The Intelligence Advanced Research Projects Activity is seeking public input to determine the feasibility of creating a generative artificial intelligence model that, after being trained on relevant data, can produce novel products based on prompts that advance, however incrementally, human knowledge in science and engineering.
Of particular interest to IARPA are generative AI models that can produce outputs in math, algorithm or signal processing modalities, such as equations or theorems, algorithms or sequences of steps and waveforms or modulations, according to the RFI posted Tuesday on SAM.gov.
IARPA underscores in the solicitation that it is not seeking generative AI that calculates math problems, executes algorithms or processes signals. Rather, the agency is seeking generative AI that produces new knowledge in MASP form, which pushes the boundaries of complex sciences, even if only marginally.
Since IARPA is at present merely evaluating the feasibility of such a model, even those that can generate already known MASP products would be acceptable responses to the RFI.
Interested parties are also being asked to address a variety of issues, including how training datasets might be acquired, how prompting would work and how an AI model might be evaluated.
Responses are due March 29.
Federal leaders and industry experts will tackle the transformative power of artificial intelligence at the Potomac Officers Club’s 5th Annual Artificial Intelligence Summit, which will take place on March 21. Register now!