How Should AI Be Regulated? Use vs. Development

Open source is vital for innovation, particularly benefiting hobbyists, academics, and startups, yet regulatory uncertainty in the U.S. is hindering the development of robust open source models. This situation is pushing these groups toward Chinese alternatives, threatening U.S. competitiveness. Current regulatory frameworks focus on development rather than use, creating loopholes due to the ambiguous definition of AI. Effective policy should prioritize the direct use of AI technologies until associated risks are better understood.

Experts discuss the need for evidence-based, technology-neutral policies that protect against misuse while promoting innovation. They reflect on the implications of recent congressional testimonies and executive orders that introduce new restrictions on computing power and express skepticism towards open source software. Shifting focus from regulating use cases to model development could stifle innovation and fail to address malicious actions.

The historical context of regulating behaviors rather than inventions is highlighted, particularly in relation to malware and cybercrime. While creating malware is not a crime, transmitting harmful software is. Techniques from malware development can also enhance system security. Distinguishing between beneficial and harmful coding practices is complex, but the emphasis should remain on curbing negative activities. The Computer Fraud and Abuse Act has effectively targeted bad actors, although some evade regulation.

Challenges in regulating coding in the U.S. arise from global development, complicating efforts to prevent misuse without stifling innovation. Current discussions often misplace focus on development rather than use, which is flawed. Unlike industries such as automotive and aviation, software regulation requires a different model. The conversation has historically included diverse perspectives, but academia and venture capital voices are currently underrepresented, leading to an imbalance in discussions about innovation and regulation.

Understanding the marginal risks associated with new technologies is crucial before implementing regulations. Experts note that the marginal risks of AI remain unclear, presenting a significant research challenge. Defining risks is essential for effective policy development, especially in AI. While existing regulations apply to certain applications, new regulations must be informed by a clear understanding of specific risks. The conversation also explores the legal responsibilities of users who misuse these technologies and differing perspectives on the potential emergence of artificial general intelligence.

Concerns about existential risks associated with AI systems highlight the need for effective regulation. Some engineers argue that current AI does not approach General Intelligence and may never achieve it, while others believe existing engineering challenges hinder advanced AI development. There is a consensus on the necessity for sensible regulation to prevent illegal activities, emphasizing the importance of understanding marginal differences in AI systems to inform regulations. Historical practices in cybersecurity underscore the need for evidence-based policies, as seen in California's SB 1047 aimed at regulating large AI models.

Experts express concerns about the current policy process's effectiveness in addressing risks, drawing parallels to past mistakes in social media regulation. The unpredictability of social media's evolution suggests that risks can only be identified as they emerge, underscoring the need for ongoing stakeholder conversations about regulating harmful technologies. The dialogue also emphasizes the need for a balance between innovation and safety, particularly in AI applications and medical innovation.

There is a call for strong justification for any changes to the current equilibrium, considering all trade-offs involved. Concerns are raised about Europe's early regulatory efforts, with Italy's ban on ChatGPT seen as detrimental, potentially causing Europe to lag behind the U.S. and China in AI development. The relationship between regulation and AI success is questioned, suggesting that regulatory differences are not the sole factor in China's advancements. The chilling effect of an uncertain regulatory environment on researchers and the shift in attitudes toward open source development are also noted.

The challenges surrounding the release of open-source AI models by U.S. companies are highlighted, as regulatory uncertainties lead to a reliance on Chinese models, with 80% of new companies using them. This raises concerns about competitive advantages and geopolitical implications. Policymaking involves trade-offs, with an emphasis on robust debates about immediate risks. Understanding historical precedents is crucial when regulating new technologies.

The conversation stresses the importance of distinguishing between regulating the use of AI and its development, as development-focused regulations can disproportionately impact startups struggling with funding and compliance. Existing regulations already apply to AI, and lawmakers should focus on identifying gaps rather than creating new, potentially obsolete laws. Proposed legislation should be technology-neutral and emphasize use-based regulations to prevent misuse without hindering innovation. The balance of regulatory power between federal and state governments is also crucial, as states can enact consumer protection laws. Ultimately, the conversation calls for a framework that addresses misuse while fostering innovation in the rapidly evolving AI landscape.