Microsoft Support for the National AI Research Resource Pilot

We’re excited by the breakthroughs we’ve already seen leveraging state of the art models for health (opens in new tab), molecular (opens in new tab)and materials discovery (opens in new tab), and earth studies (opens in new tab). Building on this momentum, we’re eager to share the tools we have for advancing science with AI to support research projects relating to all NAIRR pilot focus areas (opens in new tab). We’re inspired by research that aims to tackle grand challenge problems by using AI to accelerate science and discovery across domains, and to drive progress in areas like health and life sciences, globally equitable systems, sustainability, AI evaluation, agentic systems, and multimodal studies. We provide details below. Please note that larger, custom GPU configurations, including with InfiniBand (opens in new tab), will be available for select proposals to develop world-class AI for science.

AI for Accelerating Science and Discovery

Research using AI to design experiments and advance the scientific process in and across domains including materials sciences, chemistry, physical sciences, weather and forest fire prediction, agriculture, and other science-related fields.

Health and life sciences

Research aimed at building cellular simulations, protein design pipelines, and related topics such as AI and biosecurity.
Research to address neurodegenerative diseases including ALS and Alzheimer’s Disease. In addition to compute and AI model access, researchers can access data sets: Neuromine Data Portal (opens in new tab) (clinical data on 1200+ participants from both control and neurodegenerative disease populations) and through the Alzheimer’s Disease Data Initiative (opens in new tab).

Globally equitable systems

Research to advance globally equitable AI systems and technologies.
Research to adapt AI to the linguistic and cultural nuances of underrepresented languages.
Research that advances efficient tokenization mechanisms for non-Latin scripts.
Research that advances responsible and equitable data collection practices for underrepresented languages.
Research that advances safety post-training strategies for equitable improvements across languages.
Research that identifies languages that are uniquely effective at circumventing safety systems (e.g., low resource languages, morpho-syntactically complex languages, non-Latin script languages).

Sustainability

Research to quantify, reduce or remove carbon emissions, water-related issues, ecosystem problems or waste through new or optimized molecules/materials, systems and processes.

AI evaluation

Research to revolutionize verification and evaluation for AI systems, comparative analyses of frontier models as they evolve, and studies in mechanistic interpretability.

Agentic systems

Research on novel applications of agentic systems, such as in healthcare, e.g., tools to assist a Multi-disciplinary Tumor Board.
Research on ways agentic systems help address global challenges like scientific discovery and climate change.
Research on how we might we assure that outcomes driven by multi-agent systems are safe, correct, and match human intent.
Research on how systems to verify AI agent identity and authorizations granted can complement existing synthetic media authentication approaches.

Studies of multimodal models

Research directions in multi-modality, molecular dynamics, and quantum chemistry.