The Virtual Teacher paradigm, a version of the Human Dynamic Clamp (HDC), is introduced
into studies of learning patterns of inter-personal coordination. Combining mathematical
modeling and experimentation, we investigate how the HDC may be used as a Virtual
Teacher (VT) to help humans co-produce and internalize new inter-personal coordination
pattern(s). Human learners produced rhythmic finger movements whilst observing a computer-
driven avatar, animated by dynamic equations stemming from the well-established
Haken-Kelso-Bunz (1985) and Schöner-Kelso (1988) models of coordination. We demonstrate
that the VT is successful in shifting the pattern co-produced by the VT-human system
toward any value (Experiment 1) and that the VT can help humans learn unstable relative
phasing patterns (Experiment 2). Using transfer entropy, we find that information flow from
one partner to the other increases when VT-human coordination loses stability. This suggests
that variable joint performance may actually facilitate interaction, and in the long run
learning. VT appears to be a promising tool for exploring basic learning processes involved
in social interaction, unraveling the dynamics of information flow between interacting partners,
and providing possible rehabilitation opportunities.