The key facts

  • Neurobiologist Pavel Balabanov considers the value of AI for science 'zero': these are ordinary programs based on principles described in 1986–1987.
  • Neural networks make mistakes significantly more often than classical programs — an error rate of about 5% is the norm for them.
  • An erroneous result can become the starting point for subsequent work and lead an entire research field astray.

Neurobiologist assessed the role of AI in science

The use of artificial intelligence in science can lead researchers into a dead end, warned neurobiologist, Doctor of Biological Sciences, chief researcher at the Institute of Higher Nervous Activity and Neurophysiology of the Russian Academy of Sciences Pavel Balabanov in a conversation with Gazeta.Ru. In his opinion, the hype around AI for science is 'simply zero,' since neural networks remain ordinary programs. 'Yes, we use all kinds of programs, including those involving artificial intelligence, but they remain just programs,' the scientist said.

Nothing new since the eighties

Balabanov noted that setting up an experiment without a computer is already unprofitable, although possible, and working with AI is more productive, but it brings nothing fundamentally new. According to the scientist, modern AI is based on principles of brain function described in the eighties: the first articles on what is now meant by AI were published in 1986–1987.

The danger of errors

The main danger of AI is the error rate. Neural networks, unlike classical rigid programs, have more capabilities, but they also produce significantly more errors: an error rate of about 5% is the norm for such systems. 'Such a number of errors is normal for artificial intelligence. But in science this is unacceptable. One out of twenty experiments will be set up incorrectly,' Balabanov warned. A dangerous scenario arises when an error ceases to be local: a false result becomes the starting point for subsequent work, and an entire research field is put at risk. 'It can lead all the other experiments astray,' the neurobiologist explained.

How AI is used in the laboratory

In the work of his laboratory, Balabanov uses AI very sparingly and only as a service program that saves time. The future of neuroscience, he added, is connected not with AI but with the study of the body's physiological reserves, including the brain's ability to restore its own cells — here AI can so far help only indirectly. Earlier, scientists warned that 10 minutes with a neural network reduce a person's cognitive abilities.