For now, GAAIA is imperfect, but it is a step in the right direction in taking a proactive approach to the challenges that AI is beginning to present.
The article begins by introducing the Great American AI Act (GAAIA), a comprehensive federal effort by Congressman Jay Obernolte and Congresswoman Lori Trahan to regulate AI development through harm mitigation standards, independent auditing, and a framework for addressing workforce disruptions. The discussion draft aims to prepare the federal government for the impact of AI on labor by commissioning reports and studies on displaced workers and retraining programs. However, Congressman Obernolte explicitly states that the bill does not aim to prevent workforce displacement, viewing technological revolutions as historically job-creating rather than job-eliminating. He advocates for government support through financial assistance and retraining, but not intervention to prevent job losses. The author raises concerns about this optimistic outlook, citing conflicting data on AI's impact on employment, with some studies predicting significant job eliminations while others find negligible effects. Obernolte draws a parallel to the agricultural revolution, arguing that AI will similarly enhance human productivity, though acknowledging that AI's impact could be much faster, occurring in years rather than decades. The author concludes this section by suggesting that actively preserving jobs against technological progress may be unworkable, and the focus should instead be on worker welfare once the aggregate impact of AI is understood.
A major point of contention within GAAIA is its provision to preempt state laws regarding AI regulation. While GAAIA focuses on mitigating catastrophic risks like loss of life or property, it does not directly address concerns such as deepfakes, AI-driven scams, or biased outputs from AI models used in critical decision-making like hiring. The author questions whether this preemption is warranted, as it could prevent states from protecting their residents against various AI abuses. Congressman Obernolte clarifies that he perceives the federal government's role as regulating AI's *development*, leaving *deployment* regulation to the states. He suggests that issues like deepfakes and scams fall under deployment. However, the author points out a significant problem with this approach: many states, including California with its Assembly Bill 489, are already enacting laws that regulate AI at the *development* stage, for example, addressing AI chatbots misleadingly posing as healthcare professionals. Such state legislation would likely be overridden by GAAIA, prompting organizations like the ACLU to call for the removal of the preemption language. The author argues that regulating AI at both development and deployment stages is crucial, as developers can institute necessary guardrails, and failing to include deployment-stage remediations or allow states to regulate development-stage issues in GAAIA would eliminate a powerful tool to limit AI abuse.
The article also delves into the potential for intrusive government surveillance through AI. It highlights a recent incident where developer Anthropic was blacklisted by the federal government for refusing to allow its AI to be used for mass surveillance and fully-automated weapons systems, indicating the government's interest in such applications. The author questions Congressman Obernolte about the political will in Congress to protect citizens against government AI use that violates privacy and unreasonable searches. Obernolte acknowledges that existing Bill of Rights protections, such as those against unreasonable searches and seizures, will need to be updated by the courts and eventually by Congress. However, the author counters by pointing out historical precedents where Congress and courts have allowed government surveillance programs like the NSA's PRISM, and recent examples of the Trump administration employing AI-enabled services for targeting activists and ICE allegedly using personal information to intimidate protesters. Obernolte expresses pessimism about government self-regulation, stating that asking a bureaucracy to regulate itself is a 'fool's errand.' Despite these grave concerns, Obernolte concludes the conversation on an optimistic note, emphasizing AI's potential as a powerful tool for disseminating human knowledge, enhancing productivity, and fostering prosperity, though the author does not fully share this positive outlook. The article concludes that while GAAIA is imperfect, it represents a necessary proactive step toward addressing AI's emerging challenges.