💻 AI Learns to Turn Thoughts into Images

Israeli scientists have created an AI tool that uses brain activity scans to reconstruct images that people have looked at.

The decoder model was trained using open-source data from eight volunteers. Each of them viewed about 9,000 images while lying in an MRI scanner. Based on their brain scans, the model learned to predict how colors are arranged in an image and its content (for example, "bananas on a plate"). Based on these clues, a separate model generates the final image.

The second model, the encoder, solves the inverse problem: it predicts how the brain will react to a given image. Both models are trained in tandem: the encoder generates a predicted scan for a new image, and the decoder attempts to reconstruct the original image from it. This allows the system to be trained on images that have never been shown to anyone in an MRI scanner.

💡 In the future, researchers hope to learn how to "read minds" using AI and even turn thoughts and dreams i…

@aitopics