A team of four scientists at Japan’s Kyoto University succeeded in using artificial intelligence to decode images of thoughts.
The study, released in preprint server bioRxiv (“bio-archive”) by researchers Guohua Shen, Tomoyasu Horikawa, Kei Majima and Yukiyasu Kamitani, used deep neural networks (DNN) to generate “hierarchical” images — visualizations composed of multiple layers of colors and shapes — based on brain activity.
These images appear more sophisticated than visualizations rendered by existing techniques such as magnetic resonance imaging (MRI).
The experiments, conducted over 10 months, had three subjects view natural images, artificial geometric shapes and alphabetical letters for varying durations.
They also had one mental-imagery session, which asked them to remember one of 25 images (10 natural and 15 geometric) presented earlier.
Researchers refer to their methodology as deep image reconstruction, which essentially combines “the DNN feature decoding from fMRI signals and the methods for image generation recently developed in the machine learning field.”
The results look straight out of a sci-fi movie. Artificial intelligence literally read minds, as seen in the following figures:
The mental-imagery session was more challenging for the AI, but only because the subjects had a harder time remembering images:
Kamitani told CNBC Make It:
“We have been studying methods to reconstruct or recreate an image a person is seeing just by looking at the person’s brain activity.
“Our previous method was to assume that an image consists of pixels or simple shapes. But it’s known that our brain processes visual information hierarchically extracting different levels of features or components of different complexities.
“These neural networks or AI model can be used as a proxy for the hierarchical structure of the human brain.”
It’s safe to say that the technology behind deep image reconstruction is still in its early stages, but applications are no short of promising. For one, it can be used to visualize hallucinations of psychiatric patients to improve treatment.
Then there’s the prospect of visualizing dreams, which can prove useful in both medical and non-medical interests.