AI and Data Science Education Research to Achieve a True Fusion of Humanities and Science
Professor, Faculty of Science and Engineering
Director of the AI and Data Science Center
I'm writing this as media reports come out suggesting that the number of patients infected with the novel coronavirus may now be on the decline (during Japan's Golden Week holiday period in May). I've been teaching online classes from home. It's been somewhat of a challenge trying to do everything in ways I'm not accustomed to--it's like having to run your own live television show. Meanwhile, I've been trying to make calm judgment regarding future forecasts based on the various prediction models that are being reported in various media outlets, including social media.
Some time ago, our research team conducted and reported on the world's largest pandemic simulation in an academic report, and I have some knowledge and computational skills in pandemic prediction models. I've also long been involved in collaborative research projects and human resource development projects with Professor Nishiura of Hokkaido University, who has been a major figure in government expert meetings. For anyone who's interested, please read our website.
In the field of simulation, pandemic simulations fall into the category of social simulations, which are different from natural science simulations in two major ways. First, while simulation models in physics and other fields stem from governing equations called "first principles," social simulations are often based on formulas described phenomenologically. By observing the object, we "extract" the dynamics of a phenomenon, and construct equations needed for calculation by borrowing the structures of various governing equations. This is also the case with models such as the SIR and SEIR, which have been used in epidemiology for more than 100 years. Therefore, in order to lend objectivity to the act of extraction, it's essential to construct a process for acquiring data, which is the result of observing the object, in a clear and careful manner. In other words, the essence of social simulations is that it's impossible to obtain prediction results that match the current situation without using the data from the simulation.
The other difference is that in social simulations, changes in human behavior affect the mathematical model itself, which is the basic principle. It's been widely known for a long time that behavioral changes in players (also called "agents") have a decisive influence on stock price and exchange rate simulations. From an academic approach, much research has been done in the study of game theory. On the other hand, human behavior has nothing to do with weather forecasts on a short-term time scale, with the exception of climate change projections related to global warming levels. Thus, in social simulations, human behavior directly impacts mathematical models. In other words, the process of human decision-making, which includes efforts to craft society in our image, and to work toward a future we want our children to inherit, needs to be included in mathematical models. Understanding these differences when compared to simulations in the field of natural science will shed light on the kind of human resources needed in the future of our society shaped by AI.
Big Data and Data-Driven Research
I specialize in mathematical and computational technologies in fields such as AI and data science. I first became interested in these fields when I was in graduate school, during the second AI boom, which I was able to experience directly. At that time, electronics manufacturers were competing with each other to launch products dressed up in technical terms such as "neuro" and "fuzzy." That boom quickly died down as AI technology failed to reach the levels expected by the industrial world and the general public. Since then, my focus has consistently been on developing and systematizing methods for inductively finding mathematical models that represent the behavior and function of phenomena from data. The concept underlying this is data-driven inferencing. Statistical machine learning, which is the foundation of modern AI technology, is also a form of data-driven inferencing.
Deep learning, which has become a topic of much discussion, so much so that it's covered in programs on NHK General TV, is only one example of statistical machine learning in terms of its technological applications. Statistical machine learning has a massive number of unknown variables (called parameters) inherent in its mathematical models. The sheer size of this variability makes it possible to represent various phenomena and respond to a wide range of needs. Determining unknown variables naturally requires massive amounts of data, and without big data, deep learning could not have been successful. Considering this, it's clear that building systems for the efficient collection of big data and effective information services is essential in business. In the early 2000s, when Google and Facebook were founded, and Amazon achieved rapid growth, a new era coinciding with the new millennium emerged, led by young people who foresaw structural changes in society resulting from the advent of big data. In comparison, at this time, statistical machine learning was technologically advancing at a slow pace. Then, with the emergence of deep learning in 2006, big data and data-driven inferencing led American and Chinese companies to great success as tools for enabling information services that targeted individuals around the world.
Human-Centric AI Society
Data-driven inferencing is a fundamental technology that supports an AI society. The level of technology has far surpassed what the industrial world and the general public expected at the time of the second AI boom, and AI technology is driving major changes in a wide range of industries and in everyday life. As mentioned earlier, while big data is essential for statistical machine learning, the tendency of US-based IT mega-corporations to use big data to monopolize wealth has come under strict scrutiny, because it's often individuals who generate big data, as is the case in e-commerce and social media. Perhaps, this is a situation that was bound to happen.
IT mega-corporations have internalized the algorithms that make information services possible, and are updating them on a per-millisecond basis with big data. Indeed, these corporations possess social simulation technology, are capable of observing society through big data, and are constantly changing their simulation models. As described above, in social simulations, the process of obtaining data and the measures to reflect it in model modification are important. From this perspective, demands were made to the US-based IT mega-corporations, mainly by European countries, to place strict regulations on big data collection processes. This is how the GDPR framework was established.
The direct involvement of human behavior in mathematical models also provides insight into future trends in the algorithms and information services of IT mega-corporations as social simulations. Society has many restrictions, including laws and social norms, as well as restrictions on a range of actions rooted in human ethics. Differences in values between countries with varying religions and cultures also lead to differences in human behavior. As long as algorithms play a role as social simulations, social changes and behavioral changes in players should be reflected in the algorithms. Now that AI technology is deeply ingrained in our social lives, we must create a human-centric AI society, not a technology-driven one. In order to achieve this, there's a strong need to visualize the needs and concerns of society and reflect them in actual AI technology development (especially in algorithms), and in relevant legal systems.
Fusion of Humanities and Science, and AI and Data Science Education
In order to create a human-centric AI society, it's important to train people with a background that incorporates both humanities and science--people who understand the basic mechanisms of AI technology, and also have knowledge in the humanities as well as in the social sciences. From the perspective of the areas of human activity in our future AI society, people who are extremely proficient in AI technology will be crucial, but there will be an even greater need for people who have successfully integrated humanities and science. I'm not a huge fan of the term "fusion of the humanities and science," but if it'll give the general public a better understanding, I'll gladly promote the importance of developing human resources who can successfully integrate humanities and science.
The young people of today are the ones who will create the human-centric AI society of the future. The values of today's youth that centers around sharing and cooperating to maintain ecosystems can be the key to solving the social problems that we face on a global scale. In order to meet the demands of this era, Chuo University leveraged its wisdom to establish the AI and Data Science Center in April 2020 with unprecedented speed. This institution also intends to use AI and data science to discover and solve problems, and help develop human resources that will contribute to the development of society and the well-being of humanity.
Professor, Faculty of Science and Engineering
Director of the AI and Data Science Center
Tomoyuki Higuchi was born in 1961, and brought up in Miyazaki prefecture.
He graduated from the Faculty of Science, the University of Tokyo in 1985.
He completed the master's course at the School of Science at the University of Tokyo in 1987.
He completed his doctorate at the School of Science at the University of Tokyo in 1989, and earned his Doctor of Science degree.
In the same year, he joined the Institute of Statistical Mathematics of the Ministry of Education, Science and Culture as assistant professor. In the years that followed, he was promoted to associate professor, then to professor.
In 2011, he became the executive director of the Inter-University Research Institute Corporation Research Organization of Information and Systems, as well as director-general of the Institute of Statistical Mathematics.
He became a professor at the Faculty of Science and Engineering, Chuo University in 2019.
He is a liaison committee member between the mathematical science and informatics fields at the Science Council of Japan. He is a director at the Japan Data Scientist Society. He is a part-time research advisor at the Artificial Intelligence Research Center at the National Institute of Advanced Industrial Science and Technology.
His expertise is in Bayesian modeling.
His major publications include "Yosoku ni Ikasu Tokei Modeling no Kihon (Basics of Statistical Modeling Used in Predictions)" (Kodansha, 2011) and "Data Doka Nyumon (Introduction to Data Assimilation)" (Asakura Publishing, 2011), among others.