Cognitive Digital Twin CognitiveTwin: Teaching AI to Clone How You Think, Not Just What You Know

CognitiveTwin, released by the Human-Computer Interaction Lab at the Swiss Federal Institute of Technology in Lausanne, is a system that departs from conventional personalization. Traditional recommenders record what users do and infer preferences from that behavior. CognitiveTwin focuses on observing the decision-making process itself and builds a digital clone of how a person thinks based on those process data.

The project grew out of a long-running EPFL research program. The researchers observed that different choices users make in the same situation often reflect underlying cognitive styles. Some lean toward risk aversion, others toward novelty. Conventional AI systems struggle to capture this kind of "how" information and only record "what."

At the core of the system is a thought-process recorder. Through a desktop camera, wearable sensors, and screen interaction logs, the system continuously captures data on how a user reaches a decision. When the user faces a choice, such as picking between two films, the system simultaneously tracks eye movements, heart rate, hover time, and the final selection. Those data train a personal thinking-style prediction model.

The pilot study ran in 240 households across Switzerland and the Netherlands, with each household using CognitiveTwin for six months. Results show the system predicted user preferences with 87 percent accuracy, compared with 62 percent for traditional recommenders. The gap widens further when predicting decisions over time. CognitiveTwin can forecast choices a user would make in future scenarios days ahead.

In terms of applications, CognitiveTwin's potential reach is broad. Health coaches could use it to identify cognitive weak points when users face exercise temptations. Financial advisors could predict how clients will actually react to market swings. Education platforms could spot the thinking patterns students bring to hard problems and tailor guidance accordingly.

The project has triggered intense debate over psychological privacy. The European Data Protection Board has opened a dedicated review, asking the EPFL team to demonstrate that the data collected cannot leak subconscious information. Aude Billard, the professor leading the team, responded that all data is processed on the user's device and the thinking prediction model contains no interpretable psychological profile.

For commercialization, the EPFL team has spun out a company called MindClone SA. The product will be offered to European users as a subscription starting in the second half of 2026, priced at 49 euros per month. MindClone SA has raised 28 million dollars in Series A funding led by Index Ventures, valuing the company at 180 million dollars.

Critics point to manipulation risk as the biggest concern. Advertisers and political campaigns could potentially exploit detailed psychological profiles for targeted persuasion. Billard acknowledged the risk and pledged that the company will not sell any profile that identifies individual psychological traits, and all data use will be limited to purposes authorized by the user.

The project has received 12 million Swiss francs in funding from the Swiss National Science Foundation. EPFL plans to extend the CognitiveTwin pilot to all Francophone universities in Switzerland by 2027.