Predicting the conscious experience of other people

Here is a fascinating abstract for one of the lectures, by Geraint Rees and colleagues, at the upcoming 15th annual meeting of the Association for the Scientific Study of Consciousness. Mind reading with machines may be closer than we think.

There has been considerable interest in using multivariate decoding techniques applied to fMRI signals in order to decode the contents of consciousness. The use of such signals has inherent disadvantages due to the delay of the hemodynamic response. Moreover to date it has not been shown possible to generalize the decoding of brain signals from one individual to another. This limits the potential utility of such approaches. Here we used a different approach that circumvented these difficulties by using magnetoencephalographic (MEG) signals to decode the contents of consciousness, and to test whether such correlates generalized reliably across individuals. We recorded the MEG of 8 healthy participants while they viewed an intermittently presented binocular rivalry stimulus consisting of a face and a grating. Using a leave-one-out cross-validation procedure, we trained support vector machines on the MEG signals to decode the rivalry percept. Decoding was significantly better than chance in all participants. We then tested whether a support vector machine trained on MEG signals from one participant could successfully decode the rivalry percept of another. Again, decoding accuracy was significantly better than chance. These findings demonstrate that it is possible to decode perception independently of physical stimulation using MEG signals in near real time in a way that generalizes across individuals. Our findings indicate that certain neural mechanisms universally covary with the contents of visual consciousness, and mark a potentially important step in the design of devices for decoding the contents of consciousness in individuals unable to report their experience behaviorally.