A new artificial intelligence (AI) tool developed by IBM researchers predicts the products of organic reactions correctly about 80% of the time even without being taught any chemistry rules, according to an article from Chemistry World. The machine learning program is reportedly modelled on systems that treat chemistry like language, such as Google’s artificial neural network.
Researchers have been trying to create a functioning AI chemist since the 1970s, according to the article. In the latest effort, IBM focused on giving their program more than 50,000 reactions to train on rather than teaching it chemistry rules. This approach to learning organic chemistry, according to one of the researchers cited, centers around trying to find the underlying patterns in reactions and rationalizing them. The IBM program reportedly outperformed a comparable prediction program created by MIT.
Read the entire article here.