This paper describes an artificial-intelligence (AI) system built to capture what people are thinking, feeling, and sensing in specific moments of their daily life. The work tries to bridge a common trade-off in psychology: simple experience-sampling tools can reach many people but force answers into fixed categories, while deep, open-ended interviews reveal richer detail but need scarce expert interviewers and so stay small. The authors aim to scale the deeper approach without losing its care for precise moment-by-moment description.

The researchers built an AI interviewer grounded in Descriptive Experience Sampling (DES). DES is a method that catches random “beeped” moments in daily life and uses careful, non-leading interviews to help a person describe exactly what was present in that instant. The team derived the AI from the full corpus of DES transcripts and refined it together with Russell T. Hurlburt, the originator of DES. The interviewer appraises each participant message across eleven quality dimensions, keeps a conservative running record of what has been established about the moment (an “experience registry”), and always prioritizes locating the precise moment in time before asking about experience.

At a high level the system uses a framework-driven design rather than simple imitation. The AI follows a four-stage reasoning architecture: Message Appraisal, Experience Modeling, Strategy Selection, and Query Composition. At each turn it scores the interview state, selects a stage-appropriate, non-leading question in the participant’s own words, and proceeds only when temporal grounding is adequate. The interviewer runs inside an application called Introscope, which delivers the beeps, conducts the interviews, and offers a study platform with shareable links and tools for researchers to review sampled experiences. The project includes a demonstration video (https://introscope.mpib-berlin.mpg.de/video).