AI as the “third way” of doing science

Mar 18, 2019, 8:24 AM EDT
(Source: Dewan S. Rahman/flickr)
(Source: Dewan S. Rahman/flickr)

Artificial intelligence (AI) has created music — it can generate news stories; it can predict life spans of human beings to a reasonable degree of accuracy; and it can also do some serious science. In fact, AI’s potential to sift through mammoth amounts of data in a flash and pick patterns and anomalies is bringing a fundamental change to how scientists go about their business.

For example, the scientists have traditionally depended on two techniques — observation and simulation — to learn about nature, but the advances in machine learning and AI have handed them a third and more attractive tool, called generative modelling, notes Quanta Magazine.

With generative modeling, researchers can feed troves of observational data into the system and let it identify, on its own, the possible theory among competing explanations for any phenomenon. The approach is so radical and successful that many in the scientists fraternity don’t hesitate in calling it the “third way” to make discoveries.

At the same time, AI cannot be trusted blindly in making scientific discoveries, as pointed out by Genevera Allen, a statistician from Rice University, writes MIT Technology Review. Machine learning has sparked a “reproducibility crisis” in science, meaning many of the results derived using the technology cannot be repeated by other researchers. An explanation for this flaw is that AI algorithms pick patterns in data that may not necessarily exist in the real world.