Artificial intelligence systems like ChatGPT have become remarkably adept at generating plausible-sounding responses to virtually any query. However, these systems often fail to acknowledge their limitations or express uncertainty when appropriate—a shortcoming that poses significant risks as AI adoption accelerates across critical sectors.
Themis AI, founded in 2021 by MIT researchers Daniela Rus, Alexander Amini, and Elaheh Ahmadi, has developed a solution to this problem. Their Capsa platform can be integrated with any machine learning model to detect and correct unreliable outputs in seconds.
"We've all seen examples of AI hallucinating or making mistakes," explains Amini, co-founder of Themis AI. "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 works by modifying AI models to recognize patterns in their data processing that indicate ambiguity, incompleteness, or bias. This enables the models to quantify their own uncertainty for each output and flag potential errors. Implementation is remarkably straightforward—requiring just a few lines of code to transform an existing model into an uncertainty-aware variant.
Capsa is already being applied across multiple industries. Pharmaceutical companies are using it to improve AI models that identify drug candidates and predict clinical trial performance. Large language model developers are implementing it to enable more reliable question answering and flag unreliable outputs. Themis AI is also in discussions with semiconductor companies to enhance AI solutions for edge computing environments.
"By automatically quantifying aleatoric and epistemic uncertainty, Capsa is a transformative technology that enables model errors to be caught before they become costly mistakes," says Rus, who also serves as director of the MIT Computer Science and Artificial Intelligence Laboratory. "It expands the uses of AI systems in applications where safety and reliability are key, such as robotics and autonomous driving."
As AI continues to evolve and permeate critical sectors, solutions like Capsa will be instrumental in building more trustworthy systems that acknowledge their limitations—a crucial step toward responsible AI deployment in high-stakes environments.