Social AI: Turn-Taking Predictor (2D)

A group of AI avatars sits around a virtual table exchanging the conversational floor. This top-down 2D simulator renders the mechanism a real social-signal-processing model uses to decide who speaks next: while one avatar holds the floor, every listener accumulates a live "anticipatory gaze" score toward whichever candidate looks most likely to take over, and that score is combined with a silence-fairness term into a softmax distribution the moment the turn ends. Tune the gaze weight, fairness weight and softmax temperature to see the group swing between a single dominant speaker and a perfectly round-robin conversation, and watch the live dominance (Gini) index and selection-entropy readouts track how unequal or how contested the floor really is. Drag to pan the table view and scroll to zoom in on a speaker.