Recently completed
Topics
Foundation models are Artificial Intelligence (AI) models designed to produce a wide variety of outputs. They are capable of a range of possible tasks and applications, such as text, image or audio generation. Foundation models can be standalone systems or can be used as a 'base' for many other applications (i.e. text and image generation).
This project explores the state of the art of foundation models, especially pertaining to their use within the Department of Energy—and how to support the field as it expands.
Description
A National Academies of Sciences, Engineering, and Medicine consensus study will assess the state of the art in foundation models and their use across science research domains relevant to the Department of Energy mission. The study will address the following questions:
- What are some exemplar use cases where foundation models could impact scientific discovery and innovation?
- How can foundation models be used in conjunction with traditional modeling, computational, and data science approaches?
- How can challenges such as verification, validation, uncertainty quantification, and reproducibility best be addressed to advance trustworthy foundation models?
- What are priority research areas for investments to advance the development and use of foundation models in scientific applications? What are the tradeoffs in investing in foundation models versus other mathematical and computational approaches?
Contributors
Committee
Chair
Member
Member
Member
Member
Member
Member
Member
Member
Member
Member
Blake Reichmuth
Staff Officer
Conflict of Interest Disclosure
The conflict-of-interest policy of the National Academies of Sciences, Engineering, and Medicine (http://www.nationalacademies.org/coi) prohibits the appointment of an individual to a committee authoring a Consensus Study Report if the individual has a conflict of interest that is relevant to the task to be performed. An exception to this prohibition is permitted if the National Academies determines that the conflict is unavoidable and the conflict is publicly disclosed. A determination of a conflict of interest for an individual is not an assessment of that individual’s actual behavior or character or ability to act objectively despite the conflicting interest.
Dr. Koumoutsakos has a conflict in relation to his service on the Committee on Foundation Models for Scientific Discovery and Innovation because he is currently on sabbatical at Google DeepMind, a subsidiary of Alphabet Inc. He also holds stock in Google.
The National Academies has concluded that for the committee to accomplish the tasks for which it was established, its membership should include at least one member with current experience working directly with industry leaders developing products or services using foundational models in the US and internationally. As described in his biographical summary, Dr. Koumoutsakos is working with researchers at Google’s DeepMind, a leading developer of foundation models, including tools that have contributed to research at Department of Energy national laboratories.
The National Academies has determined that the experience and expertise of Dr. Koumoutsakos is needed for the committee to accomplish the task for which it has been established. The National Academies could not find another available individual with the equivalent experience and expertise who did not have a conflict of interest. Therefore, the National Academies has concluded that the conflict is unavoidable.
The National Academies believes that Dr. Koumoutsakos can serve effectively as a member of the committee, and the committee can produce an objective report, considering the composition of the committee, the work to be performed, and the procedures to be followed in completing the study.
Dr. Kearns has a conflict in relation to his service on the Committee on Foundation Models for Scientific Discovery and Innovation because of his position at Amazon Web Services, Inc., a subsidiary of Amazon. A portion of Dr. Kearns’s compensation is in Amazon Stock.
The National Academies has concluded that for the committee to accomplish the tasks for which it was established, its membership should include one or more members with direct current experience working with different aspects of industry leading foundational models. As described in his biographical summary, as an Amazon Scholar and Founding Director of the Warren Center for Network and Data Sciences, Dr. Kearns has extensive experience working with the intricate aspects of algorithm design and computational learning for foundation models and their use in scientific discovery and innovation.
The National Academies has determined that the experience and expertise of Dr. Kearns is needed for the committee to accomplish the task for which it has been established. The National Academies could not find another available individual with the equivalent experience and expertise who did not have a conflict of interest. Therefore, the National Academies has concluded that the conflict is unavoidable.
The National Academies believes that Dr. Kearns can serve effectively as a member of the committee, and the committee can produce an objective report, considering the composition of the committee, the work to be performed, and the procedures to be followed in completing the study.
Sponsors
Department of Energy
Staff
Blake Reichmuth
Lead