By Dr. Laurent Hermoye
Who has never dreamt reading a boss's, a colleague's or a relative’s mind? Will new brain imaging techniques replace divination and other fortune-tellers?
When you are thinking about something (let's say a bird), fMRI can show which voxels are activated (let's say voxels 33-52-20 and 34-12-40). Mind reading through functional MRI is inverting this relationship: if fMRI shows you are activating voxels 33-52-20 and 34-12-40, can we guess you are thinking about a bird?
Several groups have taken up this challenge, using artificial intelligence techniques to infer subjects' thoughts or actions from patterns of pixels activated in fMRI images.
For example, the hippocampus is known to process experience into memories, and to be involved in the recall of spatial locations. With stimuli generated by a virtual reality system, Hassabis et al. (Current Biology 2009) asked subjects to virtually move between 8 locations within 2 rooms. Using a pattern classification algorithm to analyse the fMRI results, they were able to guess at which location a subject was standing at a given moment, from the pattern of activation of specific voxels in the hippocampus and parahippocampal gyrus.
Haynes et al. (Current Biology 2007) used similar pattern classification algorithms to predict the subject's intention to perform either an addition or a subtraction of two numbers he was shown. Decoding the activity in the anterior medial prefrontal cortex, they were able to predict the subject's intention with 71% accuracy, which is significantly above chance level.
Kay et al. (Nature 2008) used receptive-field models to build a visual decoder from fMRI data. In the algorithm training step, they established how the activity of each voxel in the visual cortex responded to locations, orientations and spatial frequencies presented in 1750 images. In the image identification step, they presented images out of a set which was not used during the training session. By measuring the response of each voxel to the novel image, and comparing it with the predicted response for each image out of this new set, they were able to guess which image was actually seen by the subject. Out of 1000 images (for example, bird 1, bird 2, bird 3,..., dog 1, dog 2, dog 3,...etc.), fMRI can decipher which one you are seeing (for example, dog 2). The next (but far more complex) step would be to reconstruct the image you are seeing or thinking about from scratch. For instance, imagine a monster, then ask the machine to draw a picture of it.
The 3 experiments described above are a first step towards mind reading. Despite their experimental complexity, the scenarios described remain relatively simple: it is a matter of guessing what you are seeing, doing or planifying within a pre-defined set of possibilities. But the number of thoughts is infinite. A mind reading experiment in a broader context would be much more complex.
In addition, all of these experiments are based on a direct relationship between a feature of the stimulus (for example, the locations, frequencies and orientations in an image) and a neuroanatomical location. This relationship is clear for some functions (somatotopy, retinotopy, tonotopy), it is more than uncertain for other functions.
To conclude, functional MRI is a promising tool for potentially reading a mind. The possible applications go beyond the imaginable: reading unconscious thoughts; mind reading in a patient with an altered state of consciousness; lie detector; and so on. This is a powerful kind of tool, which deals with the most private aspect of Self, hence it must be manipulated with care and ethics.
Version française: L'IRM fonctionnelle peut-elle lire dans vos pensées ?
No cost. No obligation. 30-day free trial.