Artificial intelligence scanning ancient scientific research captures discoveries that have been overlooked by humans
Scientists have used machine learning to uncover information contained in old research documents. The machine learning algorithm, which uses language in millions of old research documents, has begun to make new discoveries.
Researchers at lawrence Berkeley National Laboratory detailed their work in an article published in Nature on July 3. Researchers who scanned old scientific research documents to create scientific discoveries that humans might have missed, shared the results of an algorithm called Word2Vec. After his scan, the algorithm began to speculate on possible thermoelectric materials.
For those of you who don’t know, these thermoelectric materials, which convert heat into energy, have been used so far for many heating and cooling processes. In the meantime, it is worth noting that the algorithm does not know the definition of thermoelectric and does not have any material science training.
Using only relationships between words, the algorithm reveals prospective thermoelectric material candidates that are promising and may be better than those currently being used.
Researcher Anubhav Jain said of the algorithm:
Finally, to train the algorithm, it uses the language in the 3.3 million theories summaries of material science. In this way, the algorithm has a vocabulary of around 500 thousand words and can analyze the relationship between words thanks to machine learning.