AI News

2026 Bund Summit: AI Reshapes the Labor Market—How Should Institutions Adapt Proactively?

Sina TechUpdated Nov 16 19:53

AI summary

At the 2026 Bund Summit, guests including Cai Fang, Mário CENTENO, Mark PEARSON, and Ekkehard ERNST discussed the impact of AI on the labor market, noting that macroeconomic data have not yet shown a large-scale employment shock, but job turnover and structural adjustment are already inevitable. An OECD survey of 8,000 employers in 14 countries shows that among companies using AI, only about 1% to 2% have laid off workers because of AI. The guests advocated government intervention, redistribution policies, and reform of the skills training system, and warned against AI benefits being concentrated in a small number of companies.

Body · AI translation

Related special topic top start

Special topic: 2026 Bund Annual Meeting — Technology, Geopolitics and the Global Economy Related special topic top end Futures promotion begin Futures promotion end Miaopai begin Miaopai end Video player start Video player end Market chart begin Market chart end

Artificial intelligence is reshaping the labor market. Although at the macro data level AI has not yet caused a large-scale employment shock, structural contradictions have already emerged in recruitment, youth employment, income distribution and other areas. Whether AI can raise productivity, and how the productivity dividend can be translated into more job opportunities and welfare for society as a whole, has become an unavoidable question for policymakers in every country.

On October 21, at the 2026 Bund Annual Meeting, Mário CENTENO, former Governor of the Bank of Portugal and former Minister of Finance, Cai Fang, Chairman of the Academic Committee of the China Finance 40 Forum and Member of the Academic Divisions of the Chinese Academy of Social Sciences, Mark PEARSON, Acting Director of the Directorate for Employment, Labour and Social Affairs of the Organisation for Economic Co-operation and Development, and Ekkehard ERNST, Chief Macroeconomist of the International Labour Organization, discussed "Reshaping the Workforce: Where to Place Humanity?", moderated by Du Yang, Director of the Institute of Population and Labor Economics of the Chinese Academy of Social Sciences.

October 21, 2026, 2026 Bund Annual Meeting, Bund Roundtable "Reshaping the Workforce: Where to Place Humanity?", from left to right: Du Yang, Mário CENTENO, Cai Fang, Mark PEARSON, Ekkehard ERNST

The panelists generally said that although the impact of AI on employment has not yet fully shown up in macro data, the labor market is bound to undergo job replacement and structural adjustment, and the urgency of institutional preparation and policy intervention has risen sharply. Cai Fang stressed that in responding to the employment impact of artificial intelligence, one cannot count on technology to correct itself; government intervention and the joint participation of all parties, including finance, are needed to promote AI for good.

Cai Fang explained that this AI employment shock differs from the two previous technology shocks in history: the first two, despite the pain, ultimately created a large number of new jobs, whereas AI is beginning to replace relatively high-end jobs, and those replaced can only move downward, which may cause labor productivity to decline against the trend and incomes to deteriorate. Although the labor market can self-adjust, it is both too time-consuming and harmful to workers. Therefore, now more than ever, government policy intervention and institutional change are needed.

Mário CENTENO expressed similar concerns. He believes that traditional skill-biased technological change mainly hit low-skilled workers, whereas this round of AI revolution bears first upon high-skilled groups. Large-scale job replacement may widen income gaps and aggravate inequality and intergenerational tensions. At the same time, AI is giving rise to an atomized labor market, in which workers tend to become dispersed and isolated, making it difficult to form collective bargaining power; information asymmetry between firms and workers is amplified, and traditional protection institutions such as trade unions and collective bargaining face challenges.

Based on OECD surveys across multiple countries, Mark PEARSON pointed out that among firms applying AI, currently only about 1% to 2% have cut jobs because of AI, while more firms plan to use AI to expand their workforce. AI has the potential to bring benefits to workers, yet workers are anxious. The reason for this paradox is that existing institutions and policies have failed to ease market concerns. He also argued that the distribution issue is very important, that pressure on youth employment has long been evident, and that AI further amplifies the contradiction, so safeguarding youth employment should be a core policy priority.

Ekkehard ERNST also mentioned the distribution issue. He believes that the biggest difference between the AI era and historical technological transitions is that, for the first time, the dividends of new technology are captured by a small number of companies, and under the current institutional framework it is very difficult for market competition to dissolve this spontaneously. Wealth is highly concentrated in the hands of a few companies, and the distribution problem will be more enduring and harder to solve.

On policy and institutional development, the panelists offered proposals from multiple dimensions: Cai Fang advocated promoting AI equity, guiding AI for good, cultivating in advance industries with high demand elasticity but currently low productivity, and pushing the labor market system to connect with new forms of employment, enhancing the inclusiveness of social protection systems.

Mário CENTENO stressed the importance of redistribution policies to hedge against the risk of widening income gaps brought by AI job replacement; attention to education quality under the digital shock, so that declining education quality does not weaken labor market matching; and vigilance against AI further amplifying intergenerational tensions, with policies focusing on achieving a smooth labor market transition.

Mark PEARSON and Ekkehard ERNST both recommended reforming the skills training system. Mark PEARSON proposed promoting the introduction of AI in enterprises with supporting employee training, making the improvement of AI literacy for all a priority of skills policy, and preventing the entrenchment of training inequality.

Ekkehard ERNST proposed that training should not focus only on short-term job transitions, but should draw on Singapore's experience to build long-term career development paths for workers, and combine effective social welfare systems and resilience-building to enhance workers' adaptability. He also proposed expanding the definition of labor value so that social "activities" are remunerated, and promoting the concept of "full activity." In addition, new institutions such as legal framework reform, data markets and sovereign digital wealth funds should be used to address wealth concentration and distributional inequality.

The following is a translation and compilation of this Bund roundtable, slightly abridged, for reference only.

Du Yang: Welcome to this roundtable discussion. The theme of this session is "Reshaping the Workforce in the AI Era: Where Do Humans Belong?" We are very honored to have invited four distinguished guests to this discussion. The first is Mr. Mário CENTENO, former Governor of the Bank of Portugal and former Minister of Finance of Portugal. The second is Professor Cai Fang, whom everyone knows very well, Director of the Academic Committee of the China Finance 40 Forum. The third is Mr. Mark PEARSON, Acting Director of the Directorate for Employment, Labour and Social Affairs at the Organisation for Economic Co-operation and Development. The fourth is Mr. Ekkehard ERNST, Chief Macroeconomist at the International Labour Organization. Thank you all for joining this discussion.

As we discussed in the previous session, artificial intelligence has a major impact on the microeconomy, productivity and employment. Next, we will move to the first round of discussion. We invite each guest, in light of their respective research fields, to take about five minutes to share their views on the relationship between artificial intelligence, productivity and employment. First, let us welcome Mr. Mário. October 21, 2026, 2026 Bund Annual Conference, Bund Roundtable "Reshaping the Workforce: Where Do Humans Belong?", remarks by Mário CENTENO

Mário CENTENO: Thank you. Thank you very much for inviting me to this discussion. I would like to share the following views. From a micro perspective on AI's impact on employment and its role in the economy, it will have a positive effect on productivity. This is almost self-evident, and that is precisely why we embrace AI and invest substantial resources in it. When this effect transmits to the labor market, it will cause some jobs to gradually disappear, but at the same time it will also create new jobs. Thanks to the infinite possibilities that AI brings to the labor market and to firms' production functions, many new occupations that we had never even imagined in the past will emerge.

Jobs will be added and reduced, and the result is job replacement in the labor market. Replacement is often accompanied by a certain degree of dynamic turnover, and this is exactly what labor economists most welcome, because leaps in productivity stem precisely from this, both at the firm level and at the macroeconomic level. So in my view, what is worrisome is not this process itself, but its scale and intensity. This may again be a case of going too far, and we must plan public policy in advance to respond.

AI will bring some fundamental changes to the labor market. As we acquire more and more information, the potential concern is that this process may generate "economic rents." Why would there be "economic rents"? Because firms may capture most of the surplus value in the market. If this happens, social welfare will suffer. Another worrying issue in the labor market is whether we expect too much of the carrying capacity of the "matching function." The matching function is the mechanism that connects workers and firms. When we place excessive demands on this mechanism, it does not necessarily increase total employment, and may ultimately lead to higher unemployment and greater inequality.

Therefore, I believe it is crucial to understand the initial state of each market and the evolution behind it. As for the initial state, labor markets in most economies today are at full employment. In this economic cycle, global labor market performance has been impressive: jobs have increased, unemployment rates are low, and labor force participation has risen.

Therefore, if the labor market is hit at a high level of full employment, we must be prepared for a net reduction in jobs, because it is difficult to immediately and precisely channel every displaced worker into a new job. In the context of aging, the same logic applies. We know that when a productivity shock occurs, older workers are often the hardest hit. This is because this group does enjoy a certain high-wage "rent" in the labor market, and firms have the strongest incentive to cut them. However, older workers are precisely the group that has contributed the most to employment growth in recent years (at least in Europe and the United States). If the above situation occurs, we will once again face the difficult problem of how to properly place these workers.

Du Yang: Thank you, Mr. Mário CENTENO. Next, let us invite Professor Cai Fang to speak.

October 21, 2026, 2026 Bund Annual Conference, Bund Roundtable "Reshaping the Workforce: Where Do Humans Belong?", remarks by Cai Fang

Cai Fang: Let me first talk about the overall situation. We are all now saying that AI will affect employment. But whether in China or in other countries, if I understand correctly, so far we cannot see from macro data that AI has caused an employment shock. However, we have also seen some signs and some effects from recruitment data and from micro-level links. Some of these are real, for example, the number of new hires for programming has decreased; but some are because employers have an expectation, so they begin to wait and see, and they neither lay people off nor hire new people, forming a stalemate. I think this is a brief quiet period. After this, I think there may be more obvious labor market effects. So this gives us an opportunity to analyze the characteristics of AI's impact on employment and to make some theoretical and policy preparations. This is a rare opportunity.

I would like to discuss three key terms in this first round. The first is productivity. Because, from a very macro perspective and from a purely theoretical perspective, if AI can clearly raise labor productivity, and if the labor productivity dividend appears instantly, then we can use this productivity to create new demand and new jobs. So there is job destruction, and there is also job creation. We know that AI will certainly raise productivity—whether it accelerates scientific and technological progress and brings technological efficiency, or helps us reallocate resources with less waste and less trial and error and brings allocative efficiency, or drives our enterprises to survive the fittest, letting high-productivity firms survive and grow and low-productivity firms exit—all these channels for raising productivity can be accelerated by AI.

However, there is a problem here: like any past technological progress, the application of AI is uneven across industries, sectors, and firms, over time, and even across different processes, different labor tasks, and different links in the chain. Therefore, when you reduce labor input at one link or in one segment in order to save costs, job destruction begins. But productivity is constrained by overall equilibrium and does not necessarily rise immediately. Therefore, we see inconsistency between productivity and employment shocks, inconsistency across sectors, and so on. So this is what we should address. And we have also seen that AI is now replacing many relatively high-quality, high-human-capital positions, so that those replaced need to move elsewhere. This flow is not what we labor economists hope to see.

The second is the J-shaped change in productivity—this is not something I created; it is the conclusion of research by teams at Stanford University and elsewhere. The labor productivity raised by AI will certainly follow a J-shape. So what does this J-shape mean? It means that for a period it does not raise productivity, but as I just said, before productivity rises, labor substitution may already have occurred. Therefore, our labor market cannot wait for this J-shape, that is, it cannot wait until productivity truly rises before making institutional arrangements to solve employment problems; we must respond in a timely manner. In particularly bad cases, perhaps during the period at the bottom of the J-curve, there may be increased destruction, slow income growth, worsening income distribution, and so on, and the national economic circulation may in turn be damaged, so that the improvement in labor productivity is instead hindered.

Du Yang: Mr. Mark PEARSON, I have seen that some colleagues at the OECD have discussed the issue of OECD countries using AI to raise productivity. At the same time, we have also seen that the OECD has done some empirical research, and the empirical results exactly confirm the view just put forward by Professor Cai Fang. Could you please elaborate further on this?

October 21, 2026, 2026 Bund Summit, Bund Roundtable "Reshaping the Workforce: Where to Place Humanity?", remarks by Mark PEARSON

Mark PEARSON: Of course. Good afternoon, everyone. It is a pleasure to be here. Usually in discussions like this, someone will at some point raise the "job apocalypse" thesis, arguing that almost all the work we do in the labor market will be replaced by AI in the future. Fortunately, my colleagues are not so pessimistic; their views are much more balanced—although this morning we did indeed hear some similar arguments. In the long run, perhaps the future really will be like that, but in the short run, when we actually look at the specific data, clearly there is no such sign yet.

Mário CENTENO just mentioned that employment rates in various countries are at historic highs and unemployment rates are extremely low. Looking more closely, this summer (between June and August) we conducted a survey in 14 countries (covering the United States, Japan, and several European countries), examining the actual situation on both the employer and employee sides. This survey covered 8,000 employers and tens of thousands of employees. We found that among firms that have already applied AI, looking at their employment situation, only 11% said they had reduced the scale of their workforce; and when asked further what proportion had laid off workers because of AI, that figure was only 9%. In other words, among all employers applying AI, only about 1% to 2% said they had actually reduced positions because of AI. Of course, it cannot be ruled out that large-scale unemployment may really occur in the future. We all know that some positions may indeed be completely replaced by AI in the future, but looking at the overall situation at present, even if there really is a so-called "job apocalypse," it is still in the very, very distant future.

However, as my two panel colleagues just said, the fact that the total number of jobs has not changed overall does not mean that no changes will occur. In fact, future changes will be very obvious. But I think we need to examine exactly who will be affected. The example I often give is: think about office administrative staff in the late 1970s, think about the work they did back then—taking dictation, typing, photocopying, mailing things... almost everything they did in the 1970s has now been automated. But if we look at a labor market like that of the United States, the actual number of office administrative staff today is even greater than in the 1970s. This shows that work tasks are changing, and the specific work tasks people undertake in particular positions will change. I think this is something we will see a great deal of in the future.

Of course, I would also like to make one point: the distribution issue is very important. We have just heard the discussion about older workers. Next, I want to talk specifically about young workers. Many views hold that entry-level positions are continuously disappearing because of AI, but I think the reason is not only AI. If we look at the data, the deterioration of young graduates' employment situation actually began around 2014. At that time, the main influencing factors were offshoring and changes in corporate hiring practices—companies began to place more emphasis on practical skills and work experience, and people gradually stopped believing that formal academic credentials have a signaling role. But AI is very likely another factor making employment difficult for young people. I think this should be the core of our policy response, namely ensuring that young people can smoothly enter the labor market. Thank you very much.

Du Yang: Thank you. Mr. Ekkehard ERNST, as I understand it, the International Labour Organization has already carried out a number of studies exploring the impact of AI on the global labor market, especially on developing countries. These countries generally have weak capacity to independently develop models. What new evidence have you seen? What impact does AI have on productivity and employment in the Global South?

October 20, 2026, 2026 Bund Summit, Bund Roundtable "Reshaping the Workforce: Where Do We Put Humanity?", remarks by Ekkehard ERNST

Ekkehard ERNST: Thank you very much. The guests just now have already offered quite illuminating views, and the patterns of employment change we observe in low- and middle-income countries are similar. At present, global employment is at a historic high and the unemployment rate has hit a new low, but the other side of the contradiction is that productivity has not improved, and the growth dividend promised by artificial intelligence has not yet shown up in the data.

Regarding the research you mentioned, we conduct enterprise surveys in different countries. The surveys found that some enterprises have achieved huge productivity gains on specific tasks, such as handling customer work orders and screening and reviewing resumes. Productivity on such tasks has risen significantly, but this cannot generate a macro-level aggregate effect.

Part of the reason is that enterprises still need to restructure workflows internally and figure out which work can be scaled up with artificial intelligence and which needs to be reduced, and this adjustment itself takes time. But more important is an interesting phenomenon: the total time you save on certain tasks is not net time saved. On the one hand, this is because of the hallucination problem of large models; on the other hand, many steps still require humans in the loop to verify information. The impact of this verification cost cannot be underestimated.

In addition, many tasks that can be automated account for a very low share of employees' full-day working time. For example, software engineers are a very typical occupation affected by artificial intelligence, but coding may account for only one-third of a software engineer's working time. A large amount of a software engineer's time is spent on meetings, task planning, and other matters, and these tasks are hard to complete with artificial intelligence. This is the "Jagged Frontier" problem, and it explains to a large extent what we call the artificial intelligence aggregation paradox.

Next is the key point. The development model you mentioned is an important factor that must be taken into account. From the perspective of AI exposure indicators, AI mainly affects high-skilled positions, so high-income countries are more worried about this, because such countries have a larger share of high-skilled positions. But AI will also affect businesses such as offshore customer service exports. For example, India and the Philippines have a large amount of customer service outsourcing for overseas clients, and these services are not supplied to their domestic markets.

If these high-skilled service jobs will eventually be automated, and some of them indeed can already be realized, it will be very difficult for these countries to follow the traditional path of structural transformation: gradually moving from low-value-added, low-skilled jobs to high-value-added, high-skilled jobs. I call this phenomenon "premature loss of space for industrial upgrading," similar to the difficulties some emerging countries encountered in their early industrialization paths, and now the same phenomenon is unfolding in services.

On the road to transitioning to high-skilled services, artificial intelligence is erecting more and more obstacles. At present, artificial intelligence has not yet produced an overall effect, but this itself is a problem. Artificial intelligence hinders a country's growth potential and also puts certain groups in the labor market in a severe situation, making it difficult for them to find stable paid work. It will also slow the process of structural transformation. In China the impact may be limited, but in the Philippines, India, and low-income countries in Africa, the impact will be very obvious.

October 21, 2026, 2026 Bund Summit, Bund Roundtable "Reshaping the Workforce: Where Do We Put Humanity?", Du Yang moderating the roundtable

Du Yang: Thank you. Next we move to the second round of discussion, focusing on policy issues.

Mário CENTENO: Okay, let me first draw an analogy between the AI transformation and the technological transformations we have experienced before. Labor economists have for decades studied the impact of skilled-biased technological change on the labor market. Interestingly, we can still use this analytical framework, but the logic inside it must be completely reversed.

In the past, skilled-biased technological change was unfavorable to low-skilled workers, because market demand for skills continued to rise; but in the current AI revolution, the first to bear the brunt are high-skilled groups, as Mr. Ekkehard ERNST illustrated with the example of computer programmers. We have sent millions of young people to university to train computer programming talent, and now market demand for such positions has suddenly been hit.

Mr. Mark PEARSON mentioned an episode from the 1970s. I am reminded of another case that helps predict future trends and clarify policy direction: the fluctuation in demand for physics talent in the 1970s. During the international space race at that time, a large number of top talents in the United States pursued physics and obtained doctorates in physics, but the race ultimately came to an abrupt halt due to insufficient funding, plunging the field into crisis: on one side, the supply of physics talent continued to expand; on the other, market demand contracted rapidly.

What was the result? Unemployment and falling wages. These workers could only move into other industries. The only positive aspect of this episode was that the workers involved were highly capable—among the most outstanding students at American universities—so they were relatively easy to reemploy, and that is precisely where policy needs to focus. I will focus on three points.

First, inequality. Large-scale job displacement and labor turnover may mean that income gaps among employed workers are likely to widen, because wages for high-productivity positions may soar, while wages for positions unaffected by AI may remain unchanged. If the entire population is taken into account, inequality will intensify further, because some people will be forced to change careers, and some will not find work at all. So inequality is an issue we must take seriously. We already have policy tools to address it, and societies around the world already have experience implementing redistribution policies; it is just that this time, the scale of the impact is surprising.

Second, education, that is, the signaling problem in the labor market. Education is the core human capital supplied to the labor market; how will digitalization affect the quality of education? We have already observed warning signs: at least on average across OECD member countries, the quality of examinations has declined. Data from the Programme for International Student Assessment clearly confirm this, and it is very likely related to digitalization. Will digitalization exclude people from the labor market at the capability level? This is precisely the core concern of the labor matching function, so the policy level must also pay attention to the education sector.

Third, generational conflict. Previously, we have already experienced a social crisis whose aftershocks persist to this day, and part of the younger generation has gradually formed an expectation: that they will not live better than their parents. If the AI revolution causes this trend to continue, generational conflict may intensify and trigger social unrest. So at the policy level, we must think carefully: what support can we provide to the younger generation. As Mr. Mark PEARSON said, young people deserve special attention.

So I emphasize three directions: redistribution policy, education quality, and a smooth transition in the labor market. That concludes my sharing. Thank you.

Cai Fang: We have discussed a great deal about the extent of AI's impact on the current labor market, but the current data support does not seem particularly obvious. However, we should also recognize that AI is developing rapidly, and what we see happening today is not at all what tomorrow will bring; the high-exposure jobs calculated today also reflect only today's situation. High exposure does not mean that an impact has already occurred. Therefore, the high-exposure situation estimated today may no longer hold tomorrow. So I think we cannot accurately predict what will happen in the future, nor can we accurately estimate the scale of the impact.

In this sense, it is very important to prepare policies well now and build response platforms at the institutional level. Whether it is government intervention, policy intervention, or the participation of other macroeconomic policies, all should point toward common goals and form coordinated means. Accordingly, I offer recommendations in four areas.

First, we should focus on promoting equal AI rights. In any era, the emergence of new technologies creates divides and so-called K-shaped divergence; the divide brought by AI may be unprecedented. A large part of the structural employment contradiction in China's labor market is concentrated at the two ends of "the old and the young," namely older workers and young job seekers. For older workers, the technological divide they have not faced before is now very likely to be further widened; for young workers, the entry-level skills they previously mastered are depreciating in the face of AI; therefore, addressing the structural employment contradiction starting with these two groups has become an urgent task.

At the same time, we must also recognize that if the impact of AI has already manifested today as a technological divide, then it can be expected that tomorrow's impact of AI will be different, and therefore the range of people it may affect may be larger. So the corresponding employment support policies should be continuously upgraded—this is a key point. In addition, promoting equal AI rights is also key to ensuring that labor productivity improves, and it helps avoid the so-called Solow paradox.

Second, AI should be well guided. AI scholars raised the so-called alignment problem from the very beginning. At present, the requirement for this alignment should be expanded from the technical level to the economic and social, macroeconomic policy, industrial policy, social policy, and other fields, comprehensively requiring AI to achieve alignment.

In particular, we need to recognize that the AI we are discussing today is in fact still just a set of foundation models. What will truly play a role and truly affect our lives in the future is the practical application of these models. When they are transformed into embodied intelligence, and when humanoid robots truly acquire intelligence, their impact on workers is likely to be the greatest. Therefore, in this process, guidance, incentives, and regulation should be fully upheld, and a bottom line should truly be safeguarded: from an employment perspective, AI should be made an enhancer of workers' capabilities, not a destroyer of workers' jobs. In this sense, there is a great deal we can do.

Third, how should humans be accommodated? Although it is somewhat ahead of its time, this is also a topic of current urgency.

We should focus on cultivating in advance those industries with high demand elasticity but currently low labor productivity, such as many service industries. Since the purpose of using AI is to raise labor productivity, why should we also encourage industries with low labor productivity? Because raising overall labor productivity is precisely to leave room for those industries with high demand elasticity to survive, and we must not let the development of these industries, the creation of jobs, and people's employment and income become decoupled from individual labor productivity. "How humans are accommodated" is both the direction of future efforts and a firewall today, an important mechanism for buffering employment shocks. To this end, we can create new demand, open up new markets, and create new jobs, all of which come from people's all-round development.

Fourth, judging from China's current situation, my personal assessment is that AI's impact on employment has already begun to appear. Although it has not yet appeared on a large scale, it has already become intertwined with structural employment contradictions and labor market mismatch problems. These structural contradictions will further intensify, and they also reflect the negative impact of AI on employment.

Accordingly, we should start with structural employment contradictions to respond to the shocks that AI will bring in the future. The most important point among these is how labor market institutions—such as the minimum wage system, the minimum living security system, and the labor contract system—can connect with new forms of employment. Whether labor market institutions or social security systems, in China's past they were all closely tied to unit-based employment. As forms of employment become increasingly flexible and the share of non-unit employment gradually rises, we must explore new institutional arrangements so that labor market institutions, worker rights protection systems, and social security systems can adapt to new forms of employment. This involves many urgent institution-building goals.

At the same time, our social protection, social security, and social welfare systems should be more inclusive. This is because, in the process of AI's impact on the labor market, it is increasingly difficult for us to identify whether the shock a person suffers is due to personal reasons or macro factors. Therefore, to ensure that everyone can receive adequate protection, the inclusiveness of institutions should become ever stronger. In this sense, the 15th Five-Year Plan emphasizes investing in people; then, resolving structural employment contradictions and responding to the future impact of AI should also be incorporated into the overall strategy of investing in people.

Du Yang: Thank you, Professor Cai Fang. I think we have just heard a very important point: in the AI era, the Baumol effect may no longer be "Baumol's disease," but rather a way for people to share the productivity gains brought by AI. Next, I invite Mr. Mark PEARSON to speak.

Mark PEARSON: Everyone has already put forward some very important suggestions, and I agree with them all. I cannot elaborate on all policies in detail, so I will quickly add three points.

First, guide employers on how to introduce AI. Existing evidence has already shown that if introduced properly—for example, by consulting with employees in advance and providing training—the results are much better.

Second, support displaced workers. We know that labor markets will see worker flows, some people will lose jobs, and at the same time new jobs will be created. Some countries are better than others at making this flow produce positive effects.

Third, pay attention to development paths in the early stages of careers. We need to develop new apprenticeship systems and the like, and promote them much more widely than at present.

But what I really want to focus on is skills policy, and what kinds of skills we actually need. Because what we need is not only technical skills related to AI. If we look at the occupations most affected by AI and then gather employers' requirements for the corresponding skills online, we will find that employers need data analysis and interpretation capabilities and digital skills—I think neither of these is very surprising—but they also need creativity and problem-solving ability. So we are talking about abilities that you might not think AI possesses: the skills we need, on the one hand, are to help employees use AI more effectively, and on the other hand, are to enable people to do work that AI may not be able to handle. This is the direction of skills development.

In addition, improving AI literacy across the entire workforce is crucial. This should become a main goal of skills policy: not to cultivate AI technical talent, but to ensure that the broad mass of workers can use AI effectively.

Our survey shows that among enterprises that provided training when introducing AI, 89% of employees said AI improved their productivity; among enterprises that did not provide training when introducing AI, this proportion was only 69%. The effect of training has already become evident, but in fact only half of employers introducing AI have provided relevant training. This is simply hard to understand. You cannot introduce a major new technology like AI and still expect employees to figure out how to use it on their own. Providing training will produce better results.

At present, there is a strong push to develop more AI-related training, and that is good. But existing approaches may entrench the inequalities already present in the training field: those who currently have the least access to training are low-skilled workers. In OECD countries, about 20% of low-skilled workers participate in training each year, compared with about 50% of high-skilled workers. If we want to introduce AI without worsening inequality, we must ensure that those with lower skill levels and fewer training opportunities can also access AI-related training. Otherwise, existing inequalities will become further entrenched.

We also know what should be done: offer short courses, attach modular certificates to each course, and certify the quality of training institutions. We already have many methods for improving training outcomes. The development of AI means we must advance this work with greater urgency than ever before.

Ekkehard ERNST: I would like to add three points on the policy issues everyone has just discussed.

First, we all seem to agree that labor force transition is very important and needs support. But surprisingly, what we actually see is that transitions from one job to another are decreasing, whether job changes or other forms of job switching. This phenomenon exists in both advanced economies and many emerging economies.

The question is, how has this happened? One factor closely related to policy is that labor market concentration is rising at different income levels across countries. This is not just a problem brought by AI, because this trend already existed before generative AI models emerged, and it is closely related to corporate policies, such as non-compete clauses.

Professor Cai Fang also mentioned platform employment, where some workers are misclassified as non-employed. Therefore, there is a policy problem here that needs to be addressed, and only by handling it well can we help people move from one job to another, including between different high-skilled positions.

The International Labour Organization adopted a new convention on platform work this summer, which will help countries formulate and implement relevant legislation to clearly define the employment status of platform workers. This is a positive development, helping to clarify how workers should be classified with regard to social security contributions, occupational safety and health, and other labor market-related regulation.

This is the first point—we need to clarify why labor market mobility has not increased, and what related policy issues urgently need to be resolved.

Second is the skills issue. We should use AI as infrastructure to promote more integrated, longer-term, and more forward-looking skills policies. The skills retraining policies already implemented have had surprisingly limited actual effects. Part of the reason is that we often focus only on the next job, or only on the small step forward a person can take next.

Singapore offers a noteworthy experience in this regard. Its approach is to map career development paths in the labor market and help workers understand not only where the next job is, but also what work they might do in five or ten years. This allows them to prepare in advance and clarify which skills and capabilities they need to master. Ideally, these arrangements should also be incorporated into the corresponding institutional system, cultivating broader foundational capabilities while providing technical skills training.

Resilience is a buzzword right now, but it is not just a buzzword. Resilience can genuinely help people adapt to change, especially when combined with effective social welfare systems. Europe has good examples, such as Denmark and Switzerland. These countries have active and effective social protection systems. If you lose your job, you not only receive income compensation, but also a full set of incentives and support to help you find your next job.

Finally, I want to return to the question mentioned at the beginning: what should countries really focus on? This question is especially important as emerging economies transition toward higher-value economic activities. We need to figure out how to raise productivity in economic sectors and areas traditionally classified as low value-added and low productivity.

I am thinking in particular of one area: health care. Health care is a highly labor-intensive activity, and it is the only sector where employment has continued to grow over the past 20 to 25 years. Demand for health services is constantly increasing, and will continue to grow as populations age. We need to find ways to provide health services with greater efficiency and productivity.

Related to this, and something I pay close attention to, is that people are actually all very busy. They may not be in employment, but they do many things for which the labor market currently does not pay. I call this activity. So we need to shift the policy focus from full employment to full activity. We are now close to full employment, and we should ensure that activity, social activity, and care activity are also properly compensated. Likewise, AI, as an infrastructure tool, can help us identify where these people provide value to society—these are activities with social value.

And as Mark PEARSON said, AI can certify and track such activities, helping people obtain real social value for their activities. So I think countries should not focus only on headline sectors such as finance, AI companies, and industry, but should also pay attention to a large number of sectors currently regarded as low value-added and low productivity, and make them more attractive and more productive in the future. Thank you.

Du Yang: With a few minutes left, I would like to ask the panelists a big question. Since the first industrial revolution, technological progress has been labor-saving, and the same is true in the age of artificial intelligence.

But we have also seen that since the first industrial revolution, both productivity and employment have grown substantially. So in the age of AI, what lessons can we draw from history? And what will the future look like?

Mário CENTENO: That is a hard question to answer. But I think you have actually already given the answer: there will be more employment, but the structure of employment will be completely different. We need to build institutions that can support this progress. Daron Acemoglu has proposed "pro-worker AI." I will not elaborate on that concept, because frankly I have not fully understood it, but there are indeed some regulatory measures moving in that direction. There are also two institutions that are very important and interconnected, which we have not mentioned so far but are worth discussing. These two institutions are trade unions and collective bargaining.

The labor market in the age of AI is a completely atomized market. In my first-round remarks I mentioned firms extracting rents, and this is exactly what happens in such a labor market: firms use an agentic AI module to face each worker one-on-one, precisely knowing how many hours they work and which days of the week they prefer to work. In this very, very atomized labor market, information between firms and workers is completely asymmetric. This may be the biggest hidden concern for both equilibrium wages and workers' bargaining power. Historically, as you just accurately pointed out, workers and society as a whole have been able to build new institutions in the labor market while introducing new technologies. Now, in my view, two of these institutions—trade unions and collective bargaining—are facing great risk.

Cai Fang: How is the potential employment shock we face different from those in history? I think the biggest difference is that historical employment shocks are other people's history, whereas what we face now is our history, our reality. Historically, there was indeed a time gap between job destruction and job creation. But in history books, looking at the data, we may overlook this and assume that new technologies will eventually create more jobs. But now, when we suffer job destruction, if new jobs of the same level and the same number are not created to bridge the gap, we will face structural employment contradictions and possibly lower incomes. So for us, the most realistic thing is that we must know what to do before job destruction occurs.

As for specific features, we can consider three scenarios. The first is the Luddite movement during the first industrial revolution. The background of the Luddite movement was mainly the destruction of relatively high-end, well-paid jobs held by artisans. At the same time, more jobs were created that anyone could enter and do, requiring only cooperation with machines to produce. This also absorbed a large amount of cheap labor, the industrial reserve army, and women and child workers. Under these circumstances, wages did not grow, and the condition of the working class deteriorated. Until it improved, society was in a rather bad state for nearly half a century, which people call the "Engels' pause."

The second comparable situation is the phase when the United States and Europe suffered the shock of globalization. During this period, a large number of middle-level jobs, such as skilled and semi-skilled manufacturing workers, were lost, and the labor market became polarized. Some workers moved to lower-level jobs, and their incomes fell accordingly; others may have moved into higher-level services. Although employment was also resolved, the income gap widened accordingly, causing dissatisfaction.

This time, with the AI technological transformation, I think at least for emerging market economies, substitution also starts from relatively high-end jobs, leaving workers no way back. That is, when workers exit relatively high-end jobs, there are no better jobs to go to, and they can only move downward to lower-level jobs. The result of this downward movement is that labor productivity generally declines, going against the structural path of labor productivity evolution. At the same time, workers' conditions may also deteriorate. Historically, even under the most ideal circumstances, the process of employment recovery and adjustment took a long time, during which workers' living standards were adversely affected. At present, if we rely only on the self-adjustment of the labor market, it will both hurt many workers and take too long; society cannot afford to wait. Therefore, compared with any period in history, government policy intervention and institutional change are needed now more than ever.

Mark PEARSON: I think there is a paradox here, and this is also one of the important lessons history gives us. We estimate that over the next decade, the application of AI could raise real incomes by 0.15% to 0.95% per year. Compared with changes in real incomes in recent years, this increase is quite substantial.

I mentioned earlier that among companies adopting AI, fewer than 2% say they are cutting jobs. One point I did not mention earlier: compared with companies that are laying off workers, the number of companies planning to increase employment because of AI is much larger. Seen this way, people should be quite optimistic about the impact of AI on the labor market. In fact, workers do not think so: in OECD countries, 16% of workers are very afraid that they will lose their jobs because of AI within the next year. Sixteen percent, and within just the next year. Another 40% of workers have a moderate level of concern about this. So here is the paradox: AI has the potential to bring real benefits to workers, yet workers are deeply worried. I think the reason is precisely that the labor market will inevitably undergo job turnover, and workers do not believe that existing institutions and policies can help them get through these changes.

I think their worries are not unfounded, and you have already heard the reasons in today's discussion. In the face of job turnover in the labor market, many workers will be affected, and we do not yet have corresponding policies to help them cope. So I think the lesson of history is, as Professor Cai Fang has already mentioned, that the pattern this time is very similar to history: from the initial shock to the broader emergence of benefits, there is always a time lag in between. This is a common phenomenon. We need corresponding institutions. During the First Industrial Revolution, it took us 50 years to establish these institutions. To better manage the AI transition, we cannot afford to wait 50 years.

Ekkehard ERNST: Yes. I think there is one big difference between this AI transition and past historical transitions: for the first time, the dividends of these new technologies are captured by a small number of companies, and they will not be spontaneously dissipated by market competition, at least not under the current institutional framework. I think on this point I am very much in agreement with Mário CENTENO. Distribution issues will be much more important, last much longer, and be harder to solve. Although I come from the International Labour Organization, I do not think collective bargaining and trade unions can play much of a role, precisely because wealth has already become highly concentrated in the hands of a few companies.

What I think needs to be done is a comprehensive reform of the legal framework so as to create entirely new markets. For example, Daron Acemoglu recently proposed establishing data markets so that other companies can use existing data to develop their own models, thereby allowing the excess returns currently monopolized by a few companies to be redistributed among market participants. Another proposal I have consistently raised is to establish a sovereign wealth fund, that is, a sovereign digital wealth fund.

Interestingly, Anthropic already took a first step in this direction earlier this week. Through such funds, governments can invest in these companies, share in the returns from this massive accumulation of wealth, and then distribute it, for example directly to citizens, or indirectly by investing in infrastructure and so on. These approaches are all feasible, but we must reshape existing institutions to address the problem of highly concentrated wealth. This concentration is rooted in the existing data infrastructure, whose architectural design is largely skewed in favor of the few companies that currently benefit from this technology.

Du Yang: All right, thank you. Time is almost up. Please join me in thanking all the speakers. article_adlist[

]article_adlist news_keyword_pub,stock,sz300024 iframe below the main text iframe below the main text general statement general statement QR code at the end start QR code at the end start

QR code at the end start QR code at the end start Editor name and work code

Editor name and work code

Source

Sina Tech