The AI tools we build, have, undoubtedly, an impact on our cognition, and more specifically, attention, memory, engagement, among many other cognitive and affective states. We explore ways to better understand what happens to us when we engage with these tools, but also if we can adjust the AI interaction to match better the current state of its user.
We do:
(1) empirical studies to understand how brain activity is affected by LLM use, for example, check our project “Your Brain on ChatGPT”; where we measured and analyzed functional brain connectivity of students engaging in the educational task of writing an essay using an LLM.
(2) design tools and technologies to detect cognitive states from EEG, fNIRS and physiological sensors to dynamically adapt AI interaction, for example, check our project NeuroChat, a neuroadaptive AI tutor that combines genAI with real-time EEG data to create a personalized, dynamic learning experience. Unlike traditional conversational AI, which relies on explicit user input, NeuroChat infers engagement levels directly from brain activity, modifying its responses to match the user's attention level, cognitive state, and learning preferences. When engagement drops, NeuroChat modifies its conversational strategy — adjusting response complexity, pacing, and level of interactivity — to re-engage the learner and maintain an optimal cognitive state.
(3) Design and open-source privacy-first tools for the community to explore and build their our real-time proactive agentic systems, using foundation EXG models, text and image embeddings models, by directly leveraging biophysical and brain signals, recorded using Brain-Computer Interfaces (BCIs). Check our project NeuroSkill.