The latest research in artificial intelligence seems to be closer to science fiction. A Japanese team has presented the model of a machine, based on Stable Diffusion, capable of reading minds, reports BFMTV. A detailed study was published last December on the Biorxiv website.

To achieve this, two researchers combined technology based on Stable diffusion, the tool for generating images from simple text, with functional magnetic resonance imaging (fMRI) recordings. It's about reconstructing visual experiences from fMRI data, so translating what a person thought into an image.


I'm speechless.

Not peer-reviewed yet but a submitted paper.

The 'presented images' were shown to a group of humans. The 'reconstructed images' were the result of an fMRI output to Stable Diffusion.

In other words, #stablediffusion literally read people's minds.

Source 👇 pic.twitter.com/sne2LhF8wZ

— deiniolb 🐻👉 aisuite.io (@danberridge) March 3, 2023

Access to this content has been blocked in order to respect your choice of consent

By clicking on "I ACCEPT", you accept the deposit of cookies by external services and will thus have access to the content of our partners

I AGREE

And to better pay 20 Minutes, do not hesitate to accept all cookies, even for one day only, via our button "I accept for today" in the banner below.

More information on the Cookie Policy page.



Eight volunteers

For a specialist in the decryption of IMRf signals interviewed by Libération, the two Japanese researchers do not make a big departure from current research, "whether in terms of application, methodology or scientific discovery".

To build these images, Yu Takagi and Shinji Nishimoto drew data from the Natural Scenes Dataset (NSD). Created in 2022, it accumulates fMRI readings from eight volunteers who have been subjected to tens of thousands of images.

Text, then images

Thus, 27,000 associations were encoded. The researchers then had to find a mathematical method for the fMRI recordings to be translated into text, so that Stable Diffusion could produce an image.



Conclusion: when we observe the difference between the images that had been presented to the group of volunteers and those designed by Stable Diffusion from the recorded signals, the result is very close. At most, it can be conceded that the visuals provided by the AI are less sharp and precise than the original images.

  • Tech
  • Artificial Intelligence (AI)
  • Japan