U.S. Congressman Beyer on AI challenges facing America and the World

Congressman Don Beyer of Virginia, a returning guest on the Practical AI podcast, is enrolled in a PhD program in AI at George Mason University, which host Chris Benson describes as unusual among members of Congress. Beyer describes the AI policy landscape as almost entirely changed since the Biden administration, driven by both the new administration and the rapid acceleration of AI itself. Trump rescinded Biden's executive order on AI, which Beyer describes as the largest executive order ever written by a president, then issued his own order that was largely similar in content. Beyer considers Trump's decision to preserve the AI Safety Institute at NIST the most important AI policy action of the new administration, noting it was renamed and had leadership changed but retained much of the same staff. David Sacks was appointed AI czar and Mike Krasios was appointed chief science advisor, both described by Beyer as Silicon Valley business people rather than scientists. The early Trump administration pushed toward accelerationism on AI, opposing new laws or restraints and framing the priority as beating China, including reversing Biden's policy restricting Nvidia chip exports by approving the sale of H200 chips to China.

An AI model called Mythos, introduced a few weeks before the recording, served as a wake-up call within the administration. Beyer says Mythos demonstrated the ability to analyze and unravel existing cybersecurity protections, a capability he had previously assumed would not arrive until quantum computing. Beyer credits Anthropic with acting responsibly by sharing the Mythos code with a limited group in advance of wider release to allow security defenses to be strengthened. He expects OpenAI, Gemini, and Chinese developers to develop similar capabilities in the near term and suggests Mythos may prompt a wholesale rethinking of cybersecurity architecture, describing it as a phase shift.

The only federal AI-related legislation passed so far is Ted Cruz's Take It Down Act, covering nonconsensual sexual imagery. A bipartisan congressional task force convened by Mike Johnson and Hakeem Jeffries produced 80 specific AI legislative recommendations, of which only one has been enacted. Beyer argues that in the absence of a federal framework, state and local governments should not be restricted from acting, citing over 700 pieces of state AI legislation already introduced. Republican Congressman Jay Obernolte has argued against a Tower of Babel scenario where Virginia, California, and Texas each have different AI regulations. Beyer cites California's HB 53 as an important first step in state-level AI regulation, though it was vetoed or amended by the governor before becoming law. Beyer also proposes a new Geneva Convention-style international agreement on AI involving the U.S., China, Europe, and Middle Eastern nations, while acknowledging that strong U.S. regulation without equivalent Chinese regulation would be unworkable in the long run.

Beyer cites Dario Amodei's predictions of 25 to 50 percent white collar job displacement within two to five years as a major shared concern across party lines, contrasting that speed with the agricultural and industrial revolutions, which unfolded over decades or a century. He told enrolled agents and small accounting firm workers that AI handling accounts payable, accounts receivable, and payroll could affect not just their roles but those of people working for them, with implications across 18 million white collar workers in America. Beyer is co-leading a House bill for a commission on the future of the economy focused on what to do if AI displaces massive numbers of workers, with a Senate companion led by Mark Warner. Beyer says the likely policy response is not regulation banning AI use in specific jobs but rather investments ensuring people have both a means to live and meaningful activity. He expressed concern about the concentration of AI-generated wealth among a small number of very wealthy individuals and said that two-thirds of the population cannot be left behind.

Beyer identified surveillance as a potential bipartisan concern, citing the Doge episode in which Social Security and tax records were allegedly copied and loaded into Grok, Elon Musk's AI system. He referenced Tim Berners-Lee as working on a project called a pod, a personal data container that would require permission or payment to access, intended to let individuals monetize their own information rather than companies like Meta doing so. Beyer warned that AI makes financial scams significantly more sophisticated and harder to detect, noting that fraudsters already cost older Americans trillions of dollars.

On autonomous weapons, Beyer noted that DeepMind, OpenAI, and Anthropic all originally opposed their AI being used in systems without a human in the loop, while the current administration does not want a human in the loop for autonomous weapons decisions. He posed the direct strategic dilemma that a weapons system without human authorization will likely defeat one that requires it, creating pressure to remove human oversight entirely, while adversaries including China, India, and North Korea may deploy fully autonomous systems.

Beyer identified job displacement as the immediate concern and existential risk as the longer-term concern. He described the alignment problem, meaning whether artificial superintelligence will want what humans want, as a central existential challenge. He noted that neither Demis Hassabis nor Jeff Hinton signed the 2023 letter signed by approximately 700 people calling for a halt to advanced AI development, with Hinton declining largely because he believed it would be unenforceable given the diversity of global actors.