That cat experiment took place in 1999, and modern brain decoding has made a massive leap from blurry, black-and-white static to crisp, full-color, high-resolution images.
Instead of relying solely on simple mathematical formulas to map individual neurons, today’s neuroscientists combine brain scans with generative AI—similar to tools like Stable Diffusion. [1, 2]
How Modern Visual Decoding Works
- Non-Invasive Brain Scans: Unlike the cat experiment, which required electrodes implanted directly into the brain (0:18), modern human studies primarily use fMRI (functional MRI) scanners. These machines track blood flow and oxygen levels in the visual cortex without surgery. [1, 2, 3]
- Separating Structure and Meaning: Modern AI decoders split brain activity into two streams. One part of the brain scan reveals the basic layout, shapes, and colors of what a person is seeing, while another part decodes the "semantic" meaning, such as recognizing whether an object is a dog, a clock, or a baseball game. [1, 2, 3]
- AI Image Generation: The decoder feeds both the rough layout and the conceptual clues into a generative AI model. The AI uses that data to fill in the missing textures and details, producing a sharp, realistic reconstruction of what the person is looking at. [1, 2, 3]
What They Can See Today vs. 1999
| Feature | 1999 Cat Experiment | Modern AI Brain Decoding |
|---|---|---|
| Image Quality | Grainy, low-resolution, black-and-white silhouettes (0:49). | Sharp, full-color, high-resolution scenes. |
| Hardware | Invasive electrodes implanted into animal brain tissue. | Non-invasive MRI scanners (and experimental MEG helmets) for humans. |
| Speed | Real-time video playback (0:33). | Mostly still images, though recent breakthroughs like Princeton's MindEye2 can reconstruct images within seconds. |
| Generalization | Custom-calibrated to one specific animal's electrodes. | Newer models like Brain-IT can transfer learned brain patterns across different people. |
No comments:
Post a Comment