Dual-EEG of joint finger tapping: what can two interacting brains teach us about social interaction?

The underlying neural mechanisms of real-time social interactions remain largely unknown. Only a small number of recent studies have explored what goes on in brains of two people simultaneously as they interact. The question still remains whether such quantification can better reveal the neural signatures of social cognition.

In our study, we wanted to address this question by quantifying whether we gain more information about the interaction from the two brains. We measured dual-EEG from pairs of participants as they engaged in an interactive finger-tapping task. They were asked to synchronize with an auditory signal coming from the other member of the pair or the computer.

They experienced two conditions: an interactive ‘coupled’ condition, each receiving feedback of the other person’s tapping; and an ‘uncoupled’ computer-control condition, each receiving feedback from a non-responsive computer. Time-frequency analysis revealed a left-motor and right-frontal suppression at 10 Hz during task execution, when carrying the task out interactively in contrast with the uncoupled computer-driven task.

We used machine-learning approaches to identify the brain signals driving the interaction. The raw-power at 10 Hz during tapping emanating from electrodes of member one and member two were used as features. We combined data from both participants in each pair, and applied logistic regression using feature selection in order to classify the two conditions.

The first seven (frontal) electrodes consistently emerged as good classifiers, with 85-99% accuracy. There was a tendency for one member’s frontal electrodes to drive the classifier over the other’s, which predicted the leader of the interaction in 8/9 pairs. This study reveals new neural mechanisms underlying two-person interactions.

It also shows how analyzing two interacting brains can give better classification of behaviour; and hence that the whole of two brains is indeed better than the sum of its parts, at disentangling neural signatures of interaction.