Cedars-Sinai Software Helps Build, Monitor AI for Research
Three New, Open-Access Software Platforms Design Agentic AI Systems and Fine-Tune Them After Creation
Cedars-Sinai investigators created three software packages that build, modify and monitor agentic AI systems and produce more transparent research results.
The agentic AI packages, called ASAREE, Memstara and GENASYS, are available online and free of charge. The hope, investigators say, is for members of the public to use these tools, learn from them and repurpose them.
“I view this accessibility as integral to our mission at Cedars-Sinai, which is dedicated to improving the human condition,” said Jason Moore, PhD, chair of the Department of Computational Biomedicine at Cedars-Sinai. “Our team also hopes to learn a lot from the community. As they engage with the software, they will give us ideas that help us improve and extend it.”
Moore said that commercial agentic AI systems often lack transparency, which is a special concern for scientific endeavors.
“Agentic AI systems are complicated,” Moore said. “They must choose and organize AI agents—programs that can autonomously complete complex tasks—to do the work, and then also critique the work, perform quality control and evaluate ethics.”
For example, Moore said, if you give the system a problem to solve, and it produces a result, but you have no idea how it got there, how can you trust the result?
“Our software aims to combat this lack of transparency,” Moore said.
The first two software packages help design agentic AI systems.
ASAREE is short for Analytical Sandbox for Agentic Research, Engineering, and Experimentation. Investigators can direct ASAREE to investigate a hypothesis such as: Would a particular type of software improve the reliability of the workflow? The system tests whether the hypothesis is correct by trying out various options.
Memstara draws on existing knowledge about relationships among genes, diseases, biomarkers and other factors to generate original, biologically informed hypotheses, much as a human scientist does. The hypotheses can then be tested using real-world biomedical data.
The third software package fine-tunes the agentic AI system after it is created. GENASYS is short for Genetic Evolution of Networked Agent Systems. It automatically modifies agentic AI workflows, repeatedly evaluating alternative configurations and selecting those that perform best for a given biomedical task.
Moore said that although ASAREE and GENASYS were developed with biomedicine in mind, they could be applied to agentic AI systems for studies in economics, engineering, meteorology and other scientific disciplines.
He said Memstara at present has a more limited potential because it incorporates a specific knowledge base, but it could be adapted to work with different knowledge bases.
Cedars-Sinai Health Sciences University is advancing groundbreaking research and educating future leaders in medicine, biomedical sciences and allied health sciences. Learn more about the university.