AI Makes New Scientific Discoveries By Analyzing 3.3 Million Scientific Abstracts

A machine learning algorithm was able to make completely new scientific discoveries by analyzing scientific papers for connections humans may have missed.

Researchers from the Lawrence Berkeley National Laboratory used an algorithm called Word2Vec, which allowed their algorithm to create predictions for possible thermoelectric materials. These materials are used to convert heat to energy and are used in heating and cooling appliances. The algorithm did not know the definition of thermoelectric, but it was still able to provide candidates for future thermoelectric materials.

Researcher Anubhav Jain stated that “it can read any paper on material science and can make connections that no scientists could.”

The Word2Vec algorithm was trained by assessing the language in 3.3 million abstracts related to material science and learned about 500,000 words. Jain explained that “the way that this Word2Vec algorithm works is that you train a neural network model to remove each word and predict what the words next to it will be. By training a neural network on a word, you get representations of words that can actually confer knowledge.”

In one experiment, the algorithm analyzed only papers published before 2009 and was able to predict one of the best modern-day thermoelectric materials four years before it was discovered in 2012.

Lead author of the study Vahe Tshitoyan states that “this algorithm is unsupervised and it builds it won connections. You can use this for things like medical research or drug discovery. The information is out there. We just haven’t made these connections yet because you can’t read every article.”

Read the original article by Nature here

Like the post? Share it:
Andre Moncayo
Andre Moncayo
Marketing Associate