The impact of technological progress on employment has been a focus of society and research since the Industrial Revolution. Movements and ideologies with Luddite characteristics have repeatedly appeared in various forms. However, whether looking at causes and results or seeing the essence behind appearances, the impact of AI on employment this time is truly different from the past. Preparation is key. Facing the possible impact of AI on employment, it is necessary to clarify some understandings and establish certain policy principles.

AI Employment Impact Differs from Historical “Technological Unemployment”

Firstly, this time it is not the specter of “technological unemployment” that has repeatedly appeared in history but a terminator that can replace almost all professions. From scientifically savvy entrepreneurs like Elon Musk to economists concerned with AI development like Lawrence Summers, they all believe that AI will comprehensively replace jobs. Once artificial general intelligence (AGI) emerges shortly, no job, whether simple, complex, physical, or intellectual, will be spared.

Secondly, the rapid pace of AI technological progress increasingly gives a feeling of significant leaps in short periods. For example, from the “Mechanical Turk” (a hoax in 1770, which can be considered the starting point of this idea) to Turing’s 1950 paper, it took 180 years; then to 1997 when Deep Blue defeated Kasparov, another 47 years; and then to 2016-2017 when the chess robot AlphaGo defeated Lee Sedol and Ke Jie, it was about 20 years. However, only one year has passed between the emergence of ChatGPT and the launch of Sora. We do not need complex models to predict the future, but rather only need to look at this speed and acceleration to reasonably anticipate the appearance of AGI.

Lastly, the “development paradox” of large model AI predestines a nearly inevitable large-scale loss of jobs. All camps, countries, and companies recognize that whether they can occupy the commanding heights of AI technology and industry is a matter of life and death. This has led to a competition around AI development similar to the space race, arms race, and nuclear weapons race during the Cold War. Additionally, large model AI is highly energy-consuming and costly. The inevitable direction and method of exploring model applications, expanding user bases, and increasing return rates is to improve labor productivity, thereby reducing the use of labor and human capital.

Two Fundamental Ways for Humanity to Respond to Job Replacement Have Not Changed Fundamentally So Far

However, as long as human labor has not been completely replaced or decided by artificial intelligence, or as long as “human-machine integration” has not been universally realized, some things will not change. Moreover, these unchanged aspects are becoming increasingly valuable, providing us with a time window.

The most important point is that humans are still the dominant party, still “telling machines what to do,” which is the fundamental reason for our confidence. This has both technical and institutional implications. In other words, the two fundamental ways for humanity to respond to job replacement have not changed fundamentally, although they need to keep pace with the times and constantly correct directions.

First, human capital is still the foundation for resisting AI impact, but humans need to know their strengths and weaknesses and adopt the basic strategy of leveraging strengths and avoiding weaknesses in the AI era. So far, human intelligence or natural intelligence still has advantages over AI in: (1) soft skills rather than hard skills; (2) non-cognitive abilities rather than cognitive abilities; (3) emotional intelligence rather than IQ; (4) humanistic understanding and empathy rather than problem-solving ability in math, physics, and chemistry, or even coding skills; (5) implicit knowledge rather than explicit skills.

Second, the social welfare system is still the fundamental safety net, and the material conditions for fulfilling such functions are increasingly enhanced. Marx observed from the development of early capitalism that once labor becomes a commodity, workers are institutionally bound to a fate of exploitation. When the Nordic countries established welfare states, they emphasized “de-commodification” in institutional design, weakening the attribute of labor as a purely private element and strengthening the social rights of workers and their families. Under conditions where the “job destruction” of AI increasingly surpasses and accelerates “job creation,” this concept and approach become increasingly important.

Several Paths for Workers Facing AI Employment Impact

Whether old methods or new ideas, the ways employment responds to AI replacement mainly include the following. Before summarizing these methods, we first give a reasonable presupposition that AI development will eventually increase labor productivity to an unprecedented extent. Based on this, workers usually have the following paths.

First, transitioning to positions requiring higher skills. This is an outcome that optimistic economists have always believed in, and it has been continually proven since the Luddite movement. However, this requires workers to have higher skills to match. In other words, those who obtain these new positions are usually not the same people who lost the old positions, not even the same cohort or generation.

In the future, the gap between losing old jobs and getting new ones will only widen. Many economists, including former U.S. Treasury Secretary Summers, have shifted from their previous confidence in technological progress creating jobs to recognizing that Luddism has its logic. Considering former U.S. Treasury Secretary Mnuchin’s still “optimistic” view on AI’s employment impact, and the rarity of such attitudes today, we can call these positions that will not naturally emerge “Mnuchin-style positions.”

Second, transitioning to industries with lower labor productivity and consequently lower pay. Kuznets’ normal direction of labor transfer to higher productivity sectors implies that positions with decreasing productivity belong to those with “reverse Kuznets” characteristics. Objectively, the new positions are less formalized than the previous jobs.  Subjectively, the decency of the new position is also lower than the original job. In short, job quality is reduced.

Third, transitioning to positions in industries with high demand elasticity. These are industries where people maintain great demand, yet their labor productivity is inherently hard to improve. Economist William Baumol exemplified this with performing arts. Whether such industries and positions can continue and expand depends on the demand for corresponding products and services and their elasticity. However, the scale and number of such positions will not infinitely expand.

Fourth, transitioning to positions induced by new consumption. Our current consumption, unimaginable a few years ago, simply did not exist earlier. The same goes for jobs. With future increases in labor productivity, people’s tastes will change, new things and ideas will continuously emerge, expanding consumption fields and diversifying job types. Since these positions arise from supply-side productivity improvements, creating a “supply creates demand” phenomenon, we can call them “Say’s law positions.”

Fifth, transitioning to positions redefined as employment. Activities previously not considered employment can now be socially recognized as such and compensated through transfer payments, supported by overall labor productivity. For example, if someone considers themselves a “writer” without published works or earnings, according to unemployment survey definitions, this state of “not being engaged in an hour or more of paid work in the past week” does not count as employment. However, if society can afford it, this can be considered employment.

Similar situations include those who do not declare themselves as “working” and are not seeking employment, such as those choosing to “lie flat” supported by other means. This includes two scenarios: One is where the individual has a source of support, such as relying on family and choosing to “lie flat.” The other is where the individual does not need employment but can receive universal social welfare support. For example, if a universal basic income system is implemented, it creates an environment where some people affected by job disruptions may choose not to participate in traditional work. Given that Keynes explored ways to share the benefits of productivity increases a long time ago, we can also refer to these as “Keynesian jobs.”

Response Strategies

Drawing on long-term experiences in economic history addressing technological replacement of employment, several principled suggestions can be distilled to guide technological development and market behavior through institutional construction, policy adjustments, and system reforms, aiming to maximize synchronization: 1. Synchronizing the pace of job destruction and job creation, making transitions feasible in numbers as much as possible. 2. Synchronizing productivity improvements across industries, avoiding the Solow paradox (IT improving productivity in only some industries, not permeating others). 3. Synchronizing AI labor replacement speed with training speed, minimizing reemployment friction periods. 4. Synchronizing productivity improvement and productivity sharing, embodying the unity of fairness and efficiency.

From a governmental function perspective, several important and urgent countermeasures can ensure these principles’ implementation.

First, accelerate building a Chinese-style welfare state. Emphasize: 1. Speeding up the improvement of the social welfare system with a sense of urgency akin to Moore’s Law. 2. Designing and perfecting welfare systems based on universal principles, moving away from strictly identifying welfare beneficiaries since the accelerated job loss era makes distinguishing those “lying flat” increasingly difficult. AI-driven labor displacement has strong externalities. 3. Using social mutual assistance, protection, and rights guarantees to offset the adverse effects of informal employment’s spread on workers.

Second, significantly extend the years of compulsory or free education. Competing with AI demands increasingly high human capital and a greater focus on non-cognitive abilities. Harvard’s Center on the Developing Child research shows the brain can establish over a million neural connections per second in the first few years of life, unmatched later. The optimal time for non-cognitive ability development is ages three to four, ideally extending compulsory education to this preschool age. Expected significant productivity improvements can substantially expand public educational resources, supporting longer school durations for children.

Finally, eliminate institutional barriers in children’s development, education and training, mobility and employment, social security, and other basic public services, particularly addressing systemic causes of migrant and left-behind children issues. Studies indicate parental care has irreplaceable roles in children’s human capital development, particularly in acquiring non-cognitive abilities and lifelong social mobility opportunities. Therefore, prioritizing household registration system reforms to address these issues should be paramount.