Will Artificial Intelligence Lead to Mass Unemployment or 'Less Work for More Pay'?
-
Could the future of human work simply be 'playing'?
In the future, people may have two ways of working:
The above is speculative, but not without basis.
The consensus in the industry regarding the impact of artificial intelligence (AI) on human jobs is that, similar to previous waves of machines replacing human labor, AI will eliminate certain positions while simultaneously creating new ones. However, the relationship, quantity, timing, and process between old and new jobs may undergo disruptive changes.
Last year, the American Economic Review published a paper titled The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and Employment, authored by Professor Daron Acemoglu from MIT's Department of Economics and Professor Pascual Restrepo from Boston University's Department of Economics. The former is widely considered a future Nobel laureate in economics.
Despite my limited English proficiency, I managed to read the entire paper with the help of Baidu Translate, as I have long been interested in the topic of 'AI's impact on employment.' Acemoglu's paper uses a model that effectively illustrates the dynamic process of job displacement and creation, which is highly valuable.
Traditional growth theory posits that growth depends on two endogenous factors—'capital' and 'labor'—and one exogenous factor—'technology.'
Acemoglu's framework examines how 'technology,' under the influence of 'capital,' affects 'labor.' Thus, the approach is purely economic, but the subject matter is highly relevant and forward-looking.
In his model, 'mass unemployment' and 'less work for more pay' could coexist as two possible scenarios.
Acemoglu divides AI's impact on the labor market into two parts: one where AI replaces human jobs and another where AI creates new jobs.
Technology enhances labor productivity, reducing businesses' demand for labor—this is AI's first effect: automation-driven substitution. This is the terrifying scenario depicted in many films where 'AI dominates humanity' or the tragic reality where 'capitalists use AI to amass wealth while the masses face unemployment and extreme wealth inequality.'
In reality, AI's substitution effect is merely a continuation of the 'automation replacing labor' trend that began with the Industrial Revolution. However, Acemoglu argues that incorporating capital into the equation changes the picture.
First, new AI technologies inevitably attract additional capital, leading to entirely new job opportunities—this is AI's second effect: technology-driven labor supplementation.
The chart above shows that between 1980 and 2007, total U.S. employment grew by 17.5%, half of which was due to new occupations.
Businesses invest capital in AI to replace human labor to improve returns. As efficiency rises, the remaining non-AI-replaced jobs see increased wages.
Some might argue that this is misleading because most of the wealth generated by AI replacing labor flows into capitalists' pockets, while displaced workers may never adapt to new roles, leading to declining incomes and accelerated societal inequality.
Acemoglu believes this is only a temporary imbalance in the first phase, as capital forces will eventually restore equilibrium. The process might unfold as follows:
Capital replaces labor with AI when labor becomes too expensive. However, if unemployment rises too much in a sector, wages will drop, making human labor relatively cheaper and reducing the incentive for AI adoption.
This leads to an 'AI technology plateau,' during which a completely opposite process emerges in adjacent industries.
As capital flows into new AI-created roles (e.g., in software), high-paying jobs appear due to labor shortages. These attract talent from other industries, which in turn creates vacancies that absorb displaced workers at various levels.
Thus, workers displaced by AI don’t necessarily need to reskill. After a period of unemployment, they can quickly find similar jobs—much like the opportunities brought by Fuyao Glass in American Factory.
This is the 'AI technology plateau,' where, like a queue, everyone moves forward to find a better position, eventually reaching a new equilibrium.
Later, as labor costs rise in the industry, a technological breakthrough may disrupt the equilibrium, reintroducing cost-effective AI solutions and restarting the cycle of automation.
In summary, AI's transformation of human employment will be a long, gradual, and irreversible process, with capital, labor, and technology mutually constraining each other. Periods of technological leaps and plateaus will alternate:
When capital's long-term rental rate is cheaper than labor, it's a 'technological leap,' where automation accelerates and labor becomes redundant. This continues until labor becomes cheaper than capital's long-term rental rate, entering a 'technological plateau' with the complex changes described earlier.
Overall, Acemoglu's model is optimistic. AI-induced unemployment is merely temporary pain during rebalancing. Although AI eliminates more jobs than it creates (due to rising labor efficiency), increased capital returns benefit more people through taxation. Thus, society as a whole works fewer hours but enjoys higher average incomes.
It seems the world is moving toward 'less work for more pay,' but what will this future actually look like?
Acemoglu's model doesn’t provide a clear answer. However, the paper reminded me of my firsthand experience with 'AI-induced unemployment' a few years ago—Alibaba's transformation of ad design.
A marketing campaign involves pushing numerous ads to consumers, a complex process spanning 'strategy—creativity—production—distribution,' handled by different ad professionals.
The biggest issue is the complexity and time required. When I first joined the ad industry, serving an automaker, ad planning for a new model began a year before launch.
But since 2014, I witnessed AI reshaping the industry—Alibaba's 'Luban AI' for Taobao sellers allowed users to define key elements, and the system auto-generated ad banners, eliminating the need for designers and copywriters.
It even replaced 'creative directors,' as the system provided real-time performance testing, letting users select high-click, high-conversion, or high-engagement options for different ad goals.
In theory, this system helps small businesses save labor, mid-sized sellers improve ad performance, and large brands optimize workflows and decision-making.
Clearly, if 'Luban AI' matures and becomes cost-effective, it could expand beyond Taobao, erasing many ad industry jobs and causing mass unemployment.
On the other hand, the system elevates the importance of 'strategy' and 'evaluation,' creating new roles.
First, generating ads requires precise keyword definitions—a task previously scattered across planning, design, and copywriting. In the AI era, this is centralized, requiring a dedicated role:
A new job emerges, possibly still called 'planner,' but with a completely different focus—deeply understanding and collaborating with the AI system.
Of course, I believe that with the advancement of 'machine learning,' future AI design won't require 'defining keywords.' It should actively learn your needs from your past advertisements—meaning this 'new position' could also be replaced.
But the problem is, 'strategic thinking' in the advertising industry will never disappear and cannot be learned by AI. It can only be pushed forward to earlier stages like product strategy, consumer strategy, channel strategy, and even corporate strategy—higher-level thinking domains.
The same applies to the 'testing and evaluation' phase in AI design. Previously, 'ad testing' was cumbersome and of limited accuracy, used only for major campaigns. But with AI design, 'ad testing' will become standard. This will require designing test plans for AI and interpreting results, effectively creating another new position.
Then there's 'ad performance evaluation,' which was previously just an ancillary task for media personnel, mainly to enrich final reports and impress clients. In the era of AI creativity, it will directly impact 'machine learning' capabilities, effectively becoming another newly created role.
Looking back, every time AI 'conquers' a work segment, it generates new positions upstream and downstream to make its own work more efficient. Of course, new roles will also include those directly serving AI, like 'artificial intelligence' and 'machine learning.'
Many might ask: After all this, it seems the new roles are mostly in planning and analysis—have creative professionals been eliminated? In reality, creative professionals haven't been eliminated; they're just doing work that better utilizes their abilities.
This is called 'play'!
In the foreseeable future (the 'weak AI' era), AI cannot create any new ideas. It merely breaks down existing ideas into data and relationships, then recombines them. It doesn't understand these outcomes; it only grasps data correlations, not logical causality—the latter being a uniquely human ability.
Not only can AI not create new ideas, it also needs to be 'fed' massive amounts of fresh creativity to work efficiently. Currently, AI is fed with works accumulated over millennia, but future competition will demand even more new ideas. Generating these new ideas will become the creative professionals' new job.
In past and current advertising workflows, creatives had to engage in commercial thinking, understanding how clients—not consumers—viewed their ideas. Some even earned high salaries and promotions based on this skill rather than genuine creative talent. This model wasted the potential of many talented creatives. Thus, AI is their liberator, not their executioner.
Once AI becomes the bridge between creatives and clients, creatives will be freed from commercial thinking—which never truly belonged to them—and return to purer artistic innovation. How these ideas are commercialized will be AI's concern.
I don’t know what future creative work will look like—that’s territory for sci-fi writers. It might resemble today’s painters or writers, working in complete freedom, or it might take an entirely new form. How will they earn income? I don’t know either. It likely won’t be monthly salaries or traditional models like selling paintings or collecting royalties. A new, unimaginable payment method will likely emerge.
In the highly specialized industrial era, idle hands were shameful. But future creative work may blur the line between work and leisure while still generating income.
During the recent U.S. election, Democratic candidate Andrew Yang proposed a core platform of '$1,000 per month to address AI challenges.' While seemingly populist, this idea is remarkably forward-thinking given AI’s irreversible impact on employment—hence his strong support from Silicon Valley, including Elon Musk.
Universal basic income, now a temporary relief measure, may become a future 'national income.' On this foundation, everyone could earn additional income through 'play' rather than formal work.
Work feels tedious because it involves repetitive, monotonous tasks and forces us to suppress our personalities to survive in authoritarian corporate structures. AI excels at the very tasks we dislike—it’s efficient at repetitive work and has no personality to sacrifice. Once these tasks are offloaded, the nature of human work may fundamentally change.
This may explain why 'total societal labor hours decrease while average income rises.' Labor will become a true right for workers, not an obligation.