MIT's AI Governance Framework: How It Could Shape US Policy for the Future
Published on: March 10, 2024
Addressing the need for comprehensive governance of artificial intelligence, a committee of MIT experts has released a set of policy briefs. These briefs outline a pragmatic framework for AI regulation, suggesting the extension of existing regulatory and liability mechanisms to oversee AI developments.
The primary paper, titled 'A Framework for U.S. AI Governance: Creating a Safe and Thriving AI Sector,' proposes that existing U.S. government entities, already overseeing relevant domains, can effectively regulate AI tools. The focus is on identifying AI tools' purposes, tailoring regulations to fit these applications.
Dan Huttenlocher, dean of the MIT Schwarzman College of Computing, emphasizes the practical approach of this framework, building on existing regulatory structures for high-risk areas. Asu Ozdaglar, Deputy Dean of Academics at the College, also played a pivotal role in overseeing this initiative.
The policy briefs come amid a surge in AI interest and investment, with the European Union also working on AI regulations. These efforts face challenges in regulating both general and specific AI tools, with issues like misinformation and surveillance looming large.
David Goldston, director of the MIT Washington Office, underscores MIT's obligation in contributing to this discourse, given its leadership in AI research and development.
The main policy brief advocates for extending current policies to cover AI, utilizing existing agencies and legal frameworks. It highlights the importance of AI providers defining the purpose and intent of AI applications in advance, to ascertain the applicable regulations and liability.
The papers also address the complexity of AI systems existing in 'stacks,' where the responsibility may be shared among different system levels. General-purpose AI tools, as part of these stacks, should also bear accountability for specific problems they contribute to.
In addition to leveraging existing agencies, the briefs propose new oversight capabilities, including the advancement of AI tool auditing, possibly through a nonprofit entity or a federal body like the National Institute of Standards and Technology (NIST).
The policy framework suggests the creation of a new 'self-regulatory organization' (SRO) agency for AI, akin to FINRA in the financial industry, to accumulate domain-specific knowledge for responsive and flexible governance.
The committee's work extends to exploring 'human plus' legal issues, where AI's capacities exceed human abilities, necessitating special legal considerations for tools like mass-surveillance and fake news generation.
The set of policy papers covers a range of regulatory issues in detail, including labeling AI-generated content and examining general-purpose language-based AI innovations.
The policy briefs advocate for research on AI's societal benefits, with a focus on AI augmenting rather than replacing human workers, fostering economic growth distributed throughout society.
The MIT committee's initiative reflects a commitment to bridging the gap between AI enthusiasm and concerns, advocating for adequate regulation alongside technological advances.
This effort, led by Huttenlocher, Ozdaglar, Goldston, and other notable MIT scholars, represents MIT's dedication to shaping the national and global discourse on AI governance, ensuring its responsible and beneficial development.