Reading Minds With fMRI

Well, it may not be mind reading just yet, but a computer model developed by a group of neuroscientists at the University of California, Berkeley, is perhaps one (tiny) step closer to that sort of technology. In a study to be published in tomorrow’s issue of Nature, the group describes the use of the computer model to accurately identify which photograph—out of a group of many—a subject had just looked at, based only on fMRI data. Even more impressively, the model worked with photographs the participants had never seen before.

Studies of this sort have been conducted in the past with success, but involved only simple patterns or basic object recognition. In the current study, two participants were shown 1750 photographs of various scenes and objects while their brain activity was measured with fMRI. Using the data from these fRMI scans, the researchers created a computer model to distinguish patterns of activity in the visual cortex that occurred in response to specific features of the photographs. For example, the model could be used to determine which areas are typically activated in response to lines, spherical shapes, or spots of dark shadowing. To do this, they divided the fMRI representation of the visual cortex into small cubes and used the model to examine how activity in each subsection changed in relation to different photographs.

After the initial fMRI data was analyzed with the computer model, the two subjects (also co-authors of the study) then viewed 120 photographs they had never seen before while being scanned again. The researchers used the model to predict what the brain activity of each subject would be as they viewed the novel pictures. For one subject, the model’s prediction matched the actual brain activity 92% of the time. For the other, it was accurate 72% of the time. By chance alone, it would have made the correct match only 0.8% of the time. One of the subjects then viewed a set of 1000 pictures with scenes more similar to one another to further test the specificity of the model. It was still accurate 82% of the time.

While the mention of mind reading above is, of course, a bit sensationalistic, this technology is still amazing, and perhaps a harbinger of strange things to come. If we can eventually predict patterns of brain activity in response to visual stimuli with precision, who is to say we will not one day be able to dissect more complex thought processes, or at least identify sharp distinctions, like when one is telling a lie vs. telling the truth? Such technology, if determined to be accurate, could have interesting ramifications.

This is all speculation about things that may happen in the distant future, however, and only tenuously related to the computer model discussed above. After all, the model is still limited to pictures from a known set. It could not be used to interpret fMRI data and reconstruct a semblance of what a person has seen, it can only match the activity to photographs it has been exposed to previously. Regardless, it is an area of research that is worth following closely, as it involves perhaps the most precise elucidation of cognitive processes we have yet to be privy to.