Main Text
Original link https://1q43.blog/post/12190/
If you are anxious about AI causing massive unemployment, then consider this question first:
In 2025, how many people have jobs that do not require any use of computers or smartphones?
If your answer is very few, then you can read the article while also reflecting, as we will calculate this number and percentage in the middle of the article.
In the past two years, anxiety about AI has been almost like a “super contagious disease.” We worry about ourselves, white-collar workers who have made PPTs and Excel sheets for ten years, whether we will be replaced within three years by a graduate who is better at prompts or even by an Agent. We also worry about our children—whether teaching them painting, programming, or finance now will still matter when they graduate from university 20 years later.
This anxiety is not baseless—regardless of whether AI ultimately leads to universal welfare or a post-utopia, almost all authoritative organizations emphasize its short-term pains.
For example, the World Economic Forum (WEF), in its latest “Future of Jobs Report 2025,” bombards us with various charts, with the core idea being “structural transformation” and “skills turnover.” PwC’s “Global AI Employment Barometer 2025” states more bluntly that in industries “more susceptible to AI impact,” each employee’s income growth is three times that of other industries, and the speed of skill change has increased by 66%.
Even OpenAI, the “originator” of AI, does not forget to “pour fuel on the fire.” In their report “Work in the Era of Intelligence,” they enthusiastically share how Walmart uses large language models to handle product data, claiming that “without generative AI, the same work would require nearly 100 times the current manpower to be completed in the same time.”
Translated, this means: the technology is awesome, change is here, if you don’t learn, you’re out.
This logic is so airtight and unquestionable that “lifelong learning” and “embracing AI” have become the only “politically correct” attitudes of our time.
But…
There is always a “but.”
This “but” is: we assume a premise—that AI is a more advanced form of “intelligence,” which will replace human “intelligence.” The reality is that a significant portion of jobs in this world fundamentally do not require that kind of “intelligence,” not even human existing “intelligence.”
Or put differently, they do not require “cognitive intelligence.” They require “physical intelligence.”
A software engineer completes all work in the digital world. AI, as a more efficient digital processing tool, can certainly assist, enhance, or even replace them.
But what about a construction worker? Their work involves carrying, masonry, binding; what they need to learn is how to use their waist to avoid injury. A kitchen helper cutting vegetables, a security guard patrolling a neighborhood, a farmer planting rice seedlings, a worker screwing bolts on the production line, a cleaning lady tidying an office—these jobs share the characteristic of operating mainly with “physical atoms” in the physical world.
Of course, they may use smartphones to watch short videos or chat with family on WeChat. But in their core workflow, they need a pair of hands, a pair of legs, and a presence in the physical world. Unless humanoid robots can rapidly improve and drop to a very low price in a very short time, this will almost not affect their jobs.
Injecting AI’s “digital intelligence” into “physical atoms” is extremely costly. Discussing whether AI will replace programmers is a “Yes/No” issue; discussing whether AI will replace security guards is a question of “ROI” (Return on Investment). In the latter field, human physical costs remain astonishingly low.
So, how large is the group represented by this “but”?
To understand this, I had AI run two estimation reports: one about China, and one about the globe. I deliberately asked them not to cite grand reports stating “AI will impact X% of jobs,” but rather to estimate a “lower bound” based on the most fundamental global labor structure data (e.g., data from the International Labour Organization (ILO) and various national statistical bureaus).
How many people’s jobs are corely “non-digitally dependent” (completely do not rely on phones and computers)?
The result is startling; let’s first look at China’s situation:
This report uses two cross-checking approaches based on the benchmark of 740 million employed people in China in 2023.
The first is a “top-down” reverse estimation method. It relies on the officially released 2024 Report on National Digital Literacy and Skills Development Levels. The report shows that among all employed people nationwide, 67.85% possess basic or higher digital literacy.
Reversely estimated, 32.15% of employed persons (about 238 million) lack basic digital literacy.
Note, this is only a “lower bound.” It calculates how many people are unable to participate in digital work due to skill deficiencies. But in reality, many people (like neighborhood security guards) “possess” the skills (can use Excel) but their “positions” do not require it.
Thus, the second “bottom-up” estimation is closer to reality. It does not care about what people “can do,” only what the “job” requires.
It looks separately at the three industrial sectors:
-
Primary industry (Agriculture): About 163 million employed. This estimate assumes at least 85% of the work (such as field labor in planting and breeding) is purely physical labor. This part constitutes 138 million people. They might use phones to check weather or sell goods, but “farming” itself is non-digital.
-
Secondary industry (Industry and Construction): Approximately 212 million employed. This excludes highly automated “smart manufacturing” factories and management positions. The estimate mainly captures construction workers, low-end manufacturing, mining, and other manual laborers. This part totals about 106 million people. They are the brick carriers on construction sites and assembly-line workers performing repetitive labor.
-
Tertiary industry (Services): The most complex part, totaling about 358 million people. This includes highly digital finance, IT, and research sectors (about 100 million), middle ground such as teachers, doctors, drivers, and purely physical laborers. This estimate conservatively extracts physical laborers in “on-site service” jobs, such as housekeeping, cleaning, security, and kitchen staff. The estimate puts this group at about 50 million people.
Adding these three parts: 138 million (agriculture) + 106 million (construction/industry) + 50 million (services) = 294 million people.
The conclusion is: In China, about 4 out of every 10 workers (32.15% to 39%) have jobs whose core processes do not involve operating computers, tablets, or smartphones at all.
This does not even include delivery workers and ride-hailing drivers, who hardly require much operation but are strongly network-related new types of jobs.
The yellow box parts are estimated by AI using real statistical data and specific calculation methods, repeatedly computed under the Deepresearch mode with the same Prompt on different Gemini and ChatGPT products, with proportions fluctuating between 35%-45%, but never less than 35%.
Has this number already disrupted your perception of a “digitized China”?
Wait, China’s labor digitalization degree is actually quite high. Let’s now look at the global context.
The global estimate used the same “bottom-up” labor sector decomposition method, with primary data from the International Labour Organization (ILO), using a denominator of 3.7 billion total global labor population.
It divides the global “physical labor force” into three major blocks:
-
Agricultural workers: Still the largest “physical labor” group worldwide. Estimated about 829 million people, they constitute 38% of the global physical labor force.
-
Informal service workers: The “background color” of the global economy. Includes domestic workers, street vendors, small merchants, artisans, etc. They are the main body of the global poor population, and their work heavily relies on physical strength. Estimated about 766 million people (35%).
-
Industrial manual laborers and manual service workers in formal sectors: Including construction workers, workers in non-digitized factories, and service sector cleaners, security, etc. Estimated about 573 million people (27%).
Summing these: 829 + 766 + 573 = 2.168 billion people.
The conclusion is: By 2025, about 58.6% of the global workforce will have jobs that do not require computer operation at all.
The yellow box parts are estimated by AI using real statistical data and specific calculation methods, repeatedly computed under the Deepresearch mode with the same Prompt on different Gemini and ChatGPT products, with proportions fluctuating between 45%-60%, but never less than 45%.
At least half.
This means that the personal computer revolution (starting in the 1980s), the Internet revolution (starting in the 1990s), and the mobile Internet revolution (starting in the 2010s)—these three huge waves over nearly half a century—have not even managed to “digitize” half of the global jobs.
We, who work 10 hours every day in front of screens and worry about AI dynamics, are just less than half of the “digital labor force.” We mistakenly believe we represent the whole world, but in reality, we are not even the majority.
Now, we can look back at the impact of AI.
Of course, it is inaccurate to say AI has no impact on this nearly 50% “physical labor force.” For example, for drivers in the transportation industry, the emergence of ride-hailing services a few years ago caused unprecedented crises for traditional taxi drivers. But in terms of labor skills, ride-hailing drivers do not have obvious advancement from taxi drivers; it is more a shift than a replacement for that position.
It can even be said that mobile internet has created three huge “employment safety nets” from nothing: delivery workers, ride-hailing drivers, and couriers.
Most likely, the AI era will be similar. AI will reorganize digital workflows and create and explore new demands previously non-existent (for example, AI generating 1,000 kinds of personalized product schemes), but AI itself cannot cover all the “physical links” needed to realize these new demands (for example, who will complete sampling, packaging, and delivery of these 1,000 personalized products?).
These “physical gaps” created by AI efficiency gains but uncovered by AI itself are precisely one source of new jobs created by AI. Yet, even so, a laughable paradox still appears:
A “non-digital-dependent” worker who has in the past 30 years missed the computer wave, the Internet wave, and even just used a smartphone as an entertainment tool in the smartphone era…
He is actually in the safest position in this wave of AI.
The impact of AI targets precisely those who “won” in the previous wave—the people who rely on “digital intelligence” work.
That financial analyst analyzing data in a cubicle is more at risk than the construction worker plastering walls on site. That graphic designer retouching photos at the computer is more at risk than the kitchen helper chopping ingredients. That programmer tapping keys is more at risk than the cleaning lady taking care of the office.
Because the cost of replacing “digital work” with AI (computing power, electricity) is rapidly declining, while the cost of replacing “physical work” (robot hardware, maintenance) remains high.
And this exactly reveals the nature of “AI anxiety”—at its root, it is a kind of elitist anxiety, a modern version of “why don’t you eat meat porridge?”
Why do I say so?
Those “physical” jobs, which nearly half of the global population rely on, usually mean what? We know well: lower income, worse environment, heavier physical exhaustion.
The 2.1 billion people worldwide, or 300 million in China, have always lived under the “danger” described in this narrative. Not the danger of being replaced by AI, but danger from poverty, occupational injuries, heatstroke, severe cold, and physical exhaustion.
But in the grand narrative of the “AI revolution,” this half is the silent majority. Their “danger” is regarded as a backdrop or an unacceptable outcome.
Now, when AI appears, that 40% “digital labor force” (i.e., us) begins to fret. What are we anxious about? We worry about “losing jobs.”
But what we truly fear is “downshifting”—we fear having to leave our comfortable air-conditioned rooms to do the physical labor we used to “ignore,” we fear going from “analysts” and “programmers” to “construction workers” and “cooks.”
This is itself a perspective that does not regard the majority as human.
So, this fear of “downshifting”—this anxiety of “I don’t want to be a construction worker”—what it should truly drive is probably not us rampantly overworking ourselves learning AI or competing for ever fewer “digital” positions within that 40%.
Instead, it should drive us for the first time to truly face the reality of that 60%. It should push us to think about: Why are the environments and conditions of “physical labor” so poor that we treat it as “the apocalypse”?
This fear is exactly the best motivation to improve the working conditions of those 300 million or 2.1 billion laborers.
To put it more harshly, perhaps AI’s greatest merit is indeed to sweep everyone out of the “mental labor” illusion and make everyone "downshift” to the realm of “physical labor.”
Because only then will “improving labor conditions” become a social consensus for the first time.
After all, when everyone becomes a construction worker, those former meritocrats will no longer be able to ignore others’ heatstroke subsidies and occupational injury insurance with the nonsense “I work in an office because I studied hard back then.”
Therefore, when we talk about “AI changing work,” we may all have misunderstood one thing. What AI brings may not be a “cognitive revolution” targeting all humanity, but more like an “internal reshuffle” and “class anxiety” of that 40% “digital labor force.”
It cannot change the physical reality of those 300 million Chinese workers, nor the 2.1 billion global workers.
Viewed this way, AI might indeed change (our) jobs, but saying it will change (everyone’s) jobs is still far from the truth.
Next time some “digital elite” tries to sell you AI anxiety, urging you to “embrace change” and quickly buy courses,
You can just nod and then ask back, “So, have you… signed up for New Oriental’s cooking class?”