This AI Recreates Images From Brain Waves More Accurately Than Ever Before
Researchers have developed mind-reading AI technology that can reconstruct the images that a brain is seeing with unprecedented accuracy.
In 2022, researchers at Radboud University in the Netherlands revealed that they had developed “mind-reading” technology that can translate a person’s brainwaves into photographic images.
However, in a new study published this summer, the research team showed that they are now able to convert thoughts into images with near-perfect accuracy after giving AI systems the ability to focus on particular brain regions.
“As far as I know, these are the closest, most accurate reconstructions,” Umut Güçlü at Radboud University in the Netherlands tells New Scientist.
How The Researchers Did It
The researchers at Radboud University merged their previous 2022 study with the latest research to convert brain activity into images with exceptional precision.
In the first experiment, the team showed photos of faces to two volunteers inside a powerful brain-reading functional magnetic resonance imaging (fMRI) scanner.
An fMRI scanner is a type of noninvasive brain imaging technology that detects brain activity by measuring changes in blood flow.
As the volunteers looked at the images of faces, the fMRI scanned the activity of neurons in the areas of their brain responsible for vision.
The researchers then fed this information into a computer’s AI algorithm which could build an accurate image based on the information from the fMRI scan.
For the new study, the research team used brain signal recordings and an upgraded AI system for image reconstruction.
According to Interesting Engineering, the second study involved reanalyzing data from previous experiments where electrode arrays were implanted in a macaque monkey’s brain to monitor and record its activity as it viewed at AI-generated images.
This time, the improved AI system was able to reconstruct the original images with almost flawless precision. The images created from the monkey’s brain activity were almost identical to the original images.
This is because the implanted devices provided precise data on the monkey’s brain activity which helped the scientists reconstruct images far more accurately.
“Basically, the AI is learning when interpreting the brain signals where it should direct its attention,” Güçlü tells New Scientist.
The study is the latest example of how scientists are attempting to discover how AI models can work the brain activity to recreate images.
In October, PetaPixel reported on how Meta had developed an AI system that can scan a human brain and quickly replicate the images that a person is thinking about — in a matter of milliseconds.
Image credits: All photos sourced from “PAM: Predictive Attention Mechanism for Neural Decoding of Visual Perception” by Thirza Dado, Lynn Le, Marcel van Gerven, Yağmur Güçlütürk, and Umut Güçlü.
