Through machine learning, or machine learning, an algorithm has allowed the discovery of materials as yet unknown. The program did this after analyzing a database of 3.3 million excerpts from articles related to materials science. The name of this artificial intelligence is Word2Vec.
It was designed by researchers at the Lawrence Berkeley National Library.
Specifically,the researchers incorporated a glossary of about 500,000 words from 3.3 million excerpts from articles at Word2Vec. AI seems to have understood complex concepts such as the table of elements and the chemical structure of molecules. By linking words, she was able to draw up a list of possible new thermoelectric materials.
The materials predicted by the algorithm have not been described in any scientific paper. These are essentially materials used for heating or refrigeration. The results of the study were published in Nature.
AI makes links that no scientist could make
Anubhav Jain, a researcher who participated in the study, explained that Word2Vec is able to “read any scientific paper, and thus make links that no scientist could make.”
The researchers taught the neural network to recognize a specific word and predict the words that follow each word. They think that some of the materials cited by AI could be better than those we already know today.
“You don’t control the algorithm, it makes the connections on its own,” they wrote in the article..
The promise of many more discoveries?
Wanting to ascertain Word2Vec’s true ability to predict scientific discoveries, the team gave it a test en by erasing recent data. . En Based on the analysis of old articles, the AI passed the evaluation.
For example, she was able to predict the discovery of one of the best thermoelectric materials available today. According to the computer simulation, if the researchers had this tool at the time,the material in question could have been found four years before its “official” discovery in 2012.
The team believes that this new use of machine learning could have applications in different fields. “It could be used for medical purposes, or in drug research,” , a-t-she wrote. “The information is there,we just haven’t made the links yet because we can’t read all the articles.”