Skip to main content
Создан искуственный интеллект, способный «читать мысли»

Created an artificial intelligence that is able to "read minds"

A group of researchers from Japan representing Kyoto University presented an artificial intelligence capable of "understanding" what a person thinks about and trying to portray it. Naturally, the technology is far from perfect, but the results achieved sometimes look pretty fascinating.

For ten weeks, researchers have shown volunteers various images. In parallel with the help of different equipment, the reaction of the participants' brain to each of the pictures was recorded, and the algorithm worked out by the scientists processed the accumulating information in order to then "recognize" them. The information was collected not only when the volunteers were directly examining a picture, but also when they were asked to recall the drawing they had seen before.

After scientists found that artificial intelligence was sufficiently "learned", specialists began to test it. Participants were asked to look at the image of a geometric shape or image of an animal, and the algorithm tried to "guess" what they were looking at. In most cases, looking at the image created by the program, it is difficult to say with certainty what the person looked at, but sometimes the similarity between the "original" picture and the one that was created on the basis of the volunteer's thoughts has a fairly obvious similarity.


Deep image reconstruction: Visual imagery, GIF version

Left: Imagined images
Right: Images reconstructed from brain activity (being optimized)

— 'Yuki' Kamitani (@ykamit) 4 января 2018 г.

Although researchers do not deny that their project requires a very significant improvement, they believe that in the future artificial intelligence of this kind expects exciting prospects - it is possible that one day it will be possible to view the "video recording" of a person's dream.

The scientific work devoted to the results obtained, the specialists published in the online library of preprints BioRxiv.

Add new comment

Plain text

  • No HTML tags allowed.
  • Lines and paragraphs break automatically.
  • Web page addresses and email addresses turn into links automatically.