A Paradigm Shift in Scientific Research: The Intersection of Science and AI

The methods of scientific research have evolved through several paradigm shifts. The first paradigm was empirical methods based on experiments and observations, the second was theoretical methods based on the formulation of natural phenomena, and the third was computational science methods that make full use of computer simulations. In recent years, another paradigm shift has occurred, and a fourth method is being established: the data-driven method that uses machine learning and artificial intelligence (AI).

To explain how these methods are actually used in research, let's look at the development process for luminescent materials. The terbium (Tb) compound in Figure 1 (left) exhibits green luminescence, but its intensity is weak. Where should we begin to develop a compound that shows strong luminescence from this point? One method is to synthesize compounds by gradually changing the molecules around the Tb (called ligands). However, it is quite difficult to blindly repeat experiments without knowing which changes are a shortcut to the goal. (However, repeating experiments is very important, as unexpected results can be obtained from such repetitions.) Next, we focus on the factors that determine the intensity of the luminescence. After a molecule gains energy by absorbing light, it releases that energy as either light or heat. The former is called luminescence, and the latter is called thermal deactivation. Through accumulated theoretical research, it is known that while the probability of luminescence does not depend much on the ligand, the probability of thermal deactivation varies greatly depending on the ligand. In other words, to control the luminescence intensity, we need to control the probability of thermal deactivation. At this point, it is time for computational science. The structural and energy changes during the deactivation process are calculated using a computer. As a result, as shown in Figure 1 (right), it was found that the out-of-plane bending vibration of the C=N double bond is the cause of thermal deactivation. In fact, it has been confirmed that simply replacing the C=N double bond with a single bond significantly improves the luminescence intensity.

From what has been discussed so far, you might feel that the first through third research methods are sufficient, but ideal cases like the one described above are rare. In reality, there are many cases where multiple factors are intricately intertwined, making a theoretical explanation impossible, or where calculations cannot be performed with high accuracy using existing computer simulations. This is where the fourth paradigm—the data-driven research method—is expected to exert great power. In the data-driven research method, a machine learning modely=f(x) is constructed using a dataset of many materials (x,y), where the material's information is the explanatory variablexand the material's function or properties are the objective variabley. Once the modelfis created, the function/propertyyof an un-experimented material can be estimated by inputting itsx. In other words, even if the mechanism is unknown or simulation is not possible, it becomes possible to screen materials within a computer. The success of this strategy depends on (1) how to represent materials numerically and (2) how to collect material data. Regarding problem (1), methodologies that leverage the knowledge and techniques cultivated in theoretical and computational science are being proposed one after another. To solve problem (2), automated experiments by robots are attracting attention. Starting with "Mahoro" at the National Institute of Advanced Industrial Science and Technology (AIST), robots are beginning to advance into various fields of science. When I was a student, I never expected that a day would come when knowledge of AI and robots would be necessary for scientific research. Furthermore, in the world of computational science, a paradigm shift due to quantum computers is about to occur. Being able to witness a turning point in the scientific paradigm is the greatest joy for me as a scientist. The amount of knowledge required to catch up with cutting-edge research has suddenly increased, and I am struggling, but I see this as an opportunity and want to enjoy this new science to the fullest.