevant policies to date have been mainly executive rather than legislative, leaving them comparatively reversible. Even within EA, support for this arms race aggression has not been unanimous.

Katja Grace, for instance, a lead researcher at the Machine Intelligence Research Institute, warns against those who compare the “AI situation to a classic arms race” between “researchers in the West” and “more reckless lab[s] . . . in China.” “The best individual action,” she adds, “could be to move slowly and cautiously. And collectively, we shouldn’t let people throw the world away in a perverse race to destruction.”

But without successful internal or external resistance, effective altruism as a major player in US politics continues to convert tech infrastructure into a new cold war front, with all the global dangers that conversion entails.


This post was originally published on Jacobin.