The goal of this studio is to explore and discuss the potential benefits and the open challenges of giving a physical form to Artificial Intelligence (AI), towards the definition of tangible, or graspable AI. In this studio, we want to carry out a hands-on exploration of the potential of tangible interaction to help people grasp how transformer-based AI models, such as Large Language Models (LLMs), make predictions. Starting from a basic understanding of the core mechanisms of these models – input/output embedding, positional encoding, self-attention, and multi-head attention, we will explore how tangible interaction properties, such as data physicalization and tangible manipulation, may help designing graspable interfaces for facilitating the understanding and control of LLMs. Fostering AI transparency and human agency, we believe that tangible interaction may help increase the trustworthiness of Large Language Models. Through brainstorming and prototyping activities, we aim at identifying effective tangible representations and manipulation techniques that will help users grasp and trust LLMs.
