This ‘Mind-Reading’ AI Is a Wiz at Figuring Out What You See
The trick is, you have to be in an MRI machine — and it only works with pictures.
Israeli researchers have created an AI model named Brain-IT capable of reconstructing and predicting what people are seeing with remarkable precision. This artificial intelligence system is akin to mind reading, although the technology is still confined to laboratory environments. The researchers are optimistic about the potential for future uses, including medical applications that could help individuals communicate despite injuries or disabilities.
Brain-IT is the brainchild of Professor Michal Irani and her colleagues at the Weizmann Institute of Science in Israel. Their findings have been published on GitHub and presented in a scientific conference paper. The AI model leverages pattern recognition, utilizing data from functional MRI brain scans, which measure brain activity via changes in blood flow and oxygen.
This data is then used to recreate the images that people were viewing when the scans were taken. Additionally, Brain-IT can predict what a brain scan would look like if it were shown a particular image. While other AI models can also translate brain activity into images, preserving the general feel of an image, Brain-IT excels by accurately recreating both the content and details of the images.
The AI system is notably faster, requiring only one hour of fMRI data from a new subject to match the results achieved by methods trained on 40 hours of recording. The researchers trained Brain-IT on thousands of brain scans from a public dataset collected from eight volunteers who were asked to look at specific images while being scanned.
This allowed the AI to identify patterns in brain activity, with researchers identifying 128 "functional regions" in the brain that light up when people think about specific things. For instance, some regions activate when individuals view food while others activate when they see sports. By identifying these patterns, Brain-IT can determine what the person being scanned is observing.
Although Brain-IT has significant potential, it also has limitations. Currently, it can only generate images from fMRI brain scans, requiring time and voluntary participation in an MRI machine. The researchers are also exploring the possibility of using simpler EEG devices to achieve similar results more easily. While Brain-IT is impressive, it is not mind-reading in the traditional sense.
The technology does not capture thoughts, memories, or language. Instead, it is essentially a more efficient and dependable method of reconstructing a scene based on the brain activity of the person viewing it. Professor Irani notes that the term "mind-reading" is a "cute, jazzy name." The team now aims to achieve similar results with auditory information and, ultimately, dreams.
However, this would require overcoming the challenges of decoding video, as fMRI scans take about two minutes, whereas dreams involve dozens of images changing every second.
Written by urgent.news from CNET's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
