Computers may not be able to hack our brains the way we do with their systems, but we may be one step closer to neural hacking and mind reading. University of Helsinki researchers have developed a technique for computers to model visual perception by monitoring human brain cells. This means computers will be able to predict our thoughts and even illustrate them.
The research shows that computers can communicate with us via our thoughts through a process called neuroadaptive generative modeling (NGM) that can adapt to our neural activity. This is done through the interaction between brain responses and a generative neural network. NGM requires the computer information presented and brain signals to be modeled simultaneously with artificial intelligence to generate images that match the visual characteristics participants focused on.
To prove its effectiveness, 31 volunteers were shown A.I.-generated images of different people. The electrical activity in their brains was recorded through an electroencephalogram (EEG). Participants concentrated on certain features while their EEGs were fed to a neural network to indicate when the brain detected those features. This information was then fed to an A.I. system to estimate when the participant’s thoughts matched what they were looking for. The researchers presented them with computer-generated images and asked if the images matched their thoughts.
The experiment proved to be 83% accurate Researchers believe that such findings may help science better understand human perception and the underlying processes in our minds. Senior Research Michiel Spapé says this method “responds to the associations we have with mental categories,” and “may provide a new way of gaining insight into the social, cognitive and emotional process. One direct benefit is that this method can help those with artistic ambitions but who lack the talent to draw and illustrate.
Although the ability to print and create digital images from mere thoughts seem exciting, there are potential legal concerns such as the issue of privacy in using personal data. Until the implications of neuroadaptive generative modeling are worked out, we can reassure ourselves and maintain some level of privacy in our one-sided conversations with our computers.