Of all the research into artificial intelligence, the ability to represent what we think about in images seems to be the most worrying for fundamental freedoms. However, Meta says it has made significant progress in this very technology…
This will also interest you
[EN VIDÉO] The 10 Most Dangerous Artificial Intelligence Threats Artificial intelligence has enabled advances in health, research and…
This is an announcement that will probably give you major goosebumps… The ability to read minds is a theme that has inspired many science fiction writers or directors. In Minority Report, the vision of three “Precogs,” humans connected to a machine, is used to predict crimes before they happen. In Inception, “dream thieves” use advanced technology to enter characters’ dreams and extract or integrate information. And let’s not forget the famous “thought police” deployed in 1984 to track down subversives.
For better or worse, three researchers from Meta (Facebook) have just published a scientific article in which they take stock of a revolutionary AI that would be able to decode the brain’s visual representations in a few milliseconds…
A real-time representation of images from the brain
Meta’s technology is based on a MEG (Magnetoencephalograph), a neuroimaging technique based on magnetic fields resulting from neuronal activity in the brain. By capturing thousands of measurements per second, researchers say they could display images that correspond to the activity of the intellect.
The hellish trio
The meta system consists of three components:
Image encoder: a system that can convert any image into a format that can be processed by AI.
Brain Encoder: This matches MEG signals with images created by the Image Encoder.
Image decoder: The latter generates a “plausible image” from MEG data.
Imperfect images! To better reassure us, the meta-researchers insist that the images thus created are of a generic level and may have weaknesses in certain details.
For what purposes?
What can such technology be used for? Improved virtual reality applications or even expressive aids for those who have lost the ability to speak. Hmm… How can we not anticipate all sorts of misuses, such as trying to track down dissidents in non-democratic states, or on a smaller scale, stealing passwords or confidential information?
And now GPT can read minds!
A non-isolated initiative
Furthermore, Meta’s initiative is not an isolated case. Two Japanese researchers, Yu Takagi and Shinji Nishimoto, claimed that by coupling stable diffusion with fMRI (functional magnetic resonance imaging) signals, they achieved similar and encouraging results. At the University of Texas, Professor Alexandre Hut is working on artificial intelligence that could sooner or later generate a text as soon as a person imagines a story.
For its part, the University of California (Berkeley) demonstrated a similar technology related to audio: participants had to think of the song “Another Brick in the Wall” by Pink Floyd, and the AI generated music that was close to the original. We also know that the company Neuralink, founded by Elon Musk, is developing electronic components to be implanted in the brain for various applications, including mind reading.
A legal arsenal
This ability to translate thoughts into images is undoubtedly the greatest threat that AI poses to the privacy of all people, and it seems urgent that the legislatures of major nations address the issue to prevent any unauthorized use of such technologies .