NY Assemblyman Alex Bores: The AI super PAC doesn't want there to be any regulation whatsoever

Key Concepts

• Responsible AI Safety and Education Act (Res Act): A New York State bill aiming to codify voluntary AI safety commitments into law.
• Executive Order: A directive issued by the President of the United States that has the force of law.
• Super PAC: A type of independent political action committee that can raise unlimited sums of money to influence elections.
• AI Safety Plans: Commitments made by AI companies to develop and implement measures to ensure the safe development and deployment of AI.
• Critical Safety Incidents: Significant events or failures related to AI safety that require reporting.
• Unreasonable Risk: A threshold for AI model disclosure, indicating a significant potential for harm.
• Laboratories of Democracy: A concept referring to states as testing grounds for new policies and ideas.
• Federal Standard: A uniform set of regulations or guidelines established by the federal government.
• Interstate Commerce Laws: Federal laws that regulate business activities that cross state lines.
• Substantial Risk of Killing 100 People / $1 Billion in Damage: The specific legal standard proposed in the Res Act for preventing the release of high-risk AI models.

New York's Res Act and AI Regulation

Assemblyman Alex B. Boris, a computer science graduate and former engineer at Palantir, is a sponsor of New York State's Responsible AI Safety and Education Act (Res Act). He views AI as a powerful technology with immense potential for good, such as curing diseases, but also acknowledges its potential for harm, like the development of bioweapons. His primary concern is managing these risks and mitigating the downsides while maximizing the benefits.

Boris argues that the Res Act is not introducing new burdens but rather codifying existing voluntary commitments made by AI companies to the White House and in international summits. These commitments include developing safety plans, reporting critical safety incidents, and refraining from disclosing models that pose an unreasonable risk.

Opposition and Super PAC Criticism

A super PAC has targeted Assemblyman Boris, accusing the Res Act of slowing American progress and potentially ceding the global AI race to China. Boris contends that this opposition stems from a desire to avoid any regulation whatsoever. He believes that if the super PAC genuinely supported a national AI standard, they would endorse his congressional platform, which advocates for such a standard. Instead, he sees their actions as an attempt to silence him through significant financial pressure.

States' Rights vs. Federal Standards in AI Regulation

The debate surrounding the Res Act touches upon the broader issue of states' rights versus federal oversight in AI regulation. Boris highlights the role of states as "laboratories of democracy," capable of moving faster than the federal government. He points to the proliferation of state-level privacy laws as an example of states taking the lead while federal legislation lags.

While Boris agrees that a federal AI standard is ultimately desirable, he questions whether states should be prevented from making progress in AI safety while waiting for the federal government to act. He advocates for a collaborative approach where states and the federal government work together.

Boris believes that interstate commerce laws suggest the federal government should have the authority to preempt state regulations. He envisions a federal framework that mirrors the Res Act, requiring AI companies to:

• Make safety standards public.
• Adhere to these standards.
• Disclose critical safety incidents.

He cites a recent incident where Anthropic reported China using their AI model for cyberattacks on U.S. government institutions and chemical manufacturing plants. Boris argues that such incidents should be legally mandated disclosures for all major AI developers.

The "Unreasonable Risk" Standard and Pushback

The most significant pushback from AI companies, according to Boris, centers on the legal standard of not releasing models that their own tests indicate have an "unreasonable risk of causing harm." He draws a parallel to the tobacco industry, which knew about the dangers of cigarettes but denied them publicly. The Res Act proposes that if AI developers conduct tests and fail to meet their own safety benchmarks, this should constitute a legal standard preventing release.

Boris clarifies that the Res Act's standard is not about minor risks but about "a substantial risk of killing 100 people, or leading to $1 billion in damage." This threshold is designed to prevent the release of AI models with catastrophic potential and to hold companies accountable for acting irresponsibly, especially when they are the experts in assessing these risks.

Conclusion

Assemblyman Alex Boris advocates for a proactive approach to AI regulation, emphasizing the need to codify existing safety commitments into law through measures like the Res Act. He believes that while a federal standard is ideal, states play a crucial role in driving progress. The opposition, represented by a super PAC, appears to favor a complete absence of regulation, which Boris argues is detrimental to public safety and national interests. The core of the debate lies in establishing clear legal standards for AI development and deployment, particularly concerning the disclosure of models that pose significant risks.