In a landmark move that could potentially change the landscape of artificial intelligence (AI), U.S. Representatives Anna Eshoo (D-CA) and Don Beyer (D-VA) have proposed the AI Foundation Model Transparency Act.
The legislation aims to increase the transparency of AI foundation models by requiring companies to disclose sources of training data, describe the limitations or risks of the model, and outline alignment with the National Institute of Standards and Technology’s (NIST) planned AI Risk Management Framework.
Under this bill, companies would also be mandated to report any attempts to “red team” the AI model, which involves scrutinizing its security robustness. The legislation is seen as an enhancement of the Biden administration’s AI executive order, which establishes reporting standards for AI models but isn’t a law. The AI Foundation Model Transparency Act would serve to bring these transparency requirements into federal law.
The bill also incorporates provisions addressing copyright concerns, given the increasing public use of foundation models and the ensuing legal issues.
Furthermore, the proposed law would direct entities such as the Federal Trade Commission, NIST, the Copyright Office, and the Office of Science and Technology Policy to develop transparency standards for AI deployers. This incorporates unveiling the training data used, explicating the training process, and clarifying if user data is harnessed during inference.
Many view this bill as a critical step towards empowering consumers, enabling them to make well-informed decisions when leveraging AI. This is especially significant when biases in models can lead to detrimental impacts, like rejections for housing or loan applications or misguided medical decisions.
Moreover, the AI Foundation Model Transparency Act would necessitate reporting of the computational power used to train and run the AI model. AI developers would also need to publicize specific information to consumers.
The proposed bill has not yet been assigned to a committee for discussion. Its progress may be delayed with the upcoming election campaign season.
As the world keenly watches the unprecedented move, it remains to be seen how this legislation will shape the AI landscape by enforcing unprecedented transparency and accountability measures.