For her research, Dado used AI-generated images. ‘This allowed us to know the exact AI code behind each image. We then linked this information to brain activity', she explains. Test subjects were shown pictures of faces while inside an MRI scanner. Dado examined what happened exactly in the brain as they viewed those images. By observing which “neural code” was triggered by each AI code, and then training a decoder to link them, she could determine how specific patterns of brain activity corresponded to certain AI codes. ‘This enabled us to reconstruct quite accurately which image someone was looking at, using only their brain activity.’
A Picture of Your Mother
The results were remarkable. ‘During my PhD, AI technology has advanced tremendously. Even in the last five years, the quality of reconstructed images has improved enormously', Dado says. By interpreting brain activity, she can now reconstruct with impressive precision what kind of image a person is viewing. ‘Theoretically, we no longer need the AI codes to do this. If I step into the scanner and look at a picture of my mother – an image unknown to the decoder, without any associated AI code - I'm fairly confident that the reconstructed image based on my brain activity will come quite close.’
