With each new news on AI, we tell ourselves that it becomes almost impossible to go further. Meta proves the opposite, since the company is working on the development of an AI system capable of decoding brain activity to translate it into images.
The human brain is a true marvel of evolution, capable of processing an immense amount of information every second. According to the most recent research, it is estimated that its bandwidth would be equivalent to 1 exabyte per second. A figure much higher than the power of the most powerful supercomputers ever designed, like Leonardo for example.
Computational neuroscience and genomics of 21th century have greatly advanced the understanding of this complex organ. Despite this, a large part of the treatment process in the human brain remains unknown. In this nebula, Meta is currently working on a system with strong potential: understanding the internal processes of our mind using AI.
The technology behind the innovation
Meta’s discovery is based on the principle of magnetoencephalography (MEG). It is a non-invasive brain imaging technique (which does not require physical intervention on the brain) capable of recording thousands of measurements from brain activity every second. The team working on the project therefore developed an AI system based entirely on this principle and which can decode the visual representations of the brain very precisely.
At the heart of this matrix sits a tripartite architecture: an image encoder, a cerebral encoder and an image decoder? The first encoder acts as variation constructor of the representation of a single image, functioning independently of brain activity. The brain encoder, for its part, plays the role aligner and synchronizes the MEG signals to the representations sent by the first encoder. Finally, the decoder reconstructs the final image based on the data previously retrieved from the brain representations.
One might believe – rightly – that a computer capable of reading the human brain would be classified in the category nightmares of SF. But it must also be admitted that if this technology is more widely developed, the implications could be revolutionary. First, the area of image processing by the brain is largely untapped. Such technology could therefore offer unprecedented insight into this process. If we look even longer term, it could also be a therapeutic alternative offered to people suffering from brain injuries. Alzheimer’s, fronto-temporal dementia or amyotrophic lateral sclerosis, individuals suffering from these many conditions could potentially benefit from this innovation.
A technology still in its infancy
For the moment, there is no denying that the results are impressive. However, they are far from perfect and the images generated are often inaccurate in detail. If we compare MEG to fMRI (functional Magnetic Resonance Imaging), the latter is slower, but its spatial precision is higher. Currently, this is only the beginning, and there is still very much room for improvement. The beginning is in any case very encouraging.
At the crossroads of biology and AI, Meta is reaching a very important milestone. Understanding the brain is a quest undertaken since Antiquity, a period when our ancestors had already understood that it was the seat of the mind. Software imitation of the functioning of the human mind is perhaps one of the ways to explore its foundations. Everything still remains to be played out.
Source : Meta
0