The classical approach to artificial intelligence consists in analysing human intelligent behavior, in developing formal models explaining this behavior, and in coding these models into computer systems. We follow a radical bottom-up approach. We believe that artificial intelligence is only credible and feasable when embodied artificial agents autonomously bootstrap their competence in interaction with physical environments. We believe that embodiment and language are both crucial keys for enabling and speeding up this bootstrapping.

The Talking Heads experiment is a novel Turing test. Because humans can play the role of agents, we can monitor whether people believe they play against human or artificial agents. The experiment goes beyond the Turing test because the artificial agents are grounded through visual sensors, must exhibit learning capacity, and language creativity