Stanford HAI Funds Groundbreaking AI Research Projects

Stanford HAI is pleased to announce the recipients of the latest Seed Research Grant awards. The seventh cohort of this program represents all seven Stanford schools and more than 31 academic departments. In the coming months, these teams of scholars will explore the possibilities of AI technology in diverse fields, including organizational culture, AI for science, cybersecurity, neuroscience, and robotics.

Supported initially by Steve and Roberta Denning and for four years by Dalio Philanthropies,

seed grants are awarded to speculative ideas at the frontier of AI research. Funding aligns with the institute’s three research imperatives: human impact, augmenting human capabilities, and intelligence. And since HAI receives such a wide range of proposals each year, the program has become a recognized bellwether for the future of AI. 

New for the 2024/25 program, Stanford HAI awarded an additional $10,000 for select projects that have a public policy component. The funding will support these research teams in conducting policy activities, such as holding workshops and developing briefs that will amplify their research findings among policy audiences.

Also for the first time, Stanford HAI is co-funding two projects with the Center for Digital Health: “Empowering Patients with AI: LLM-Based Summarization of Clinical Reports for Patient Experience Improvement” and “Shuno: Accessible and Joyful Hearing Health Monitoring.”

This year’s funded projects include:

Democratizing Structural Biology and Protein Design with Multimodal AI

Possu Huang, assistant professor of bioengineering, aims to enhance our understanding of protein structures through an interdisciplinary, human-centered AI approach. By leveraging large language models (LLMs), bridging the gap between text and protein modalities, and integrating their lab's protein generative models and evaluation metrics, this team of researchers aims to create a user-friendly, natural-language interface for interacting with protein structures and design tools. This work will facilitate a broad-scale comparison of methods and the development of new ways to design proteins using computer simulations. 

(Re)presenting the H in HAI: A Disability Justice Evaluation of AI Detection Tools

Alfredo Artiles, a professor of education and an expert on disability and inclusive education, is leading a two-part study to improve research methods for the protection of disability rights with respect to AI. The first phase involves a literature review of how AI researchers conceptualize the human figure, focusing specifically on (dis)ability representation in natural language processing research. In the second phase, Artiles and colleagues will evaluate the behavior of AI detectors on text written by neurotypical college students compared with essays written by neurodivergent students. Common AI detection tools will be used to evaluate whether the essays were AI-generated or composed by humans. Finally, the team will use its findings to create a toolkit to assist individuals in pursuing recourse and reparations for infringement of their rights under the Americans with Disabilities Act.

AI-Driven Drought Indicators for Sustainable Urban Water Policy

Led by Sarah Fletcher, an assistant professor of civil and environmental engineering and a center fellow at the Woods Institute for the Environment, this team plans to develop an AI-driven approach to mitigate the increasing risk of drought in urban environments. Drought planning presents a challenge because current indicators of drought are developed ad hoc and not tested rigorously or updated frequently. Moreover, there are many variables to consider and many available scales to measure them. To solve this problem, the researchers will integrate a water management simulation model into a reinforcement-learning framework, guided by stakeholder engagement.

In-the-Wild Multimodal Data Collection for Compliant Assistive Robot Skill Acquisition

As the world population ages and demand for daily care increases, traditional support systems will become inadequate to meet the needs of elderly populations. Assistive household robots may help address this gap, particularly ones capable of imitation learning that leverages human demonstrations. However, deploying AI technologies in robotic systems faces a fundamental challenge: data scarcity. Mark Cutkosky, professor of mechanical engineering, has assembled a team that’s developing a low-cost, scalable solution to augment standard Universal Manipulation Interface (UMI) records with tactile data to provide multimodal information that can enhance the precision, robustness, and safety of robotic systems for the home.

Since its founding, Stanford HAI has provided approximately $14 million in seed grants that have attracted an additional $25 million in external funding. “This multiplier effect validates the program’s ability to identify promising early-stage research in AI,” said HAI Director of Research Programs Vanessa Parli. 

See snapshots of all this year’s Seed Research Grant projects.