A bill formed on the ‘AI Foundation Model Transparency Act’
has been issued by two lawmakers, that demands that creators of foundation
AI models disclose sources of their training data for clarification of
The bill is developed in alignment with the Biden
administration’s AI executive order. It has been released after recognizing the
importance of training data transparency around copyright as an increase in
lawsuits accusing AI companies of copyright infringement has been observed.
Several concerns about inaccuracy, imprecision, and biased information have
been raised against AI foundation models.
The lawmakers: Reps Anna Eshoo (D-CA) and Don Beyer (D-VA), would oversee the Federal Trade Commission (FTC) as it works with the National Institute of Standards and Technology (NIST) to determine rules around training data transparency.
Companies responsible for AI foundation models will be
required to disclose information regarding sources of training data and the
process of retaining it, as well as explain the limitations or risks of the
model, how it aligns with NIST’s planned AI Risk Management Framework, and deliver
information on the computational power used to train and run the model.
Additionally, the companies will have to report the
necessary information to “red team”- that would ensure that the model does not
provide inaccurate or harmful information around subjects of medicine / health,
biological synthesis, cybersecurity, politics, policing, financial loan
decisions, education, employment decisions, public services, and sensitive
populations such as children.