In the near future, artificial intelligencewill be able to replace many working professions. Researchers from the Lawrence Berkeley National Laboratory decided to test whether AI could replace scientists and make scientific discoveries independently of a person. The test algorithm was set up to study materials science. Millions of already published scientific articles were proposed to artificial intelligence, on the basis of which the algorithm was supposed to discover new materials.
Total algorithm, called Word2vec,3.3 million articles were published, published from 1922 to 2018. As a result, Word2vec found 500 thousand different words, each of which was transformed into an array of 200 numbers, a 200-dimensional vector was formed, allowing to establish the relationship between words.
Without having initial knowledge ofmaterials science, Word2vec studied the periodic system of elements of Mendeleev and realized the concept of the crystal lattice. At the next stage, the AI revealed several most promising thermoelectric materials. After studying the 10 proposed Word2vec new materials, scientists from Berkeley found that they all surpass the average indicators of similar materials used in practice.
Using the time limit in 2008, it was found out that of the five materials offered by Word2vec is currently open and three of them are already being used in life.