New Scientific Discoveries through Graph Theory
https://www.theatlantic.com/science/archive/2017/04/can-scientific-discovery-be-automated/524136/
In recent years, science has been facing a crisis: the deterioration in quality of scientific journals and papers. Part of it is due to the sheer volume of scientific literature that is published annually, which has hit 1.2 million in the past year; as the average number of scholarly article read by scientists hovers around 250, the ratio of reviewers to papers decreases. There is also increasing evidence that the results of several–in fact, the majority– of scientific papers are irreproducible. The author of this article, titled “Can Scientific Discovery be Automated?” argues that at the root of these problems is the limits of the human brain, and that scientists often “ask the wrong questions.” However, science is algorithmic, and thus can potentially be automated. Knowledge can be projected as a network in which nodes are concepts and edges are the relationships between these concepts; new hypotheses would thus be new edges. The idea of automated scientific discovery relies on large sets of data, however, and a reliable methodology of mapping concepts, relationships, and other relevant data.
This article relates to what we have been covering in class about graph theory. The idea of mapping knowledge as a network and formulating new hypotheses as edges is relevant to our discussion on how a network is connected. Furthermore, the Strong Triadic Closure property comes into play, as two concepts that have a strong relationship with each other is likely to have some kind of relationship as well. Though this “knowledge” network is not exactly a social network like the kinds we have covered in class, it was interesting to see how the concept of a network could be applied in ways that were not strictly social, and in such a way as to improve our understanding of science.