Pathways to AI-enabled research
CZI and Sloan Foundation announce eight new grants to transform scientific research practices through AI applications
The development of artificial intelligence (AI) technologies experienced a dramatic acceleration in late 2022 with the widespread adoption of large language models and conversational agents. While AI applications are now integrated with numerous productivity tools and tasks in daily life, the development of AI tools to support scientific research is at a nascent stage. Over the last two decades, research software has reshaped scientific work and computational reproducibility. Applied thoughtfully, we believe artificial intelligence has the power to catalyze a new generation of tools for science and ways of doing science, from managing large and diverse datasets, to enabling insights from vast amounts of published literature, reducing barriers to participation and access to scientific discourse, and generating new hypotheses or providing novel insights into specific research questions.
At CZI, we believe AI will profoundly transform scientific discovery in the life sciences. We’ve built one of the world’s largest AI clusters for nonprofit scientific research and our Technology teams across the CZI Science ecosystem are working towards building AI-powered virtual cell models to help us better understand disease at the cellular level. Last month, we released models that are a first step toward this goal. Like everything else we build at CZI, these models are openly available to researchers worldwide. We’re also opening up to the wider scientific community our compute cluster, which is optimized for AI and machine learning training at scale to power new approaches to biological discovery. Recently, we started inviting proposals from researchers working at nonprofits in the U.S. to build large-scale AI/ML models that cannot be created with conventional university resources.
In addition to building models and improving access to compute resources, we have been exploring opportunities to directly support the development and adoption of transformative AI tools. In May 2024, the Open Science Program at the Chan Zuckerberg Initiative (CZI) and Alfred P. Sloan Foundation Technology Program (Sloan) shared a joint request for community input about tangible tools and applications for AI-enabled research practices, with an eye toward principles of open science. We received a breadth of responses about applications of AI to scientific productivity at various stages of development and maturity and focused on different types of use cases and domains. These responses helped us identify some key areas for investments and are proud to announce eight new grants and partnerships:
- Bernadette Boscoe of Southern Oregon University to develop a RAG-LLM tool that allows research groups to preserve tacit knowledge. (Sloan)
- Matthew Akamatsu from University of Washington to optimize and facilitate uptake of AI-enabled Discourse Graphs for scientific knowledge organization and lab collaborations. (CZI)
- Neuromatch’s Nicholas Halper to develop AI matching algorithms to support scientific collaboration. (Sloan)
- Abel Brodeur at the University of Ottawa to develop an LLM-based reviewer for academic code peer review. (Sloan)
- Emmanuel Iarussi and Viviana Siless of Universidad Torcuato Di Tella in Argentina to create VisDecode, an AI tool capable of automatically interpreting and providing feedback to enhance scientific plots. (Sloan)
- Penn State’s Ting-Hao ‘Kenneth’ Huang and C. Lee Giles to build and deploy a web-based system for composing better captions for scientific figures. (Sloan)
- Hugo Stephenson at the Center for Biomedical Research Transparency (CBMRT) to support uptake of Null Research Compass, an AI-enabled assistant for exploring and publishing negative results as preprints. (CZI)
- Andrew McCallum of OpenReview to integrate AI capabilities in their open source review platform to enhance the quality, reproducibility, transparency, and fairness of the scientific peer review process. (CZI)
The tools funded above span all stages of research, from developing ideas and collaboration through review and publication. They join previously existing and ongoing projects at CZI and Sloan to improve the scientific process for biomedical research and beyond.
We are learning about AI’s potential applications for scientific productivity, not only through these grants but also through research we commissioned last year. A partnership with Ithaka S+R surveyed a large sample of academics to assess attitudes towards and use of generative AI in research. This study found current generative AI usage in biomedical research is focused on exploring ideas (e. g., discovering relevant research and extracting knowledge) and writing (code and drafting/editing text). Moreover, over half of biomedical researchers expressed strong interest in generative AI tools specific for biomedical research. A collaboration with Milton Pividori (University of Colorado — Anschutz Medical Campus) provided insight into the effectiveness of current AI tools for engaging scientists, and suggested generative AI is most useful as a starting point for writing scientific text and code, as other usage is less reliable and may contain more risk. The findings from these projects, as well as ongoing conversations with other stakeholders, continue to inform new directions we’re pursuing on the use of AI in support of research.
Through these grants and partnerships, we’re learning how AI can be deployed responsibly to enhance research productivity, sustain open research practices, and transform scientific work while controlling for the potential risks and downsides of its applications — concerns that continue to be discussed in the research community. We will share more about what we learn as well as new opportunities in this space as they emerge.
