Worried About AI Replacing Human Workers? Cai Fang: "Let Grasshoppers Sing and Dance" – Jobs Will Continue to Be Created Endlessly

People’s greatest concern about the development of artificial intelligence (AI) is how this disruptive technological revolution will impact employment. From various perspectives—careers, jobs, or tasks—the rapid advancement and application of AI are expected to cause significant shocks. If AI is considered a revolutionary technological shift greater than any previous general-purpose technological breakthrough, its impact on human employment will also be unprecedented. In academic discussions and public opinion, there are pessimistic and optimistic camps regarding AI’s influence. This article does not intend to take sides but aims to break the boundaries of the past, present, and future of the relationship between technology and employment, conducting a thought experiment to reveal policy implications and propose recommendations.

Instead of betting on uncertainty,

why not take a leap of imagination?

In numerous studies by consulting firms, think tanks, and international organizations, conclusions about AI’s impact on employment are often ambiguous. Some findings are predictions in different versions, such as “jobs most likely to be impacted by AI,” or “human jobs AI is least capable of performing,” or by calculating AI exposure levels for common occupations or roles and ranking the likelihood and sequence of job impacts. However, the inherent fragility of such predictions becomes even more apparent in the context of AI’s employment effects.

These predictions carry considerable uncertainty in terms of timing, neither indicating how long human capabilities that outperform AI can be sustained nor whether AI’s weaknesses will remain forever. Their usefulness is limited because they rely on information constrained by current AI model capabilities or are simple extrapolations based on past experience.

Anyone acknowledging the revolutionary nature of AI development understands that the current heights and capabilities of AI models are only stage-wise progress, far from reaching their full potential. In this sense, the basis of these predictions only covers a small segment of AI possibilities and will become outdated as technology advances rapidly. Or, in other words, the information underlying these predictions is incomplete.

For example, Forbes lists 20 careers least likely to be automated, including emergency physicians, CEOs, pharmacists, and lawyers (Castrillon, 2026). But as the author states, this is only the situation in 2026. The faster AI develops and the more unpredictable its application paths, the more limited the relevance of such conclusions. Additionally, if work is broken down into “tasks,” roles requiring only entry-level skills—such as junior white-collar positions involved in support work—are already visibly impacted. Due to fierce market competition, tasks that AI is not yet proficient at may become targets for technological breakthroughs; some roles once known for interpersonal contact or “human touch” may be stripped of their “warmth” and become dominated by algorithms.

Theoretically, since we are destined not to see the full scope of AI’s possibilities before results manifest, practically, rather than betting on the future of careers based on incomplete information, it is better to allow ourselves to imagine freely and conduct a mental time travel. Keynes (2010, p. 114) once praised his mentor Marshall, believing that the most important and fundamental talent of an economist is the ability to be an outstanding historian and mathematician simultaneously, navigating between the particular and the general, the temporary and the eternal. Understanding AI’s impact on employment requires not only understanding the past and present but also “returning to the future,” enabling ideas to traverse time and space freely.

Jobs in Creative Destruction:

Past, Present, and Future

Technological change has both creative and destructive facets, and neglecting either contradicts the essence of innovation. The destructive impact of technological progress on employment appeared early in the Industrial Revolution, with machine disruptions igniting the famous Luddites movement. Since then, technology and employment in workers’ minds have often been incompatible. Contemporary economists tirelessly try to prove that, in the long run, job creation will outnumber job destruction (Autor, 2015). However, merely asserting this offers little direct help with the issues of job loss and worker impact during technological upheavals, because in most cases, new jobs are not necessarily filled by those who lost their roles.

The current Chinese labor market is best described by structural employment conflicts, with roles and job seekers not fully matched. The ability of job creation to offset or surpass job destruction cannot rely solely on a “trickle-down” effect. Moreover, as AI penetrates all aspects of economic activity, without targeted policy intervention, it will exacerbate structural employment conflicts. For example, among key groups like the elderly and youth, if employment skills are viewed as a proper combination of education years and work experience, both young and older workers lack sufficient human capital, making them relatively vulnerable groups facing structural difficulties such as high unemployment and low labor force participation. Once entry-level skills for youth depreciate and older workers face the digital divide, structural employment conflicts are intensified.

Regarding the history, reality, and future of human work, we should recognize that for disruptive technologies like AI, both job destruction and creation will be unprecedented. However, on one hand, job destruction is natural and deterministic—an inevitable result of increased labor productivity according to economic logic; on the other hand, job creation depends on policy intervention, with productivity growth as a necessary condition. Additionally, a crucial condition is that this increased productivity must be widely shared. In short, future employment—reflected in careers, jobs, roles, and tasks—can only stem from productivity improvements and their equitable distribution.

Breaking the “Isaac Paradox”:

Let the grasshoppers sing and dance!

An Aesop’s fable tells the story of the ant and the grasshopper: while the ant works hard storing food for winter, the grasshopper spends the day singing and entertaining itself; when winter comes and hunger strikes, the grasshopper begs the ant for food, but the ant refuses: “You sang all summer, now why not dance?” The image of the ant and the foolishness of the grasshopper, along with the implicit assumption—collecting food is productive activity—are based on the logic of low labor productivity, where all beings must devote themselves entirely to producing basic necessities to survive. Once labor productivity increases significantly and material wealth flows abundantly, the logic that one must work all day collecting food and cannot sing or dance becomes a “paradox,” otherwise, what is the meaning of life?

The jobs directly related to AI or human-machine collaboration are undoubtedly important sources of future roles, but they are not the only sources. Job creation must find new avenues. As previously mentioned, as long as there is sufficiently high and equitably shared labor productivity, new jobs can be continuously created, referring to another type of roles—those in industries affected by Baumol’s cost disease (William Baumol et al., 2023; Baumol et al., 1966). Economists have long studied these sectors (professions), trying to explain why such industries can still earn comparable rewards despite lower labor productivity, including fields like barbering, healthcare, education, and performing arts.

Engaging in what was once called literary and artistic work, or now broadly viewed as creative work, has always been an ideal pursuit for countless young people. Many have failed to realize this dream, lamenting lifelong regrets, mainly due to a traditional economic logic—that one must attract a large enough audience to sustain oneself. This logic can be extended to many industries, or even reinterpreted: once employment roles can decouple from market size and wages from individual labor productivity, roles, like human creativity, will have endless sources. Therefore, it is unnecessary to focus on whether a role’s labor productivity is high or low, or even to change the measurement of productivity; instead, adopt a new idea: recognize that ants storing food for winter and grasshoppers singing for entertainment are necessary and novel forms of division of labor, and this division need not aim solely at increasing productivity as Smith advocated.

The Moravec Turning Point:

The New Face of Human Capital

Robotics scholar Moravec discovered that tasks easily performed by humans, such as movement or social skills, are often challenging for machines; conversely, tasks mastered by only a few, such as advanced mathematics or big data analysis, are rapidly evolving (Arora, 2023). Comparing AI that defeated world champions in chess and Go with humanoid robots that fall flat during demonstrations, it is easy to understand the “Moravec Paradox”—AI and robots are not simply good or bad; they remind us that: various skills derived from human intelligence can be divided into two types based on their relative advantage over AI. Since the goal of technological change is to enable AI models and agents to approach, reach, or surpass human levels through algorithms, at some point, human abilities and artificial intelligence will diverge.

Imagine a coordinate system where the horizontal axis represents the level of human capital accumulation, and the vertical axis indicates the actual utility of human capital. Traditional education aimed at enhancing cognitive abilities has always been a demand for technological innovation in human capital. As human capital accumulates, its actual utility increases, forming a positively sloped curve. As AI and human labor gradually diverge in skill advantages, the upward trend of this curve diminishes until reaching a peak. Beyond this point, further accumulation of human capital, following the previous pattern, results in decreasing actual utility, and the slope becomes negative. From this “Moravec Paradox,” a “Moravec Turning Point” emerges regarding human capital development.

This does not mean human skills are useless. Improving human capital through schooling and vocational training remains vital, but the content and modes of cultivation need to change. Even if we understand the “Moravec Paradox” and anticipate experiencing the “Moravec Turning Point,” the direction and manner of AI replacing human abilities remain unpredictable. Changes in learning content and training modes should at least include two aspects. First, whether aiming for comprehensive human development or exploring uniquely human capabilities, efforts should be made to promote creative industries and employment creation, requiring a new balance between humanities and technology, with increased emphasis on philosophy, humanities, arts, and related fields. Second, to adapt to change and ensure that human capital in the AI era keeps pace and grows, education and retraining must form a new balance, significantly increasing the contribution of training to human capital development.

The J-Curve:

Addressing Short-term Decline and Sustaining Long-term Growth

Based on comprehensive theory and empirical evidence, a reliable expectation is that AI’s impact on employment will follow a J-shaped trajectory: on one hand, job destruction will cause significant short-term shocks, represented by the downward segment of the curve; on the other hand, job creation potential can be vast, leading to a long-term upward trend. From policy perspective, the goal and approach should focus on minimizing the duration of the short-term decline to quickly enter the upward phase and ensure long-term sustainability. The following operational points summarize policy responses, which can be called the “three organic unifications.”

First, achieve an organic unity between responding to AI employment shocks and resolving structural employment conflicts. Given the externalities and uncertainties of both challenges, leaving workers to compete internally or “lie flat” will not improve individual labor market conditions and will reduce overall employment quality. Correct strategies should be based on the government providing more effective public employment services, combined with labor market mechanisms to match skills, allocate human resources and other factors reasonably, and manage the duration of conflict resolution.

Second, realize an organic unity between school education during specific periods and indefinite vocational training. On one hand, school education should focus on non-utilitarian content that develops soft skills, lifelong learning abilities, implicit knowledge, practical wisdom, and cultural and artistic literacy. On the other hand, vocational training should closely follow labor market demands, establish lifelong learning systems, and be deployed at every stage of workers’ careers to cultivate capabilities across the entire lifecycle, fostering a more balanced new type of workforce.

Third, achieve an organic unity between the market’s resource allocation function and the protection of workers’ rights. While labor market mechanisms are essential for resource allocation, because labor is embodied in human beings, development goals centered on human beings and their comprehensive growth mean that resource allocation cannot rely solely on market clearing. Creating high-quality jobs in the AI era requires strengthening labor market functions and institutions, integrating resource allocation with rights protection.

References

Arora, Anmol (2023), Moravec’s Paradox and the Fear of Job Automation in Health Care, The Lancet, Volume 402, Issue 10397.

Autor, David H. (2015), Why Are There Still So Many Jobs? Journal of Economic Perspectives, 29(3), pp. 3-30.

Baumol, William J. and William G. Bowen (1966), Performing Arts: The Economic Dilemma, New York: The Twentieth Century Fund.

Castrillon, Caroline (2026), 20 AI-Resistant Careers with the Lowest Automation Risk in 2026, Forbes, Jan 27 Issue.

Keynes (2010), The Life of an Economist, Beijing Publishing Group, Beijing.

William Baumol et al. (2023), The Growth Dilemma: Baumol’s Disease and Its Remedies, CITIC Publishing Group, Beijing.

Editor｜Feng Biao