President Joe Biden issued yesterday a sweeping executive order aiming to impose federal regulation on the development of artificial intelligence technologies, such as large language models like ChatGPT. The executive order cites the emergency powers of the Korean War-era Defense Production Act as the justification for imposing federal regulation on AI technologies. As my Reason colleague Eric Boehm has pointed out, "the Defense Production Act has become a license for central planning." Taken as a whole, the new order amounts to federal central planning for artificial intelligence.
Among other things, the order will "require that developers of the most powerful AI systems share their safety test results and other critical information with the U.S. government," according to the White House. Specifically, the new federal AI regulators are supposed to oversee any "foundation model" that purportedly "poses a serious risk to national security, national economic security, or national public health and safety" by requiring that developers report to the secretary of commerce the results of extensive "red-team safety tests." Roughly speaking, foundation models are large language models like OpenAI's GPT-4, Google's PaLM-2, and Meta's LlamA 2. Red-teaming is the practice of creating adversarial squads of hackers to attack AI systems with the goal of uncovering weaknesses, biases, and security flaws. As it happens, the leading AI tech companies— OpenAI, Google, Meta—have been red-teaming their models all along.
