Meet Themis, The AI That Knows How to Shut Up

We have all had that one egotistical friend, you know, the one who throws “facts” out of their mouth with a certain degree of arrogance, only to discover later they were completely off. Now, picture the same enthusiasm, but it is driving the AI that is helping create a cancer treatment plan. Not so funny anymore? That is the challenge Themis AI is addressing chiefly. 

A product from a team of MIT researchers, Themis AI is used to “to quantify artificial intelligence model uncertainty and address knowledge gaps.” Founded in 2021, the startup is learning AI, a skill that most humans find difficult, admitting it does not know. 

Why does intelligent AI still pretend?

Most AI systems, including names like X’s Grok or OpenAI’s ChatGPT, do not just make errors; they make them with certainty. It is a known problem of machine learning called as “hallucination,” whereby the AI model creates erroneous or made-up data and presents it as if it is the only truth that exists. Results can be dangerous. You may start believing what is not true. In medicine, it can have serious repercussions. Themis AI’s solution? A platform called Capsa that acts like a built-in gut check. Instead of letting the model go on with its shaky guesses, Capsa steps in and says, “Hold up. This looks fishy.”

How does it work?

Capsa is intended to integrate with the majority of AI systems. It does not require altering the design of how the AI was created. It simply plugs in and keeps a watch. And the best part is that it is compatible with any ML model and at any stage of development.  If it detects indicators that the model is not sure or it is working with dubious data, it reports it.

Themis has already assisted telecom operators in avoiding costly planning errors and assisted oil and gas companies in analysing seismic data without interpreting it wrong. It has also seen collaboration with developing chatbots that understand when to shut up rather than coming up with gibberish. 

“Capsa, our proprietary technology, is built to be compatible with any ML model, seamlessly working in a matter of seconds at any stage of development,” reads the description of the technology on the website of Themis. 

Real-life application of Themis technology 

The technology has been used in drug discovery with an approach that assisted AI in determining if its predictions were founded on actual evidence or mere hypotheses. They have also focused on edge devices- the tiny, low-power computers in everything from medical wearables to drones. Even these light-as-air AIs can determine when they are in over their heads with Themis technology integrated. 

AI is doing increasingly large and dangerous jobs each year, from diagnosing illnesses to running infrastructure. But for all its capability, something has lacked- an awareness of its limitations. That is what Themis AI offers. Not more smartness- just better judgment. A bit of humility, as it happens, may be the very thing AI needs to win our trust.

Read more: Apple’s iPhone 17 and iPhone 17 Air Could Get Smoother Screens But