Friday, December 18, 2009

Actual Mind Reading is Happening!

BCIs Using fMRI


Paper: 'Predicting human brain activity associated with the meanings of nouns.'

Professors Marcel Just and Tom Mitchell at Carnagie Mellon demonstrate what is perhaps the most advanced BCI to date. Using fMRI and artificial intelligence, they are able to predict thoughts.

In their study, they catalogued fMRI images of brain states while subjects were thinking about certain concrete nouns (such as "celery" and "airplane") and verbs (such as "run" and "eat"). Each subject had his/her own separate set of fMRI images stored. Later, when a subject was asked to look at two images of the concrete nouns they had catalogued, and focus on a single one, a computer using the catalogued fMRI data and the new fMRI image was able to predict what the subject was thinking about with stunning accuracy.

Even more amazing is the fact that when a subject was presented with two images that had NEVER been catalogued, the computer was STILL able to predict what the subject was thinking about! This is because an intelligent computer system had been created which made correlations between words (e.g. it found that "celery" and "eat" have similarities in their fMRI scans) and could predict new correlations (e.g. "carrot" will look similar to "celery").

They also combined the catalogues of all the subjects to create a sort of "average" fMRI image of each noun. Much to their surprise, they found that this new catalogue could be used with new subjects to predict which noun they were thinking of when presented with two images. This means that most of us tend to use the same areas in our brains for each specific word, idea or thought.

This is the 'bleeding edge' of BCIs. Emotiv (see my previous post), and many other EEG based BCI's are actually quite arbitrary when it comes to assigning variables to brainwaves. For instance, volume might be assigned to increase with an increase in alpha waves, but when a person actually thinks about something getting louder an increase in alpha waves has nothing to do with volume. With this new experiment, we are getting closer to actually seeing what really happens in the human brain when we think about musical variables.

From here, it may be possible to catalogue fMRI images as a subject think of pitches, durations, rhythms and timbres. Then, a system could be built that predicts a subject's thoughts of musical elements. I hunger for that first experiment where we hear a swelling Bb played on clarinet, using mind power alone.

No comments: