An article in The Conversation examines a new scientific study recently published in Nature. This study states that an algorithm based on artificial intelligence can predict future scientific discoveries by extracting data from previously published scientific articles.
In particular, it would be Natural Language Processing (NLP), which is one of the many existing machine learning techniques for evaluating data and information from the scientific literature and extracting other information through essentially unattended learning. This means that the same algorithm can improve itself depending on the amount of data to be analyzed.
The researchers in this new study used complex statistical and geometric properties of the data so that the algorithm could identify names, concepts, and chemical structures by itself. To this end, they fed the algorithm with 1.5 million abstracts of scientific articles on materials science.
The machine learning algorithm classified the words according to their frequency and proximity in the texts and linked them to others. The same researchers who published the study indicated that this method could be used to capture more complex relationships and identify information that is not identifiable to the human mind due to the large amount of data.
In the context of materials science, for example, after this learning process, artificial intelligence can recommend certain materials for certain applications a few years before discovery. This suggests two things: that a good degree of “latent knowledge” is already contained in previous publications and that appropriately trained artificial intelligence with a sufficient amount of data can extract this knowledge and make it available immediately.
Background info and sources:
Latest posts by Martin Hill (see all)
- Corporates purchase ‘Global energy sector reshaping.’ - May 2, 2020
- U.S. Spaceships Will Launch Passengers Into Space in 2020 Year - May 2, 2020
- TS2 SPACE in the fight against coronavirus in EU - April 29, 2020