"The software matched their observed brain activity with the predicted activity from the decoder. When using a set of 120 images, the software got it right nine out of 10 times. With 1,000 images, the accuracy was eight out of 10. For 120 images, if the software were to simply make random predictions, its success rate would be just 0.8%.
The team estimate that if they used 1bn images (roughly the number on Google) it would have a success rate of 20%. With that many images, Gallant said, the software is close to doing true image reconstruction - working out what you are seeing from scratch. "There is no reason we shouldn't be able to solve this problem ... That's what we are working on now."
Gallant said it might be possible in future to apply the technology to visual memories or dreams"These researchers are talking about a practical, realistic, path towards actual reading of minds via technology. In fact they have demonstrated an early version of that already. In another part of the article the scientists make an estimate of 30 to 50 years for how they expect it to take this technology to mature. Want to bet their estimate is too conservative? I'd guess the real time is half their estimate, if not less.
No comments:
Post a Comment