The Biden administration said Wednesday it is looking at putting artificial intelligence “nutrition labels” on new tech products similar to the Food and Drug Administration’s mandatory food labeling regime.
The Departments of Treasury and Commerce published new reports scrutinizing the risks of AI that revealed federal officials are studying whether to impose the labels.
“Financial institutions stated that they would benefit from a standardized description, similar to a nutrition label, for vendor-provided Generative AI systems to clearly identify what data was used to train a model, where it came from, and how any data submitted to the model will be incorporated,” the Treasury Department report said.
Department officials said they “will work with the financial sector, NIST, the National Telecommunications and Information Administration (NTIA), and the Cybersecurity and Infrastructure Security Agency (CISA) to identify whether such recommendations should be explored further.”
The Commerce Department’s NTIA published a prototype of the potential label in its report released on Wednesday, spotlighting a version developed by tech company Twilio.
The sample label is titled “AI Nutrition Facts” and includes details of “Trust Ingredients” including whether there is a “Human in the Loop” referring to the level of human participation in the operation of the cutting-edge technology.
“AI nutritional labels, by analogy to nutritional labels for food, present the most important information about a model in a relatively brief, standardized, and comparable form,” the NTIA report said.
The NTIA’s report said new transparency-oriented reports, cards, datasheets and nutrition labels appeared “promising and some should become standard industry practice as accountability inputs.”
But such a labeling regime could prove challenging for the federal government to implement, as the Biden administration found there is widespread disagreement over how to define AI.
The Treasury Department’s report focused on AI-specific cyber risks to the financial sector and interviewed 42 representatives from the sector, information technology firms, data providers and anti-money laundering and anti-fraud companies. The participants did not share a common definition of what AI is.
“Treasury found that there is no uniform agreement among the participants in the study on the meaning of ‘artificial intelligence,’” the Treasury Department report said.
The federal government uses an AI definition established in President Biden’s executive order on the technology published last year. Mr. Biden’s definition said AI refers to a machine-based system given human-defined objectives that can make predictions, recommendations or decisions influencing real or virtual environments.
While the Biden administration is using one definition, some participants complained to the Treasury Department that vendors’ own descriptions of how their AI systems work do not match the reality of how financial institutions saw the AI working.
The Treasury Department said it would work with the financial sector in the months ahead on AI risk and fraud issues while also investigating other AI-related effects on “consumers and marginalized communities.”