In an era where AI systems increasingly make critical decisions across industries, the ability to recognize uncertainty is becoming essential. On June 3, 2025, Themis AI, an MIT spinout, unveiled breakthrough technology that teaches AI models to acknowledge what they don't know—a capability that could transform AI reliability in high-stakes applications.
Themis AI's Capsa platform works by surgically updating any machine learning model's architecture to enable uncertainty quantification. "The idea is to take a model, wrap it in Capsa, identify the uncertainties and failure modes of the model, and then enhance the model," explains Themis AI co-founder and MIT Professor Daniela Rus, who also directs MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL).
Founded in 2021 by Rus along with Alexander Amini (MIT '17, SM '18, PhD '22) and Elaheh Ahmadi (MIT '20, MEng '21), Themis AI builds on over five years of foundational research. The company's mission addresses a fundamental problem: AI systems like large language models often provide plausible-sounding answers without revealing gaps in their knowledge or areas of uncertainty.
The implications for high-risk domains are significant. In autonomous vehicles, research shows that integrating Themis AI's uncertainty estimation algorithms led to 16 times fewer collisions and a 93% reduction in automated requests for human intervention. For healthcare and pharmaceutical applications, Capsa helps identify when AI predictions are backed by evidence versus mere speculation, potentially accelerating drug discovery while reducing risks.
"We want to enable AI in the highest-stakes applications of every industry," says Amini. "We've all seen examples of AI hallucinating or making mistakes. As AI is deployed more broadly, those mistakes could lead to devastating consequences. Themis makes it possible that any AI can forecast and predict its own failures, before they happen."
The technology is already being implemented across various industries. Many companies building large language models are using Capsa to enable their models to quantify uncertainty for each output, allowing for more reliable question answering and flagging unreliable outputs. Themis AI is also working with semiconductor companies to improve smaller AI models that run on phones or embedded systems, achieving both low latency and high quality.