AI has been able to accurately reproduce more than 80% of images in the human brain.

The AI ​​model, called Mind-Vis, was trained on a large dataset of more than 160,000 brain scans and was able to reproduce up to 84% of what humans think.

Just a few years ago, imaging technology that simulates reading human brain activity was far from perfect. But a team of researchers from Stanford University, the National University of Singapore, and the Chinese University of Hong Kong (CUHK) are getting closer to achieving that feat.

Specifically, in one experiment, the team collected data on the brain activity of participants as they were shown a series of images. On the left are images displayed after scanning the volunteer's brain. Meanwhile, on the right are images generated by AI.

Next, the scientists used functional magnetic resonance imaging (fMRI) data to get a closer look at the participants' brain activity. The researchers then fed all of those signals through an AI model to train the computer to associate brain wave patterns with certain images.

The results from the way the AI ​​simulated the images captured by brainwaves were astonishing. Specifically, images of a house and a driveway resulted in a blend of colors that created images of a bedroom and a living room.

Meanwhile, a volunteer's image of an ornate stone tower was generated by the AI ​​as an image of a similar tower but with windows at unrealistic angles. Additionally, a bear was interpreted by the AI ​​as a strange, shaggy, dog-like creature.

According to NBC, the researchers found that the AI ​​model could generate images that matched the original photo's attributes, such as color, texture, and overall theme, up to 84%. The team believes that within a decade, the technology could be widely available to anyone, anywhere.

The AI ​​model , called Mind-Vis , was trained on a large dataset of more than 160,000 brain scans. To create highly accurate images like the one in the experiment, Mind-Vis had to undergo hours of training in the lab.

For the second part of the training process, this AI model will be trained on another dataset from a small group of participants whose brain activity was recorded by an fMRI machine while shown images.

From there, the AI ​​model learned to associate specific brain activity with visual features like color, shape, and texture. To train the AI ​​model, each participant's brain wave activity had to be measured for about 20 hours, according to NBC.

Zijiao Chen, a PhD student at the National University of Singapore and one of the researchers on the project, said that as the technology is further developed, it could have applications in medicine, psychology and neuroscience.

However, like many recent developments in AI, brain-reading technology raises ethical and legal concerns. Some experts say that in the wrong hands, this AI model could be used to interrogate or monitor people.