Notes from the President's Office

September 30, 2024

Professor Kohei Itoh
President, Keio University

On Monday, June 12, 2023, OpenAI CEO Sam Altman visited Mita campus (report and photos) and, as befitting the "father" of ChatGPT, opened his own "chat" with students, answering their questions one after another in the span of an hour (video). Yet, unlike his generative AI program, Altman's responses were all drawn from his own personal opinions and experiences, and in the last minute he asked if he could share some closing thoughts. He said, "You are all the perfect age to really benefit from this incredible [AI] revolution that's just been kicked off. These don't come along very often. Most people don't get to start off their careers with such an up elevator. Whatever you want to do to learn about it now or to get familiar with it, it's all going to be valuable to you later... And so, I wouldn't worry about [questions like], 'Do I know my total life plan?' or 'What is the exact technology I want to use?' I think any exposure to it―getting familiar with it, getting good at it, thinking about it, figuring about how to go be productive in the world in a new way―you're all just going to have an incredibly good time." When he said that, what struck me was the phrase "up elevator." Everything he said was condensed in those two words.

I'd like you to imagine the university's typical curriculum up to this point. When you enter university as an undergraduate student, you first enroll in core courses. After that, you take classes in your major field. Finally, you apply all the knowledge you accumulated to your graduation thesis. You climb up the "stairs" of university one step at a time.

But when a generative AI program like ChatGPT is implemented into the curriculum, one of the assignments you might encounter early in your first year could be a prompt such as this: "Japan currently has no penal law provision that punishes individuals for war crimes. Submit a concrete proposal that will create legislation to account for war crimes." Normally, you would need to study criminal law and international law, research the legal systems in other countries, and consider how to position war crime laws within a Japanese context. After that, you would draft and submit a detailed proposal. However, by "chatting" with ChatGPT and feeding it the right prompts, you may arrive at a relatively decent answer without the need for such prior knowledge or research. Indeed, it would be just like taking the elevator to the top floor. However, you wouldn't learn anything if that was all there was to it. What matters is being able to control, via effective chat prompts, how the AI takes you to the top, as well as knowing the process that takes you up. You just need to ask the right questions about present criminal law, international law, and the current state of legal systems in other countries as you ride the elevator up. You may also find yourself questioning how constitutional law relates to fundamental human rights. Through this elevator, you can come to understand the significance of legal systems including those mentioned above, and you will be further motivated to study law from the ground up. This will consequently allow you to pour your energy into the core courses you enroll in that pertain to those laws. While doing this, you will also learn how to use AI effectively. If you hone your skills like this, then later, as you advance in university and are confronted again with the question on war crime law, you will be able to arrive at a proposal that is far more nuanced than the one you made through ChatGPT as a first year in college who had not yet learned anything. By this time, you will have accumulated the foundational knowledge you need, be able to form critical opinions, and have the skills to analyze primary sources and other materials. Even if you do use generative AI, you will be able to feed it far more informed prompts that reflect your own thinking as someone who has climbed up the stairs of university. As a result, you will also be able to recognize the strengths and weaknesses of any proposal churned out by generative AI.

The same is true for the popular field of data science. First-year students with limited knowledge of mathematics and statistics could conceivably be given various real-world data related to business or another research field, from which they would be able to analyze different correlations and causal relationships. This is possible because as long as you know the end goal, you can use generative AI to decide which statistical method to choose, create a program to achieve your goal, conduct statistical analyses, and even visualize the results in an easily digestible format. You can take the elevator straight to the top. Through the prompts you give and the outputs you generate, you can learn what kind of analyses or formulae were used by AI. Your basic course of study would then be to learn this information for yourself. Naturally, you would also learn other core statistical methodologies and the differences between them. This will help motivate you as you set your sights on mastering the fundamentals of math and statistics.

Before the arrival of generative AI, it would have been impossible to pose these questions to newly enrolled students because they require prior knowledge or specialized research skills. Lacking the relevant knowledge and experience, students simply would not know how to proceed even if they put their heads together. It is the job of university faculty to help them grow. The tried-and-true method for doing this is to build students' knowledge from the ground up. Academics have the ability to climb up to the top floor of their respective fields without using elevators. However, as teachers, it is difficult for them to accompany their students up the stairs explaining to them individually how things work, unless their classes are small or one-on-one. The advent of generative AI is like hiring a private assistant. The reason why I won't call it a private tutor at this point is because the current generation of AI technology makes mistakes unless the student asks the right set of questions. However, once generative AI is designed specifically for educational settings, the more foundational the course, the closer AI will become to being a phenomenal private tutor.

In the future, AI and other similar technologies can also be used to increase the number of international students enrolling at Keio. Even if a class is conducted in Japanese, there could be a voice recognition system that displays what is being said in Japanese subtitles, thereby allowing for simultaneous interpretation into the foreign student's native language. Those students not used to Japanese, for example, can learn the material with the aid of Japanese subtitles displayed on their own screens or through simultaneous interpretation, and can, in turn, read aloud any questions and comments they have as their devices translate their native language into Japanese. Similarly, writing in Japanese would be easy. That’s right, international students would be able to take the elevator straight to the top floor. Because they can experience the thrill of seeing their studies being put into action from the very start, they will be motivated to learn basics of Japanese for themselves. When starting to learn the Japanese language using AI, the student can be fielded questions one after another based on their level. When they answer them correctly, they advance in level, like clearing stages of a game, and so the teaching materials will be used to the fullest in the process. Since international students will be able to use Japanese in no time, more of them will be able to play an active role in Japanese society after graduation. Conversely, it will also make it much easier to offer classes in English and other languages and accelerate foreign language acquisition for students who grew up in a Japanese environment.

It goes without saying that access to AI will revolutionize not only learning, but also research and business. This is why it is necessary to build up one's skillset in higher education to use AI correctly. We can break down how we can use AI into the following three categories:

1) Using AI to train humans

In a conversation between Kyoto University Professor Shinya Yamanaka and shogi player Sota Fujii, the latter made the following comment: "In board games such as go or shogi, where the playing field space is limited, the AI is overwhelmingly stronger. But this is why I feel that we are entering a phase where humans will become stronger through their use of computer technology." Shogi players up to now have made use of other human players' strategies as models to learn from and hone their own skill. However, since AI models have every possible combination of scenarios, they will also make moves that have never been tried by humans. Fujii is said to be improving his own skills through repeated matches against AI. But his opponents in actual matches will be against fellow humans, without the aid of AI. It will be pure a battle of wits. These are just some ways that AI can be used to hone human capabilities, making the technology indispensable in other aspects of education and work.

2) Using AI to enhance one's studies, work, and the pursuit of knowledge

As Sam Altman mentioned at the beginning of this piece, knowing how to use AI and other advanced technologies to complete a task or set of tasks will be essential in the future. As you become more proficient in its use, you will notice some things. Since generative AI is trained (and its performance improved) through big data, English language use-case scenarios will be the go-to option based on data pools. However, for studies, research, and work related uniquely to Japanese culture, it would be more advantageous to use AI in Japanese because that is where larger amounts of relevant data can be found. Rather than AI replacing humans, there will be an increasing number of areas where humans who can make full use of AI in this way will have an overwhelming advantage.

3) Pitting AI against AI

The aforementioned AI-powered shogi computers improved their capabilities by being pitted against other shogi computers. Consider cyber security and cyber warfare. By pitting AI computers against each other and making them play both offensively and defensively, they will learn to make instantaneous judgments that human capabilities would not be able to keep up with. This is also true of financial markets. Competition in automated trading based on AI decisions will increase the number of situations where the speed of human decision-making is insufficient. In other words, while humans focus their efforts on designing, making, and improving AI technology, the role humans play in the practical applications of such devices (i.e. computers) will decrease over time. This is the area where AI is said to replace humans, but we must not forget that it is humans who will create and improve AI programs.

What is clear from this arrangement is that education to date has been conducted in an analog manner, focusing on training people as mentioned in 1) above. However, from here on out, educational materials using AI will be used for learning the fundamentals as much as possible, allowing teachers to allocate time for instruction focusing more on human interactions, such as having advanced discussions with students, undergoing practical training, and conducting field work. In parallel, by increasing the number of courses that implement AI as envisioned in 2), we can create a learning environment full of surprises for even teachers. Experts in research and development that work on AI will provide support in all areas from 1) through 3). In light of categories 1) and 2) above, Keio launched the AI and Advanced Programming Consortium (AIC) in 2019 to create an environment in which students who excel in AI and programming can teach novice students to improve their skills in those areas. They are also working on the development of AI teaching materials. Furthermore, it is necessary to promote the development of cutting-edge AI focused on category 3) within the larger framework of science and technology. To that end, Keio, leveraging its experience in AI and information science, has established a partnership with Carnegie Mellon University to launch a new AI research center. Top domestic and foreign companies will join this center to create innovative core AI technologies and to propagate their use in various fields, including literature, economics, science, and medicine.

Now, I'm sure the readers who have followed along with me will ask, "Is AI really that great?" "Do you really like AI that much?" I can almost hear the reader sighing. Even I feel a bit out of breath after writing this much. AI is only a tool. Or rather, it is the responsibility of humans to continue to position AI as only a tool. Based on what I have argued so far, we have no choice but to use AI. One would think that post-COVID rendered online meetings the norm, but scholars and businesspeople are traveling the world and spending more time face-to-face than even before. That's how humans are. Things like foundational knowledge and academic ability are important, but as AI covers those skills, more importance will be placed on human qualities that make other people say, "I want to work with this person" or "I want to spend more time with this person." For that reason, the ability to cooperate and help each other, the ability to interact with others while understanding their point of view, and the ambition and moral sense to impart a better society for the next generation will become ever important. No matter how correct are the arguments you make, no matter how logically you develop your arguments using AI, if you are not someone trustworthy, people will not listen to you. When negotiations are going sour, especially when the forces of justice are needed to defeat bad actors, this is when it becomes essential for us to be able create communities and stand with our allies. In academia and R&D, no matter how much AI and other scientific technologies develop, there will always remain a frontier that can only be reached by human beings. In fact, AI is even expected to help identify and expand fields where human nature is a needed ingredient. We must accept as many foreign students as possible in Japan in order to broaden their horizons, to understand and respect different cultures, and at the same time to help the world understand the strengths of Japanese culture. To this end, it is desirable for higher education institutions in Japan to lead the world in the use of AI. Human connections are everything when it comes to building a cultural, prosperous, and peaceful society where people respect others' human rights. This is why I believe that higher education institutions in the age of AI should let AI do what it can do, and allocate the extra capacity freed up to promote human-to-human interactions in their advancement of learning and research.

Appendix