Slashdot Mirror
Microsoft Wants To Read Your Brain
Simon Night writes "Microsoft has entered the realm of brain machine interfaces, attempting to patent a method of classifying brain states from EEG input. 'Human beings are often poor reporters of their own actions,' the patent application notes, so reading directly from your brain is a preferred option."
I've spent a fair amount of time looking at this problem (as have others with certain agencies who have invested not insignificant amounts of money) and I can tell you that this is pretty much up in the night kinda speculation. Technical issues of obtaining clean EEG signals in a convenient manner aside, the origins of techniques like this to classify comes from the epilepsy literature, where folks attempt to classify interictal seizure spikes, but also from the sleep and awareness literature (both of which have been reasonably successful, yet are still lacking optimal tools). Granted, there are rather dramatic global state changes that occur with different states of consciousness, but this application focuses principally on the awake EEG. Specifically they quote the P300 signal which the FBI and CIA are using (trying to use) for lie detection, but the problem here is that the P300 is only an evoked potential that simply tells you whether or not someone recognizes an input (audio, visual, tactile etc...). So all questions or inputs into the system have to be crafted to understand that one is looking for an evoked potential with no necessary context in place to explain that evoked potential.
So, this is not mind reading per se nor is it a means to break cognitive processing into temporal windows to determine intent in any of those potential epochs. At its very basic fundamentals this is simply a rehash of signal detection theory doing simple Fourier analysis to "classify" brain waves. But the thing here is that there is no science behind using these signals to interpret what one is thinking even with the invocation of Bayesian networks. There are a number of other more promising methods for classifying data that have been in the literature and commonly used by a number of other disciplines that I am surprised have not made it into the EEG literature yet.
In short.... in my assessment, this is a patent proposal without much in the way of novelty or benefit to the problem at hand.
Visit Jonesblog and say hello.
Welcome that. They are welcome to read my mind, just so they will know HOW MUCH VISTA SUCKS ASS.