In a blog post, Meta disclosed that its attempt at using AI to reconstruct perceived images via brain activity has recorded impressive results “with an unprecedented temporal resolution.”
In the short term, Meta says the project will help the scientific community better understand how the human brain receives and interprets images. Although the project is not ready for mainstream use, Meta is optimistic that it may find applicability in assisting disabled individuals in communicating their thoughts using generated images.
“Longer term, it may also provide a stepping stone toward non-invasive brain-computer interfaces in a clinical setting that could help people who, after suffering a brain lesion, have lost their ability to speak,” read the paper.
The process relies heavily on magnetoencephalography (MEG), a non-intrusive neuroimaging technique that combines integration with Meta’s proprietary AI systems. Meta’s approach involves a three-part process involving an image encoder, a brain encoder, and an image decoder, with each stage of the process trained using a set of image modules.
After the image encoder builds a set of image representations, the brain encoder aligns the MEG signals to the embeddings before the image decoder generates a final image.
While the generated images are imperfect, Meta’s demonstration indicates that object categories are entirely accurate, preserving a rich set of values.
“Overall, our results show that MEG can be used to decipher, with millisecond precision, the rise of complex representations generated in the brain,” said Meta.
For now, the regulatory limitations remain unknown, with skeptics pointing to Meta’s potential violation of privacy rights. Still, its potential for improving the lives of disabled persons outweighs future concerns.
Meta follows industry leaders in AI race
Meta joined the AI race relatively late, but its initiatives in recent months have seen it close the gap against industry first-movers like OpenAI and Google (NASDAQ: GOOGL). Meta rolled out an open-source AI model dubbed Llama 2 for commercial application, pushing the limits by launching multiple AI chatbots with different personalities.
Despite covering impressive ground, the company has been at the receiving end of a vicious legal cycle involving disgruntled individuals over allegations of copyright violations in training its models. Meta argues otherwise, saying that its model does not produce “copyright-infringing material” and urges the court to dismiss the plaintiff’s claims.
Watch: Turning AI into ROI
New to blockchain? Check out CoinGeek’s Blockchain for Beginners section, the ultimate resource guide to learn more about blockchain technology.