Researchers at Google taught AI to smell

Engineers at the Google Brain Team taught artificial intelligence to recognize odors. An article describing the study is published on arXiv.org.

During the training, researchers uploaded data on 5,000 molecules that provoke various odors into the neural network. Each was tagged with a comment - for example, causing an “oily”, “tropical” and “faint” smell.

Then the molecules were associated with descriptors and checked the operation of the algorithm. It turned out that AI can predict odor based on the structure of the molecule.

However, while a neural network determines smells worse than a person, the whole thing is perception. Two people can describe the same smell in different ways, for example, “woody” or “earthy”. Sometimes molecules have the same atoms and bonds, but they are arranged in the form of mirror images and have completely different smells.

Earlier, a group of Russian and German neurophysiologists presented artificial intelligence, which can analyze the signals of the electrical activity of the brain and find traces of various disorders in its work.

hightech.fm