Biden’s AI executive order slams small entities
Dylan Dean
Video Embed
Christmas came early this year for Sam Altman, CEO of OpenAI. The present was President Joe Biden’s new executive order restricting artificial intelligence. Altman used to preach the benefits of open-source AI, so much so that he named his company after it. Now, he’s singing a different tune — and the White House is listening.
The order signed on Monday covers topics ranging from how intellectual property law applies to AI training data to specific restrictions on AI trained to understand DNA. The consequences of this order are unclear, as is its legality. Still, Cabinet staff have been directed to prepare various proposals and regulations in the coming months. What is clear is that the AI landscape is about to become much more difficult for smaller players to navigate.
BIDEN’S UNCONSTITUTIONAL AI ORDER
The order sets thresholds for how much computing power can be used to train models. More computing power means that the model will be able to “see” more information during training and therefore has more opportunity to learn. Reports must be provided to the secretary of commerce if trained models surpass the threshold. This is perfectly feasible for large corporations, but smaller entities and individuals who use “fine tuning” to teach AI models to perform specific tasks are at risk of being above the threshold if the count is cumulative, something the order does not specify. This is no accident since the secretary is directed to solicit input on the “risks” associated with fine-tuning, as well as policy recommendations regarding open-source models.
The government going after fine-tuning and open-source models specifically poses a grave threat to smaller businesses that rely on fine-tuning open models such as Meta’s Llama-2 and Mistral AI’s Mistral 7B. Yann LeCun, chief AI scientist at Meta, claimed the lobbying activities of Altman and others were “attempting to perform a regulatory capture.” LeCun argues that this regulatory capture, if successful, would mean “a small number of companies from the West Coast of the US and China will control AI [and] people’s entire digital diet.” He’s right, and he’s also right about the counterfactual — the world where AI is open and decentralized is a world that protects and promotes creativity, democracy, and cultural diversity.
Altman recognizes the level of power that could come from the centralization of AI, phrasing it as having the potential to “capture the light cone of all future value in the universe.” This is a strong claim, since humanity’s “light cone” represents the sum of what is reachable in the universe by humanity for all time. At the time, he was warning of the dangers of that centralization. Now, the regulatory capture he lobbied for is on its way in, and Altman is one step closer to capturing his light cone.
The order is not all bad. It also calls for the modernization of immigration pathways for AI experts and the consideration of expanding a visa renewal program to STEM students. Other areas could go either way, such as the call for recommendations to be made to the president for another executive order on copyright and AI, which could either help solve or exacerbate the difficulties intellectual property laws pose to AI. Notably, these areas where policy could end up positive for innovation are those where OpenAI stands to benefit as well.
While the ramifications of this executive order will not be clear for several months, with some deadlines stretching as far as April 2025, what is clear is that the corporations lobbying for regulatory capture are taking the lead. If Altman himself is to be believed, we only have everything to lose.
CLICK HERE TO READ MORE FROM THE WASHINGTON EXAMINER
Dylan Dean is a software engineer with degrees in electrical and computer engineering and is an advocate for the decentralization of emerging technologies.