The Confusion Around AI Diffusion

The core issues being missed around the new US effort to control the development of advanced AI globally.

Dean Ball

Hyperdimensional

has written an excellent summary of the new AI Diffusion rule on his substack this week.

Unstable Diffusion

My response to this great summary of the rule, text of rule available here in all its glory, was to agree in most places, except for the idea that all of this is somehow necessary to prevent China from leading in AI, misusing Ai, or using AI against "us". This of course is the whole rationale for all the rules targeting China and Ai over the past three years, including three massive new export control packages (Dean does a great job describing the rule making process and its pitfalls), even though there is no evidence that AI, for example, would be decisive in any future conflict, which will still be dominated by old fashioned firepower and the ability to bring it to bear en masse and accurately. Already software and other AI algorithms are doing this (see for example Israel’s use of facial recognition software in Gaza here), and generative AI in particular, which is the target of these rules, seems unlikely to lead to some decisive military advantage. The vast majority of end use cases of genAI are currently being focused on "civilian deployments, mundane things like enhancing productivity, aiding coders, etc. and generative and other types of AI in 2025 will likely contribute significantly to progress in science and healthcare. The rule makes nods to the importance of AI for a range of societal goods, but then focuses heavily on the potential and mainly speculative uses of AI for military gains. US officials on a public call this week to explain the new rule cited “autonomous jets” and cyber operations as areas of particular concern. Neither will rely on generative AI or the most advanced AI models that are the target of the rule. These contradictions, similar to claims that advanced semiconductors are widely used in military applications (they are not, virtually all the most advanced semiconductors produced at foundries like TSMC go to consumer products, smartphones, gaming cards, datacenter servers, for, you guessed, training large language models being developed by a small number of commercials firms at the cutting edge like OpenAI, Anthropic, Meta, Microsoft, Tesla, Cohere, and others.

In addition, literally no one in the industry believes generative AI alone will lead to something we can call AGI, let along ASI. Already research in other areas of AI, for example by the excellent Francis Chollet’s Ndea, is pointing towards much different approaches, such as program synthesis, for achieving something like AGI, and these and other approaches may not rely on massive GPU clusters to achieve progress. Chinese companies include DeepSeek and Minimax are demonstrating the ability to achieve advanced model performance without the need for access to the types of cluster targeted by the new rule.

In addition, and I have written a lot about this (see below), Dean failed to mention the critical Achilles Heel of the whole US approach—he alluded to it when noting that much of the hardware the US is seeking to control is actually manufactured outside the US-- but did not tackle head on: the entire hardware basis for AI development is located 100 miles from the Chinese coast in a territory Beijing considers to be a part of China, namely Taiwan. As I have written, it is naive in the extreme to believe that the US and allies can run ahead towards AGI/ASI, with the explicit goal of "winning" the AI "arms race" over China and containing China's ability to develop advanced AI for economic growth and all the good stuff, all while the Taiwan dependence is still the case, which it will be for the next decade and likely beyond.

The dangers here are stark and growing and no one seems to want to acknowledge this, least of all apparently the authors and drivers of these rules, who do not appear to understand the global technology industry or the risks inherent in this approach. In addition, the thrust of these rules will work to exclude China from participating in much needed global efforts to develop safety and risk frameworks around AI model development, yet another massive risk from this approach. I address some of his in a Wired piece this week with Alvin Graylin: https://www.wired.com/story/why-beating-china-in-ai-brings-its-own-risks/.

I wrote about the rule here last month in the Wire China, after having access to an early draft version that was available within US industry circles and which some commentaries also cited (see Oracle VP Ken Glueck). Glueck’s commentaries (see below) and were particularly critical of the thinking that has gone into the rule about GPUs, supply chains, distribution channels, data center buildouts, and other issues critical to understanding how US technology diffuses around the world across the AI stack.

On Taiwan, responding in part to the tone of discussions around the Taiwan issue among leading Taiwan analysts at a closed door Washington conference, I highlighted the dangers of the nexus of the US, China, and Taiwan, and US efforts to control AI here, also in Wire China.

This issue will be front and center as the new Trump administration takes office and likely reviews all the rules and executive orders handed down over the last 60 days, as most administrations do. The new administration will see a complex mix of agendas at play, in particular around technology in general, AI in particular, and the US, China, and Taiwan most critically. While there will much alignment between some China critics, technologists, and institutional heavyweights such as the Pentagon and Intelligence Community on AI and China, there will also be a new and complex mix of actors such as Tesla, X, xAI, and SpaceX CEO Elon Musk, whose views on these issues are likely to differ considerably from those channeling Silicon Valley financiers and influencers such as Peter Thiel and his supporters, some of whom are already part of the new administration.

As incoming Secretary of State Marco Rubio asserted this week, competition with China “will define the 21st century.” As of course so will advanced frontier AI models and where they are developed and deployed. These issues are clearly related, along with Taiwan and TSMC, and hence TSMC all leading US AI companies will remain in the middle of this dynamic.

In addition, industry, meaning companies across the AI stack, have weighed in on the rule, which did not involve the usual consultation with companies most affected.

Below, please find some of the commentary around the rule, and industry organizations and company individuals weighing in on the topic.

A range of people with differing backgrounds and positions on the issue have weighted in on the justifications for the US to control AI development globally.

Jan 7: Former Trump National Security Advisor Robert O’Brien op-ed: America Cannot Surrender Its AI Dominance | The National Interest

Jan 7: Center for European Policy Analysis blog: Biden’s Final Global Chip Controls Target China — and Allies - CEPA

Jan 6: Anthropic CEO Dario Amodei and former Trump Deputy National Security Advisor Matt Pottinger: https://www.wsj.com/opinion/trump-can-keep-americas-ai-advantage-china-chips-data-eccdce91

Dec 27: Rhodium analyst Reva Goujon in Foreign Affairs: The Real Stakes of the AI Race | Foreign Affairs

Dec 18: Jeff Westling (AAF): https://x.com/jeffreywestling/status/1869431623403643271?s=46&t=30niwQb-vtx1gFipJHTzAw

Dec 17: Robert O’Brien https://x.com/robertcobrien/status/1869104638856974572

Dec 16: Digital Liberty Blog- https://digitalliberty.net/bidens-midnight-rule-on-chips-will-only-help-the-chinese-communist-party/

Here is a compendium of articles written and letters sent to policymakers by industry and trade groups on the rule.

Jan 14: Multi-Industry Group letter: Leading Industry Associations Call on Administration to Cease Publication of Last-Minute National Security Regulations

Jan 7: ITIF Publication: Export Controls on AI Chips: Biden’s Overreach Risks U.S. Leadership in Tech | ITIF

Jan 6: SIA Statement: SIA Statement on Biden Administration’s Plan to Publish ‘Export Control Framework for Artificial Intelligence Diffusion

Jan 6: ITI Letter to Secretary Raimondo (non-public)

Jan 5: Second Blog Post by Ken Glueck, Executive Vice President at Oracle: Export Control Diffusion Confusion

Jan 3: Microsoft President Brad Smith blog post: The Golden Opportunity for American AI - Microsoft On the Issues

Dec 20: SIA letter to Commerce Secretary Gina Raimondo and National Security Advisor Jake Sullivan (with cc for Austin, Blinken, Granholm, OMB) (non-public)

Dec 20: USCBC letter to Secretary Raimondo (non-public)

Dec 20: Multi-industry letter to White House COS Jeff Zients and Sullivan (with cc for Blinken, Raimondo, OMB, Senate Banking, House Foreign Affairs) (non-public)

Dec 19: First Oracle Blog Post by EVP Glueck: https://www.oracle.com/news/announcement/blog/flood-the-zone-or-circle-the-wagons-2024-12-19/

Dec 9: ITI Letter to Raimondo, Sullivan and NEC Chief Lael Brainard (non-public)