Collaborative Coding, Better Scaling, Health Tracking: HAI Awards $2.17M to Innovative Research
Seed grants will fund 29 research teams pursuing novel research ideas across disciplines.
In the eighth year of the Stanford HAI Seed Research Grant program, 272 teams competed for funding to support ambitious AI projects across diverse fields ranging from science and health to religion, social sciences, and education. Earlier this month, 29 applicants learned they will receive grants to further their work.
For the 2025 cohort of seed grant recipients, global media company Holtzbrinck provided support for two of the award winners, while the Stanford Center for Digital Health will co-fund three projects with HAI.
“This year was our most competitive one yet for seed research grants,” said HAI Co-Director James Landay. “We received a record number of compelling proposals across disciplines. We’re excited to watch the winning teams work toward initial results that will drive human-centered impact.”
Since its founding, Stanford HAI has provided approximately $16 million in seed grants. Supported initially by Steve and Roberta Denning and for four years by Dalio Philanthropies, seed grants are given for speculative ideas at the frontier of AI research. Funding aligns with the institute’s three research imperatives: human impact, augmenting human capabilities, and intelligence. See all the recipients here and read about three of the winners below.
Scaling Laws for the Social Sciences
Today, social scientists use large language models (LLMs) for basic research tasks, including literature review, data labeling, survey design, programming, and writing. But they don’t yet trust these general-purpose models to simulate human opinions and behaviors. A team of sociology and NLP scholars, led by Professor of Sociology David Grusky, is investigating whether advancements in frontier models will translate to improved utility in the social science domain.
Using the concept of predictable scaling laws – a method that helps computer scientists anticipate an LLM’s future capabilities as it scales up – the researchers will explore whether the same dynamics observed in general-purpose LLMs could apply in sociology contexts. Unique factors in the sociology domain, such as cultural differences or gaps in population groups represented in a dataset, could cause the scaling laws to be unreliable or even inverse to the known patterns of frontier LLMs. If the project succeeds, researchers in the field will be able to better gauge when they can use an LLM with confidence and when a domain-specific model may be necessary.
Detecting Pediatric Pneumonia Spikes in Ethiopia
In Ethiopia, pneumonia is one of the leading causes of death for children under the age of five. Health facilities often are caught off-guard during seasonal spikes in the disease. Working with the Ministry of Health in Ethiopia, John Openshaw, assistant professor in the Stanford School of Medicine, is developing an early warning system for pneumonia outbreaks, using AI to anticipate surges in cases and allowing health officials to intervene proactively with oxygen, medicines, and staff. Drawing on multiple data sources from clinics, call centers, and weather models, the system will predict when and where outbreaks are most likely to occur. The scholars aim to save lives by developing modern tools for the country’s new national “situation room” for real-time health monitoring.
Collaborative Code Generation for Complex Physical Simulations
Madeleine Udell, assistant professor of management science and engineering, sees a need for a human-centered AI framework to help scientists create specialized software called Partial-Differential Equation (PDE) solvers. Solvers are essential for running complex physical simulations like climate modeling. Improving on current “brute-force” approaches with LLMs, the new method will leverage a scientist-in-the-loop to reduce compute costs and increase reliability of the software. By transforming AI from a simple code generator into a collaborative partner, this team could assist researchers, accelerate discovery, and make high-performance computing more accessible to a broader scientific community.
To see snapshots of all this year’s Seed Research Grant projects, visit our grants page.
Related News
AI Companions May Worsen Loneliness for Vulnerable Users, Stanford Study Finds
Stanford research finds that users with limited social networks who seek emotional support from AI companions experience lower well-being.
Stanford research finds that users with limited social networks who seek emotional support from AI companions experience lower well-being.
The Complexities of Governing Mental Health AI
Policymakers, academics, healthcare providers, AI developers, and patient advocates convened by Stanford HAI identify critical gaps in how we regulate AI tools used for therapy and emotional support.
Policymakers, academics, healthcare providers, AI developers, and patient advocates convened by Stanford HAI identify critical gaps in how we regulate AI tools used for therapy and emotional support.
With increased use of chatbots in mental health contexts, AI developers now rely on human experts to evaluate AI’s responses for “safety” – but experts rarely agree on what’s safe.
With increased use of chatbots in mental health contexts, AI developers now rely on human experts to evaluate AI’s responses for “safety” – but experts rarely agree on what’s safe.