Slots: 2. An organization may submit no more than two preliminary proposals to this solicitation as lead institution. This limit is solicitation-wide and applies across the groups and themes. An organization may submit up to two full proposals that correspond to preliminary proposals reviewed under this solicitation. In the event that an organization exceeds these limits, preliminary proposals will be accepted based on earliest date and time of preliminary proposal submission, i.e., the first two preliminary proposals will be accepted, and the remainder will be returned without review. A full proposal that does not correspond to a preliminary proposal reviewed in this program will be returned without review.
Deadlines
Internal Deadline: Friday, September 15th, 2023 for all groups and themes. Contact RII.
LOI: October 31, 2023 for themes listed Under Group 1; January 12, 2024 for themes listed under Group 2
External Deadline: February 16, 2024 for Themes listed under Group 1; May 17, 2024 for themes listed under Group 2.
Award Information
Award Type: Cooperative Agreement
Estimated Number of Awards: 5
Anticipated Award Amount: Institute awards will be made for between $16,000,000 and $20,000,000 for four to five years ($4,000,000 per year on average)
Who May Serve as PI: There are no restrictions or limits.
Link to Award: https://www.nsf.gov/pubs/2023/nsf23610/nsf23610.htm
Process for Limited Submissions
PIs must submit their application as a Limited Submission through the Research Initiatives and Infrastructure (RII) Application Portal: https://rii.usc.edu/oor-portal/. Use the template provided here: RII Limited Submission Applicant Template
Materials to submit include:
- (1) Two-Page Proposal Summary (1” margins; single-spaced; standard font type, e.g. Arial, Helvetica, Times New Roman, or Georgia typeface; font size: 11 pt). Page limit includes references and illustrations. Pages that exceed the 2-page limit will be excluded from review. You must use the template linked above.
- (2) CV – (5 pages maximum)
Note: The portal requires information about the PIs in addition to department and contact information, including the 10-digit USC ID#, Gender, and Ethnicity. Please have this material prepared before beginning this application.
Purpose
Artificial Intelligence (AI) has advanced tremendously and today promises personalized healthcare; enhanced national security; improved transportation; and more effective education, to name just a few benefits. Increased computing power, the availability of large datasets and streaming data, and algorithmic advances in machine learning (ML) have made it possible for AI research and development to create new sectors of the economy and revitalize industries. Continued advancement, enabled by sustained federal investment and channeled toward issues of national importance, holds the potential for further economic impact and quality-of-life improvements.
The 2023 update to the National Artificial Intelligence Research and Development Strategic Plan, informed by visioning activities in the scientific community as well as interaction with the public, identifies as its first strategic objective the need to make long-term investments in AI research in areas with the potential for long-term payoffs in AI. AI Institutes represent a cornerstone Federal Government commitment to fostering long-term, fundamental research in AI while also delivering significantly on each of the other eight objectives in that strategy. The National Security Commission on Artificial Intelligence (NSCAI) identifies AI Institutes as a key component of a bold, sustained federal push to scale and coordinate federal AI R&D funding and to reinforce the foundation of technical leadership in AI.
This program is a multisector effort led by the National Science Foundation (NSF), in partnership with the Simons Foundation (SF), the National Institute of Standards and Technology (NIST), Department of Defense (DOD) Office of the Under Secretary of Defense for Research and Engineering (OUSD (R&E)), Capital One Financial Corporation (Capital One), and Intel Corporation (Intel).
This program solicitation expands the nationwide network of AI Research Institutes with new funding opportunities over the next two years. In this round, the program invites proposals for institutes that have a principal focus in one of the following themes aimed at transformational advances in a range of economic sectors, and science and engineering fields:
- Group 1 – Awards anticipated in FY 2024:
- Theme 1: AI for Astronomical Sciences
- Theme 1: AI for Astronomical Sciences
- Group 2 – Awards anticipated in FY 2025:
- Theme 2: AI for Discovery in Materials Research
- Theme 3: Strengthening AI
For the institute themes listed in Group 1, NSF anticipates awards to start in FY 2024; and for themes listed in Group 2, NSF anticipates awards to start in FY 2025. Each group has a specific set of due dates and review timeline pertaining only to that group. More detail is found under Due Dates and in the timeline provided in the Program Description.
II.A. AI Research Institutes Scope
The vision of the National AI Research Institutes program is broad and ambitious. It is expected that each AI Research Institute will pursue this vision in ways that are uniquely suited to its selected research focus, facilities, collaborations, and other unique circumstances. Proposers are encouraged to convey the unique qualities of the proposed Institute, while addressing the following desiderata common to all AI Research Institutes proposed to this program:
- AI Research Institutes advance foundational AI research that will have broad and lasting impact, contributing new knowledge or methods toward understanding of the mechanisms underlying thought and intelligent behavior and their implementation in machines (see the definition of AI specified above). Institutes aimed at advancing established AI lines of research should demonstrate the potential to radically advance these areas beyond the state of the art. Institutes might also address new foundational AI research priorities that arise from rapid advances in AI and the increasing ubiquity of AI-enabled technology. Institute proposals that do not describe a clear plan to achieve ambitious advances in foundational AI research are not likely to be responsive to this solicitation.
- AI Research Institutes conduct use-inspired research that both informs foundational AI advances and drives innovations in related sectors of science and engineering, segments of the economy, or societal needs. Effective use-inspired research achieves synergy among a group of researchers to enable transformative advances in AI, related sectors, and the interfaces between these areas. This dimension of an AI Research Institute will feature clear and compelling goals to advance AI and to accelerate the fielding of AI-powered innovation; it also enhances the transfer of knowledge through the meaningful exchange of scientific and technical information with external stakeholders such as industrial partners, public policy makers, or international organizations, as well as with the broader scientific and educational community. Through use-inspired research, Institutes have the potential to create and share new community infrastructure, including data and software, to further research, promote reproducibility, and support education.
It is critical that proposals clearly specify how the use-inspired context for Institute research reveals the opportunities for foundational AI advances and how those foundational AI advances in turn contribute to the related sectors that define the use-inspired context. - AI Research Institutes actively build the next generation of talent for a diverse, well-trained workforce. Specifically, AI Research Institutes should leverage the visionary nature of their research foci to drive new and innovative education and development tailored toward, e.g., undergraduates, graduate students, and post-doctoral researchers, as well as through community colleges and skilled technical workforce training and other opportunities as appropriate that advance knowledge and education of AI, including public understanding of AI. This could include innovative pedagogy and instructional materials, advanced learning technologies, project-driven training, cross-disciplinary and collaborative research, industry partnerships, and new career pathways. Institutes should offer broad, deep, and diverse experiences to build the next generation of the AI workforce, with a focus on broadening participation among the full range of groups currently under-represented in science and engineering. AI Research Institutes should maximize their unique position to grow the next generation of talent that will provide new discoveries and leadership.
- AI Research Institutes are coherent multidisciplinary groups of scientists, engineers and educators appropriate for a large-scale, long-term research agenda for the advancement of AI and the fielding of AI-powered innovation in application sectors of national importance. The multidisciplinary nature of these Institutes will catalyze foresight and adaptability beyond what is possible in single research projects; further, the individual projects that an Institute carries out should meaningfully integrate into fundamental contributions beyond the sum of the individual projects.
- Each Institute will be comprised of multiple organizations working together to create significant new research capabilities. NSF and partner organizations seek to grow the network of National AI Research Institutes in lead organizations distributed throughout the country to grow new centers of AI leadership and leveraging existing centers of excellence as appropriate. Institutes are strongly encouraged to include organizations that can directly contribute to NSF’s commitment to broadening participation by engaging a diverse, globally engaged research community, integrating research with education and building capacity, and expanding efforts to include the participation of the full spectrum of diverse talent in STEM and diverse institutions across all geographical regions. Participants should be meaningfully integrated into a diverse Institute that is more than just the sum of the parts. Each Institute will have a lead PI with demonstrated vision, experience, and capacity to manage a complex, multi-faceted, and innovative enterprise that integrates research, education, broadening participation, and knowledge transfer. Each Institute will also be staffed with a Managing Director or Project Manager (distinct from the lead PI) and a suitable Management Team to oversee the operations of the Institute. An External Advisory Board is required for all AI Research Institutes. (Potential Advisory Board members should not be approached or identified until the Institute is funded.)
- AI Research Institutes are nexus points for collaborative efforts. The “nexus point” function in this program is not a mere state of being, but rather an active set of priorities, programs, mechanisms, etc., whereby an AI Research Institute pursues the continuing growth of collaborations with external partners to bring together people, ideas, problems, and technical approaches for maximum impact beyond the members and the boundaries of the Institute itself. As nexus points, Institutes have the potential to continue to connect with new partners with the best teams and approaches from institutions of higher education, federal agencies, industry, nonprofits/foundations, centers/institutes, and national networks. As nexus points, Institutes promote organizational collaborations and linkages within and between campuses, schools, and the world beyond, and further the Institute’s mission to broaden participation in research, education, and knowledge transfer activities through a network of partners and affiliates.
II.B. Institute Themes in GROUP 1 Awards anticipated in FY 2024:
Proposals are being solicited in the following high-priority areas for awards anticipated in FY 2024. Due dates listed for Group 1 apply for submissions to the themes in this group.
Theme 1: AI for Astronomical Sciences
With current and future astronomical experiments poised to flood the field with petabytes of high-quality imaging and spectroscopic data over a wide range of wavelengths of light and with a high temporal cadence, AI technology will be essential for mining and analyzing these data. The primary goal of an AI Institute in astronomy is to bring together astronomy and AI experts to tackle important challenges in astronomy, as well as the advances in AI that are needed to overcome these challenges. An AI institute will serve as a hub and resource for the broader astronomical community by making tools publicly available and by promoting the education and training of the astronomical community in AI methods.
Proposals can address any relevant combination of AI use cases. Some examples are provided below. This list is meant to stimulate thought about the many potential application areas and is not prescriptive.
- Clean raw astronomical imaging, spectroscopic, or time series data by removing sources of statistical and systematic noise.
- Derive accurate estimates of physical parameters of objects or extract statistical measurements directly from raw observational data.
- Classify objects on the fly for rapid follow-up observation.
- Find rare events using anomaly detection.
- Estimate physical model constraints directly from raw observational data.
- Predict the behavior of complex theoretical simulations to reduce their computational cost.
- Develop fast and accurate emulators that can be used in statistical modeling of data.
- Create an “AI astronomer” who can assist with exploring multidimensional data sets or who knows the astronomical literature.
Many of these applications may require foundational advances in AI to succeed. For example, advances may be required in dealing with especially large and complex data sets, in adding knowledge of physical laws into AI models, or in developing interpretable AI methods with well understood error properties. Proposals should clearly justify both the selection of the targeted astronomical use cases and the breakthroughs needed in foundational AI research. Proposals are also encouraged to discuss the potential for those AI advances to benefit AI research more broadly or to impact application fields beyond astronomy.
Proposals are expected to convey a vision and approach that is appropriate for the scale of these Institutes and that produces transformative outcomes. Proposals should also describe how the Institutes will connect with the broader community to disseminate knowledge. The proposed structure, activities, and management of the Institutes to achieve these goals should be clearly described.
This theme is partially funded by the Simons Foundation. Each institute funded under this theme will receive two separate awards of up to $10M, one in the form of a cooperative agreement at NSF as described in this solicitation, and one award from SF in accordance with SF award procedures and consistent with applicable law. See Proposal Submission Guidelines for detailed procedures on how to structure project plans and budget submissions.
II.C. Institute Themes in GROUP 2 Awards anticipated in FY 2025:
Proposals are being solicited in the following high-priority areas for awards anticipated in FY 2025. Due dates listed for Group 2 apply for submissions to the themes in this group.
Theme 2: AI for Discovery in Materials Research
AI has the potential to revolutionize materials discovery by integrating first principles from materials science, physics, and chemistry with heterogenous multi-dimensional experimental and synthetic data streams to scale and accelerate development. AI can expand the types and properties of materials considered through augmentation of human intuition and by tailoring discoveries to address societal challenges, such as sustainability and those in emerging industries. A successful Materials AI Institute will transform the materials discovery landscape, enable new AI-based capabilities, and be responsive to societal challenges and industrial needs. Advances in AI have the potential to transform materials research in several ways. Some potential lines of research are provided below. This list is meant to stimulate thought about use-inspired research in the intersection of AI and materials, and is not prescriptive.
- Multi-modal data integration and dataset development: Data streams that describe material properties and behaviors based on different types of variables are ubiquitous in materials science and span different length/time scales and represent a vast set of modalities, such as simulation, synthesis experiments, and characterization experiments. Research in AI-enabled frameworks for materials research have the potential to catalyze the generation of insights by integration of heterogeneous multi-modal data streams across different length/time scales. In addition, tools and mechanisms are needed to accelerate the development of new data sets with appropriate diversity, speed, and volume to empower ground-breaking AI methods for targeted materials science problems.
- Foundational AI advances driven by materials research: Extending and tailoring AI methodologies to materials science and its unique data streams creates an opportunity to develop fundamentally new algorithmic and methodical frameworks in AI for materials discovery. From a bottom up (i.e., data-driven) direction, foundational AI advances in this field should fully capture and incorporate the unique characteristics and interactions evident in materials science. From a top down (i.e., knowledge-guided) perspective, the principles of materials science hold the potential to ground data-intensive operations in the rich mathematical complexity and multi-scale nature of the different physical and chemical relationships inherent to materials. The integration of both data-driven and knowledge-guided AI holds even greater potential to lead to significant advances in materials.
- First synthesis to synthesis at scale: Materials synthesis at scale is a major challenge in materials discovery. The precision and level of understanding required spans various complex phenomenological challenges. Research in the intersection of materials science and AI has the potential to sustainably synthesize materials at scale while mitigating the complex phenomenological challenges related to materials properties, materials processing for reliable synthesis, efficient characterization for measurement of relevant properties, and statistics-based understanding of various stochastic elements present in large-scale systems. Use-inspired AI research for materials science has the potential to revolutionize materials discovery and lead to new technologies that can address complex societal challenges.
- Human-augmented materials design: While AI holds great potential to automate discovery, it remains critical that this discovery be guided by and responsive to materials scientists who will collaborate with AI systems. The interfaces that mediate AI-driven materials research should be guided by principles for effective human-AI interaction and collaboration. Principled mechanisms of interaction between human experts and AI-augmented technology can change how materials designers think about design challenges and catalyze human creativity in new and unexpected ways—for example, shortening the requirements-design-synthesis-experiment cycle. Effective guidance from domain experts will also help ensure that the design of novel materials is conducted ethically and safely.
- Interpretable materials AI: As AI accelerates new advances and insights in materials science, human understanding of materials will be advanced even further to the extent that the operations of the system are interpretable by materials scientists. A system with transparent and explicable operations will have a higher potential to contribute to the discovery of new fundamental principles in materials science. For example, might successful AI materials models predict the essential ingredients of microscopic Hamiltonians for quantum materials? Can they provide clues to develop new concepts that expand theory and computation to enable humans to reach the same or better solutions? The more interpretable the materials AI system, the greater the opportunity for materials scientists to explore new frontiers of research in this area.
Proposals to this theme can address these or other relevant research areas in any combination. Proposals that promise to significantly advance both foundational AI and domains supported by the Division of Materials Research will be most responsive to this theme.
Intel Corporation is providing partial support for this institute theme.
Theme 3: Strengthening AI
In recent years, AI systems built with multilayer architectures with many parameters trained on massive datasets have become increasingly capable of producing useful and impressive outputs. These developments have found their way into large scale deployment, while their developers continue to strive for higher levels of generality, performance, and trustworthiness. Deep neural networks are increasingly effective in all manner of applications from game playing to consumer recommendations to autonomous driving. Generative models have advanced significantly in their ability to produce constructions in natural language, images and video, leading to applications that automatically edit content or even produce novel images and texts. Unfortunately, these systems are not always reliable and may not exhibit justification for their behavior that is understandable to the people who interact with them. In spite of their limitations, these capabilities are becoming ubiquitous in fielded systems of all kinds. This trend presents the opportunity and necessity to research ways in which AI technologies of all sorts can be improved and integrated toward systems that are reliable and aligned with human intentions and ethical considerations.
A lens through which one might view the developments of AI systems is in terms of a continuum from narrow to general (or with similar meaning/intent, weak to strong). “Narrow AI” excels at performing specific tasks for which it has been programmed (or trained). Over the past few decades, these systems have far exceeded expectations in an increasing number of feats previously thought to be dominated by human intelligence. Still, these systems can be brittle in the face of surprising situations, susceptible to manipulation or anti-machine strategies, and produce outputs that do not align with human expectations of truth or human values. In contrast, “strong AI” is the aspirational goal of creating intelligent systems that learn and think as adeptly as humans do. Strong AI is expected to be capable of performing effectively in a diverse range of problems subject to potentially contradictory priorities, gain new conceptual understanding from limited exposure to new domains, and adapt appropriately to the expectations of human users. While such systems in principle would be more robust to situations that challenge narrow AI, no examples of strong AI have been demonstrated to date. Research in AI can no longer distinguish approaches simplistically as either narrow or strong. But AI systems of the future will need to be strengthened if they are to be as robust as we would like and if we are to keep such technology well-aligned with society’s intended uses.
Theme Goals:
This theme promotes the development of next generation AI systems that have been strengthened to provide greater usefulness, consistency, and robustness by exhibiting both the high performance of narrow AI and the general adaptability of strong AI. Proposals must address the following goals, taking into account the full context of the motivation described above, while remaining relevant to the contemporaneous, rapid progress in the fielding of large, capable AI models. Institutes funded under this theme must lead advances in theory, methods, or integrative approaches that strengthen AI in all three of the goals listed below:
1) Grounding. Systems must understand the concepts they reason over and operate with. We refer to this capacity as grounding. Grounding allows AI system to demonstrate connection between its outputs and the abstract concepts that they operate with. It will also enable systems to understand their risks and limitations. Such an improved conceptual understanding should also lead to robust AI that adapt gracefully and quickly to new domains, is robust to surprise, and resists malicious manipulation.
2) Instructiblity. Taking advantage of this firmer understanding, strengthened AI must be “instructible”. This means that systems can be proven experimentally to change their behavior appropriately in response to explicit feedback provided by even non-expert users. Related is how such instructible systems might invert the mechanisms behind this principle to implement more effective and trustworthy assistance to humans (e.g., in instruction, tutoring, and training) or in explaining their understanding or recommendations.
3) Alignment. Strengthened AI systems must be judged by how well their operations align with expectations of objective truths in a domain and correspond to societal expectations and human intentions in their operations. Proposals must include rigorous plans to evaluate this capability.
Any AI approaches that contribute to these three goals are in scope. This might include but is not limited to neuro-symbolic approaches, hybrid integrated architectures, or multi-representational learning methodologies. Proposals that rely mainly on continuing growth of data-driven models and their access to more data are not responsive to the three goals above unless accompanied by a compelling basis of confidence that true breakthroughs in those areas can be projected and evaluated. Technical approaches that integrate and process data from multiple sources and in diverse modalities as appropriate to the domain(s) of application are likely to serve the goals of this theme well. Institute concepts whose technical plans do not promise to advance all three of the above goals are unlikely to be competitive in this theme.
Use-Inspired research focus:
Any use-inspired research context may be the basis for an institute proposal to this theme. Institute research plans that strengthen AI in such a way that the techniques are generally applicable to diverse application domains are likely to serve the goals of this theme well. Proposers are encouraged to consider domains of broad significance to collective wellbeing. Examples of such domains include but are not limited to:
- Protecting the environment to ensure human safety and to safeguard natural resources and wildlife.
- Health and wellbeing, including various non-clinical aspects of physical and/or mental health.
- Civic and public good, for example optimization of infrastructure, responsible resource allocation, delivery of public services.
- Improving human flourishing, for example, reducing hunger or coordinating humanitarian assistance.
- AI advances that enable new discoveries in science, mathematics, or engineering.
- Enhancing the economic security of the U.S. through modernization of e.g., manufacturing, infrastructure, or communications.
Multiple awards are anticipated in this theme. Capital One is providing contributions for the partial funding of an award in this theme. Agency partners listed on this solicitation (OUSD (R&E) and NIST) may also elect to provide contributions to NSF for the funding of Institutes under this theme. Submitters to this theme may submit a supplementary document to indicate relevance of the proposed Institute to one or more partners. Submitters may also stipulate that the proposed Institute should not be considered for funding from specific partner(s) by uploading a single copy document. See Proposal Preparation Instructions.
Visit our Institutionally Limited Submission webpage for more updates and other announcements.