Transcription
Hello everyone. This is Sasaki from Pivot. In this Pivot Talk, we will be discussing the theme of "AI and the Great Divide." We have invited Professor Miyamoto from the Institute of Economic Research at Hitotsubashi University as our guest. Thank you for joining us.
Thank you. Professor Miyamoto, you recently wrote a book titled "AI and the Great Divide." Thank you for that. The original title was a bit different, but after much consideration, and with a slight hint of what's to come, we settled on this title.
There is a lot of discussion about AI these days, and it seems that extreme opinions and radical arguments are often flying around. In the academic and policy worlds, what kind of research is being conducted, and what are the future prospects? It felt like there wasn't a comprehensive overview of this, so today, we would like to introduce some of the research being done and discuss how we should approach the future. I believe the key message is that we can change the future ourselves.
Professor Miyamoto, have you been deeply involved in both the academic and policy worlds?
Yes, I am fundamentally a university professor. However, during my career, I was once part of an organization called the International Monetary Fund, or IMF. After that, I returned to academia. Subsequently, I was seconded to a national research institute for several years. So, I consider myself a scholar who moves between academia and the policy front lines.
Is AI a major theme in economics and policy circles today?
Yes, it is. The impetus for writing this book was actually a conference held in 2024 by the G7, the seven advanced countries. As you just mentioned, Mr. Sasaki, AI is becoming a powerful force for social change. Therefore, policymakers need to consider how to approach AI, how AI will change society, and how it will affect the economy. An expert panel was convened to discuss these issues, and my participation in that panel led to the creation of this book.
The discussion around regulation is emerging even in places like the United States, indicating a significant policy impact.
Yes, that's right. As you mentioned, in countries like the United States, there are concerns that AI might lead to job losses, which is causing hiring to become more selective. Despite graduating from good universities, many people are now finding it difficult to secure employment, according to news reports. This suggests that AI has a very significant impact, particularly on jobs and wages within the economy.
Since jobs and work are likely the biggest concerns for most people, I'd like to hear about the latest research in that area. One of the most typical examples is the IMF's research on the impact on employment in developed countries, specifically the 60% figure. Could you explain that?
Yes. In 2024, the IMF released an assessment of how generative AI will affect people's jobs and employment in the future. They estimated that in developed countries, 60% of jobs will be affected by AI. Now, when you hear 60%, you might think that 60% of jobs will be lost, but that's not exactly the case. Within that 60%, about half, or 30% of the total, might see positive effects through working alongside AI. However, for the remaining 30%, there is a possibility that AI will replace jobs. This could lead to the disappearance of certain jobs or a reduction in wages for those jobs. This is what the IMF announced.
30% is quite a significant number, isn't it?
Yes, it's very large. And this 60% figure is for developed countries; the global figure is not 60%. It's lower, around 40%. This is because generative AI has entered the realm of white-collar work. In developed countries, a significant portion of jobs depend on white-collar work, so the direct impact is greater.
That makes sense. It directly impacts white-collar workers. There's also the estimate from Goldman Sachs.
Yes. Goldman Sachs estimates that globally, about 300 million people will be affected by generative AI. There's also a study specific to the United States, which estimates that about 25% of jobs in the U.S. will be affected. This is roughly consistent with the IMF's finding that about 30% of people might have their jobs replaced by AI.
Goldman Sachs's estimate also breaks down which industries are more susceptible to AI's impact, particularly those where jobs are more likely to be replaced. As you can see, jobs like office work, administrative support, and home-related services are more likely to be replaced. These are precisely white-collar jobs. On the other hand, looking at the lower part of the graph, jobs like building cleaning and maintenance, which involve face-to-face interaction or physical labor, are less likely to be replaced. This clearly shows a division by industry, with some jobs being replaced by AI and others not.
The conclusions are quite similar, aren't they?
Yes. So, the conclusion is that rather than focusing too much on "taking away," it's important to think about how things will change shape.
Yes. It's not necessarily true that AI will lead to job losses across the board. While certain industries will be affected, new jobs will also be created with the advent of AI and other new technologies. So, while some jobs will disappear, new ones will emerge. The emergence of new jobs is extremely important. In the United States, for example, about 60% of the jobs held by the working population in 2018 did not exist in 1940. This means that over about 80 years, the majority of people are working in newly created jobs. This is the situation in the United States, and likely in other countries as well. And why were these new jobs created? Fundamentally, it's because new technologies emerged. Considering this, it's entirely possible that new jobs, which we are unaware of now, will be created by AI and absorb employment.
Is it possible to intuitively grasp whether the positive or negative impact will be greater?
That's quite difficult. Historically, when we look at the overall picture, technology hasn't completely eliminated employment or reduced the total number of jobs. New technologies emerge, and the number of jobs increases. So, in net terms, it seems that a significant number of jobs are created. However, there's a time lag. In the short term, some jobs will inevitably disappear, and those working in those jobs will naturally be negatively affected. Therefore, the impact on employment can differ depending on whether you consider it in the short term or the medium to long term.
Looking at past history, did new jobs generally emerge more often in the long run?
Yes, that's right.
I see. Many people are concerned that AI will take away jobs. However, this concern has existed for a long time. In textbooks from the First Industrial Revolution, you might find the term "Luddite movement," which involved smashing machines. They were destroying machines. In the past, skilled craftspeople who worked manually saw their jobs threatened by industrialization and the introduction of machinery, leading them to smash the machines. This movement existed during the First Industrial Revolution, and since then, there have always been concerns that new technologies will lead to job losses. However, historically, has employment actually decreased? No, it hasn't.
Is the situation with automation similar?
Yes. During the era of automation, there were ATM machines in banks. In the past, people would go to a bank teller to deposit or withdraw money. Around the 1960s and 70s, ATM machines were introduced, and there were concerns that this would lead to job losses in banks. However, in the case of the United States, the number of bank branches and employment actually increased. While the need for tellers decreased, the introduction of ATMs lowered labor costs. This led banks to open more branches and employ people in different roles, such as customer service. Therefore, even during the automation era, despite various discussions, the net effect on employment was not a decrease.
Can we assume that the situation will likely be similar this time?
That's the difficult part. The crucial difference between past technologies and AI is that AI has entered the realm of intellectual labor. Previous technologies tended to replace manual labor. So, historically, robots and machines have replaced blue-collar jobs. However, this time, AI is entering the white-collar sector, so the impact might be different.
Are there various hypotheses among economists about what might happen?
Yes, hypotheses are emerging. However, it's still very uncertain how quickly AI will advance and to what extent its capabilities will develop. Therefore, predicting its future trajectory is extremely difficult. However, precisely because it's difficult to predict, the recent approach is not to do nothing, but to consider various scenarios and then formulate economic and policy responses.
In that sense, the emergence of AGI (Artificial General Intelligence) and the timeframe for its arrival are likely to have a significant impact.
I believe it will be very significant. AGI, in simple terms, is an AI with human-like intelligence that can perform any task. There was a report released last year, in 2025, called "AI 2027." Have you heard of it?
No, I haven't.
Daniel Kottke, a former developer at OpenAI, the company that created ChatGPT, along with several other researchers, published a report suggesting the possibility of AGI being created in 2027. The report explores what the future of humanity might look like after that, presenting future scenarios in a timeline. The fact that someone at the forefront of ChatGPT development is suggesting the possibility of AGI appearing next year, in 2027, is quite noteworthy. Additionally, in 2024, Nobel laureate in Physics, Professor Geoffrey Hinton, a professor emeritus at the University of Toronto, stated that AGI could emerge within the next 5 to 20 years. The emergence of AGI or not will certainly change things.
If we were to engage in scenario planning, what would be important?
What's important is to consider whether AGI will emerge, and if so, when. Then, we need to analyze what will happen to overall economic growth, jobs, and consequently, wages. Scenario planning involves dividing these into several scenarios and conducting simulation analyses. Professor Anton Korinek at the University of Virginia has been working in this field and has presented several scenarios, about three of them. It might be better to explain this with a graph.
There's a scenario where AGI does not emerge. AI will continue to evolve, but it won't reach AGI. This is the first case. The second is that AGI will emerge, but it will take about 20 years. The third is that AI will emerge in 5 years. He has conducted simulation analyses of what will happen to the economy in each case.
There's a graph showing productivity, which refers to how much GDP will increase. The thick black line represents the scenario where AGI does not emerge. It shows that GDP will increase by about 2.5 times over the next 30 years. So, as AI evolves, the economy will grow accordingly. In the scenarios where AGI emerges, economic growth will skyrocket. As you can see from the graph, economic growth will proceed at an incredible pace.
The problem is wages. On the right side of the graph, you can see the trend of wages. In the scenario where AGI does not emerge, the economy grows gradually, and wages rise accordingly. This is a good scenario. The problem lies in the scenarios where AI emerges. Initially, wages will rise. However, eventually, wages will start to fall. Why do they fall? Because if AGI emerges, humans don't necessarily need to work. AI can do the work. Moreover, the cost of AI might be lower than that of humans. Therefore, companies will likely only pay wages to humans equivalent to the cost of using AI. For example, if AGI is expected to emerge in 20 years, the economic situation won't change precisely on that date; there will be a transition leading up to it. So, if AGI takes 20 years to emerge, wages might rise for the next 10 years, but after that, as AI gradually replaces human jobs, wages will decline. If it takes 5 years, this pace will accelerate. This is what Anton Korinek's scenario planning suggests.
These are quite extreme predictions.
They are quite extreme.
But if wages start to decrease significantly, who will be doing the consuming?
That's something we need to think about. However, the economy is growing tremendously. Ultimately, it might be possible to consider mechanisms like distributing income to people, as Professor Anton Korinek suggests in his papers. This is one possibility.
But if this is the case, some people believe AGI will emerge even sooner, perhaps by 2027.
Yes, that's right. In that case, there might be a fourth scenario, an even earlier emergence. I think that's entirely plausible.
If things develop this way, there will be a significant divide. While some jobs will be lost due to AI, new jobs that can be done with AI will also emerge, so the possibility of widening disparities is quite conceivable. This is similar to what we are seeing in the stock market, with AI-related stocks like Nvidia performing exceptionally well. This kind of structure in the stock market might be reflected in the wage market as well.
I think that possibility is quite high. Regarding the "AI 2027" future report I mentioned earlier, there's an interesting prediction. It suggests that if AGI emerges and AI capabilities further improve, stock prices like the Dow Jones Industrial Average could exceed one million dollars. These are predictions that are hard to believe from our current perspective, but they are present in the report. Of course, these are just future predictions, and it's uncertain whether they will be realized, but such possibilities are being considered.
In that case, if you just buy AI stocks, you can live without working.
That's what it would amount to. So, everyone might live on basic income and stocks alone. In fact, some people feel that way now. Yes, that's one possibility.
Given that there's a potential for a "great divide" due to AI, even if it's just a hypothesis, wouldn't society be in turmoil if we just accepted it? What kind of discussions are taking place in the policy arena?
As I mentioned earlier, the G7 convened an expert meeting on AI in 2024. Seven countries each sent one researcher. I am an economist, but there were also computer scientists and fiscal experts. Various specialists gathered to discuss the issues. After about half a year, they released a report with the following conclusions. First, the pace of AI evolution is honestly unpredictable. We don't know how fast it will evolve or how far it will go. However, if we wait until we know, it will be too late to act. Therefore, we should prepare for various scenarios before we fully understand them. Second, AI is expected to have a broad impact on society as a whole. The impact on employment and wages, in particular, is likely to be substantial. As a result, while some jobs may disappear, new jobs will emerge, potentially leading to disparities. How to address these disparities is the second point. The third point concerns finance. It's conceivable that AI will gradually enter the financial trading arena. This could lead to high-frequency trading and other possibilities. These are different from past trading practices, so we need to consider how to approach them. Also, as seen with companies like Anthropic in the U.S., as AI capabilities improve, there's a possibility of attacking the vulnerabilities of existing systems. If this were to happen to financial systems, which are interconnected globally, it could lead to widespread chaos. Therefore, maintaining the soundness and resilience of financial systems is a crucial consideration. In this regard, it's important to be able to understand why AI reaches certain conclusions. This is called explainable AI. We must not allow AI to be a black box; humans must retain control. If we cannot understand why AI produces a certain output, we cannot determine whether it is good or bad. Therefore, we need to ensure a certain level of transparency.
Finally, there's the issue of environmental impact. To further develop AI, we will inevitably need electricity and computational resources. If more electricity is required, how can we reconcile this with global warming? This is an energy issue. These are the kinds of topics being discussed in policy circles.
Regarding the example of employment and wages, what kind of policy responses are needed to protect jobs and the labor market?
This is a very difficult question, and it's unclear how far discussions have progressed at the policy level. However, in academic research, it's often emphasized how important it is for people to be adaptable to new technologies. When new technologies emerge, some jobs inevitably become obsolete or disappear. However, new jobs are also created, and the ability of people to transition from old jobs to new ones is crucial. Therefore, labor markets that allow for people's movement are better equipped to withstand such major changes. So, one policy direction is to reform labor markets to enable people to move flexibly.
In that sense, in the United States, for example, labor mobility is quite high. About 6.5% of people change jobs monthly, whereas in Japan, it's much lower. In a rigid country like Japan, what should be done?
I believe it's very important to make the labor market more fluid, or easier to move within. This is not just about AI; it's also related to other societal changes. Do you know this quiz? One in two girls, one in four boys. What is it?
Suddenly... No, no, no. Is it a profession?
It's not a profession. But the hint is that it doesn't say "women" or "men," but "girls" and "boys," so it's related to children. Ah, I see. The answer is the probability of living to age 90. Ah, I see. For Japanese girls born in recent years, one in two will live to 90. For boys, 27% will live to 90. This means longevity is increasing. And life expectancy is increasing. Moreover, healthy life expectancy has also been increasing recently. Therefore, especially for younger people, they will enter the workforce. While they are working, various societal changes will occur, not just due to AI. If we don't have systems in place to adapt to these societal changes, it won't be good. Rigid labor markets make it difficult to adapt to change. Therefore, from the perspective of increased longevity, labor flexibility is a very important point.
How will Japan respond? For example, in the U.S., it's said that new graduates are struggling to find jobs. Could this be seen as a successful response to the labor market?
Yes, that's right. However, in Japan, with the difficulty of things like lifetime employment, changing things within one's own company might be the most realistic approach.
I think that's one approach. As you mentioned, internal job postings, like those at some international organizations or overseas companies, are very good. If a position opens up in, say, the sales department, you can apply for it. Competition arises, and you can plan your career and apply. This allows market mechanisms to function. This is one approach.
That seems more suited to Japan. Although there are some drawbacks to that. So, is it necessary to have policies that support liquidity in that way?
Yes. It's not that Japan's current employment system or employment practices are entirely bad. We should retain the good aspects, but we also need to be able to adapt to changes. After all, it's the workers who will suffer if we don't.
Recently, I spoke with a former charismatic headhunter who suggested that Japanese companies might become more popular in the future. Why? Because amidst the turmoil caused by AI, foreign companies are implementing large-scale layoffs. Japanese companies, however, don't do that. Therefore, people might feel more secure learning AI and planning their careers, making employment stability a strength.
I don't think that possibility is zero. Having a certain level of stable employment is important for acquiring skills. Especially for new graduates, without it, they might not have opportunities for growth. So, that's the issue with the labor market.
Ultimately, even if jobs change, how will their content change? Is it an extreme view that entire jobs will disappear? In that sense, what kind of changes will occur, and what are the key points?
Recently, in economics, a method called the "task-based approach" is being used to analyze AI. Professor Daron Acemoglu at MIT, who received the Nobel Prize in 2024, proposed this concept. We should stop viewing occupations as monolithic entities. Instead, occupations consist of various tasks. For example, I am a university professor. My work involves research activities, teaching, university administration, and discussions like this one with Mr. Sasaki.
Even within teaching, there are various tasks. For instance, when preparing to give a lecture on economics, you first need to prepare the lecture, create slides, and prepare materials. Then, you go to the classroom, give the lecture, interact with students, and answer questions. You also create exams and administer them. I believe work is built up from various tasks. The smallest unit of work is called a task. AI is good at certain tasks. For example, creating slides, as I just mentioned, can be done by AI very quickly. AI can also definitely create exam questions.
Therefore, there are tasks that AI is good at and tasks that it is not so good at. Tasks that AI excels at will increasingly be replaced by AI or robots. Areas that only humans can do will remain. When considering an entire occupation, the more tasks that are replaced, the more the overall significance of that occupation changes. This is the approach economists are taking.
What kind of discoveries are being made from this perspective?
As I just mentioned, jobs with a large proportion of tasks that AI excels at are more likely to be replaced.
There's also another point. When we think about it carefully, we're only talking about whether jobs will be replaced or not. It's all about the 0 or 1. Professor Acemoglu's point is interesting. AI seems to have a significant impact on the economy, but he takes a pessimistic view. What does this mean? He calls AI "so-so technology." This means that AI is merely replicating the work that humans are already doing. Therefore, AI will simply take away jobs, and it won't create anything new.
That's different from the Industrial Revolution, isn't it?
Yes. So, if AI is used to create something innovative, the impact on the economy will be enormous, and that would be good for people. However, if it's used solely as a tool to replace jobs, it might not be such a good thing for the economy as a whole, he argues. Therefore, he takes a cautious view on GDP growth forecasts.
On the other hand, there are also optimistic views, such as from Philippe Aghion, a professor at a French university who received the Nobel Prize in Economics last year. He was also a member of the G7 meeting. He holds a more positive outlook. Regardless, one key message is that whether AI is used innovatively is crucial.
This also relates to what's happening on the ground, but currently, the focus is mainly on productivity improvements and cost reductions.
Yes. There are studies examining what happens when AI is introduced into existing companies, and they are quite interesting. For example, there's research on what happens when AI is introduced into call centers, which are customer support departments. Stanford University conducted a study. First, productivity increases, as you mentioned. The mechanism is as follows: customers make inquiries to the call center. These are now chat-based, not phone calls. AI provides suggested responses to the operator based on the customer's question. The operator can either use the suggestion or write their own response. For new or less experienced operators, AI's responses are very helpful. This increases the productivity of new operators, which in turn increases the overall productivity of the company. What's interesting is the productivity of experienced operators. It either doesn't increase or slightly decreases.
Why do you think that is?
I suspect that for experienced operators, AI's responses are similar to what they would have said themselves. Initially, they might respond on their own. But eventually, if AI's response matches their own thinking, they might start to follow AI's suggestions. If they do that, they stop thinking for themselves.
And their productivity decreases.
Yes. And their growth potential also seems to decrease.
So, overall, the effect might not be that significant.
That's right. It seems to depend on who uses it.
Yes, who uses it is important.
For tasks like writing, it's quite straightforward. In the previous example, it was about call centers, a specific job. However, writing is used in almost all jobs, whether it's writing business emails, proposals, or reports. This is interesting: about 500 mid-career professionals were gathered and divided into two groups. One group was allowed to use AI, and the other was not. They were given a writing task. After they completed it, a third party evaluated the results. The group that used AI had significantly higher productivity. However, what's interesting is how much AI they used. One-third of the AI-using group submitted the AI-generated responses as they were.
That might happen in universities too.
Yes, it probably does. About half of them made some modifications. However, the modifications were generally minor, such as changing the word order or a few words. They submitted these with minimal adjustments, and the third party judged them to be sufficiently good. This indicates that AI is having a significant impact on general writing tasks.
In American universities now, essays written with AI are being rejected. Do you think this will happen worldwide?
I think that's a distinct possibility. However, I think it will be difficult to track it completely. The possibility of a cat-and-mouse game is certainly there.
I am also at a university, and AI is indeed amazing. A collaborator recently told me that students in graduate school, who would normally spend about a year writing a thesis, can produce something of similar quality with AI in a very short amount of time.
So, it's strange to say "don't use it." What about consulting? It's said to be one of the fields most affected.
This is interesting. A joint study was conducted by Harvard University and the Boston Consulting Group. AI has its strengths and weaknesses. When AI is used in its areas of strength, it naturally produces good results. However, when used in areas where it is weak, its performance declines. This has been observed. Consultants were given a task, and they were allowed to use AI. Even in areas where AI was not particularly strong, they used the AI-generated answers, and as a result, their performance declined. This means that AI is not a universal tool; it has areas of strength and weakness, and it is crucial for humans to judge and utilize it appropriately. This is likely the implication of the Harvard University study.
There are many different studies.
We've looked at the impact of AI and the great divide. Conversely, what kind of jobs will remain for humans? I assume there are studies on that as well. How should we think about that?
There's a possibility that jobs that humans are specifically asked to do will remain. These are called "nostalgic jobs." Ultimately, it's humans who decide which jobs should remain, so it's very important for humans to maintain control. Furthermore, it's important to be able to judge whether the answers provided by AI are correct or incorrect, and whether they can be used. Older individuals, who have experience in their work, know how to write proposals and have gained various experiences. They are likely to feel a sense of dissonance with AI-generated outputs. The problem is with younger people. Due to their lack of experience, they may find it difficult to judge whether AI's answers are correct or incorrect. Therefore, individuals with more experience and skills might be more valuable in the context of new technologies.
With the Pivot app, you can enjoy background playback and download features. Background playback allows you to watch while commuting, exercising, or browsing social media. The download feature allows you to enjoy programs comfortably even in environments without reception or Wi-Fi. All features are free. Please take advantage of the Pivot app.