Panel
AI's Ethical Frontier: Responsible Regulation...by Whom? | Global Conference 2024
Milken InstituteJo Ling Kent, Rob Bonta, John B. Quinn, Rob Silvers, . will.i.am, Frankie Williams, Felicia
- The Department of Homeland Security is developing operational guidelines for auditing and quality control in essential services, establishing a new AI safety and security board with input from all technology sectors, and anticipating federal rules of the road to maintain public trust.
- Federal regulation regarding AI safety is viewed as inevitable in the US, though comprehensive legislation from Washington is not expected soon; instead, states like California are anticipated to act as "laboratories of democracy," creating a set of state laws and litigation that may eventually serve as a national baseline.
- The Department of Commerce is being urged to prioritize transparency, accountability, and auditing in rulemaking, while the Department of Homeland Security plans to use existing IP regimes and criminal investigations to combat likeness and content misappropriation.
- Regulatory frameworks are expected to shift from the US lag in "hard letter law" toward a European-inspired risk-based system, with the EU AI Act setting a global benchmark for governance, disclosure, and transparency through horizontal, non-sector-specific requirements.
- Under the EU AI Act, thresholds for risk levels will be reviewed in the 12 to 18-month period prior to applicability, creating a wait-and-see scenario as the EU AI Office adds detail, potentially reversing the burden of proof and applying strict liability to deployers.
- US future regulation is expected to involve a national plan institutionalizing rules of the road, potentially mirroring an FDA structure for AI, while ensuring minimum standards do not preempt states from implementing more stringent protections.
- Data practices are identified as a critical regulatory gap requiring urgency to prevent monopolies and empower individuals with their data rights, with potential tools including private rights of action and sophisticated privacy bars within the American legal system.
- There are significant concerns regarding AI representation, where current products lack diversity in voice and cultural context, necessitating solutions to prevent negative externalities in communities and ensure systems reflect the populations they serve.
- Industry is expected to face increased litigation and enforcement for violations of criminal, civil rights, antitrust, consumer protection, and healthcare laws, particularly in California, with a focus on preventing abusive content filtering and algorithmic manipulation of user agency.
- Future societal risks include the potential for systems to dictate user thought and emotion based on greed, necessitating urgent action to avoid a scenario by 2034 or 2044 where trust is lost and harmful products are left unchecked.
- While most states have legislation protecting names and likenesses, there are warnings that proposed federal privacy bills could preempt state protections, creating a tension between federal uniformity and state-level innovation and enforcement.
- The adoption of "hard law" in the US is expected to be slower than in the EU due to legislative complexity and a potential lack of understanding of the technology, though industry involvement is deemed essential to avoid unintended consequences.
- New York's requirement for annual certification of unbiased employment datasets is noted as a significant but challenging measure, illustrating the complexity of enforcing bias protections in practice.
- The US is expected to eventually create a balanced federal-state framework allowing courts to handle discrimination claims, while California aims to export vetted proposals to help other states address emerging challenges.
- Regulators are urged to move with urgency to avoid duplicating the social media trajectory where corporate greed led to societal harm, emphasizing that the American people must feel the technology works for them through protective frameworks.
- Representation gaps and the lack of education in inner cities are highlighted as structural issues that must be addressed to prevent AI systems from sounding like academia rather than the communities they serve.
- The EU AI Act's extraterritorial effects are expected to force global companies to comply with its governance standards, as major players are unlikely to exclude the European market from their operations.
- There is a recognition that the Biden administration possesses a degree of humility regarding AI dimensions, creating an appetite for guidance and a national plan that incorporates input from every key community.
- Risks include the possibility that US legislators lack the technical understanding necessary to craft effective laws, leading to solutions that may be incomplete or have unintended negative consequences if industry is not adequately engaged.