Artificial Intelligence: Helping the Majority Break Through the Lower Class or Forever Trapped at the Bottom?

Author: Zhang Feng

I. The Basic View of The New Yorker: “Artificial Intelligence Will Cause Most People to Become Permanently Marginalized”

In a widely circulated article in The New Yorker magazine, a disturbing future scenario is depicted: as artificial intelligence rapidly advances, society will split into a tiny elite that controls AI technology and a vast “useless class,” with most people permanently relegated to the social bottom. The core logic of this view can be summarized as follows:

First, AI will replace a large number of white-collar and knowledge-based jobs. Unlike previous industrial revolutions mainly replacing physical labor, AI directly impacts cognitive work, analysis and judgment, and to some extent, creative tasks. Traditional middle-class professions such as lawyers, accountants, programmers, doctors, and teachers may be massively replaced by AI.

Second, the speed of technological iteration far exceeds the pace of labor force transformation. Historically, the spread of steam engines and electricity took decades or even a century, whereas AI capabilities can leap qualitatively every few months. People have little time to learn new skills before they become outdated.

Third, the monopoly of capital over technology will intensify inequality. Large corporations that control AI technology and computing resources will become new “feudal lords,” while ordinary people will find no bargaining power in this system because AI is cheaper, more efficient, and more stable than any human.

Fourth, the so-called “creating new jobs” logic becomes invalid. While past technological revolutions eliminated old jobs, they also created more new ones. However, AI not only replaces physical work but also mental work; new roles tend to be either extremely high-end (accessible to only a few) or quickly swallowed by AI. Ultimately, most people lose their participation value in the economy and can only rely on basic income to survive, becoming “pets fed by algorithms.”

This view is not alarmist; it has sparked profound anxiety among academia, the tech industry, and policymakers. But if we examine the essence of artificial intelligence more carefully, we will find that The New Yorker’s conclusion is based on a fundamental misjudgment—it treats AI as an external force replacing human brainpower, without recognizing that AI is essentially infrastructure for mental labor.

II. The Rationality and Irrationality in The New Yorker’s Logic

Rational aspects. First, we must acknowledge the rational components in The New Yorker’s perspective. AI indeed will have a dramatic impact on employment, supported by abundant evidence. Large language models like GPT-4 perform tasks such as code generation, text writing, data analysis, and even legal consulting at levels close to or surpassing ordinary professionals. A 2023 Goldman Sachs report estimates that about two-thirds of jobs in Europe and America are at risk of AI automation, with a quarter to half of their tasks directly automatable.

Second, the speed of technological replacement is unprecedented. During the Industrial Revolution, it took two generations for textile workers to transform; AI went from failing the Turing test to passing the bar exam in less than ten years. This exponential pace makes traditional “retraining and reskilling” models ineffective.

Third, the trend toward concentration of wealth and power is indeed worrying. A few companies like OpenAI, Google, and Microsoft have gained significant advantages in foundational models, computing power, and data. Once this monopoly solidifies, ordinary people may indeed lose influence in the economy.

Irrational aspects. However, The New Yorker’s logic contains a fundamental fallacy: it equates “AI replacing certain labor” with “the labor’s human executors becoming useless.” This assumption overlooks the fact that in economic systems, labor and production technology are not simply substitutes but are involved in complex restructuring relationships.

The first irrational point is the trap of “zero-sum thinking.” Viewing AI as a competitor “stealing jobs” from humans is itself an outdated industrial-age mindset. In fact, every technological revolution has eliminated old professions but also unleashed new demands and possibilities. In the 19th century, mechanization of agriculture reduced employment from 80% to less than 2%, yet there was no 80% unemployment—people shifted to manufacturing, services, and later to “knowledge workers” unimaginable before. AI will similarly create new professional fields beyond our current imagination.

The second irrationality is ignoring the diversity of human labor value. The perspective in The New Yorker implicitly assumes that economic value only exists in productivity that can be measured by efficiency. But human creativity, emotional connection, ethical judgment, aesthetic experience, community building, and educational companionship—many activities—are still not fully or efficiently replaceable by AI. As AI’s efficiency increases, these “inefficient but unique” human abilities will become even more precious.

The third, and most critical, irrationality is a misunderstanding of AI’s essence. The New Yorker views AI as a form of “superintelligence,” as if it were an independent entity capable of taking over all human mental work. But real AI is not “another form of intelligence”; it is an “infrastructure for extracted and industrialized mental labor.” To understand this, we need to analyze AI’s fundamental characteristics.

III. The Essence of Artificial Intelligence: Infrastructure for Mental Labor

An analogy: the industrial revolution was the infrastructureization of physical labor. To understand AI, we must revisit the Industrial Revolution. It was not a mysterious “machine age,” but the industrialization of general repetitive mechanized physical labor.

Before the revolution, forging a shovel required the skill of a blacksmith—strength, rhythm, and angle of hammering, accumulated as body knowledge over generations. The revolution, through steam engines, stamping machines, and assembly lines, extracted these repetitive, rule-based physical actions from individuals, standardizing, mechanizing, and scaling them. As a result, what once took ten years of apprenticeship could be learned by a farmer in two months to operate machinery.

This is not “machines replacing humans,” but “physical labor capabilities becoming a basic infrastructure accessible to all.” You don’t need to be a blacksmith; just access the industrial system, and you can produce far more than a blacksmith. The revolution turned “physical strength,” once a scarce personal ability, into a cheap, universal public resource.

The consequence was not impoverishment of workers but, on the contrary, a historic and sustained rise in living standards. Once the bottleneck of physical labor was broken, humans shifted focus to organization, design, management, and innovation—those truly requiring human uniqueness.

AI: the infrastructureization of general repetitive mechanized mental labor. AI is a continuation of this logic in the realm of mental work. The essence of AI is the infrastructureization of general repetitive mechanized mental labor.

What is “general repetitive mechanized mental labor”? Let’s break it down:

General: Not the genius-level creativity of Einstein’s relativity, but the routine problems faced by ordinary professionals—writing a business email, summarizing a meeting, translating text, writing standard sorting code, analyzing basic trends in financial data, identifying common lesions in medical images.

Repetitive: These tasks have clear patterns, with highly similar approaches across many cases. A doctor judging 1,000 CT scans applies similar logic each time; a programmer writing 100 sorting functions uses similar structures.

Mechanized: Tasks have explicit rules, methods, and procedures, describable in “if-then” logic or algorithms. The steps are definite, with a clear input-output mapping.

This type of mental labor constitutes the main part of white-collar work today. It requires expertise, training, and thinking—but not the most advanced creative breakthroughs or work involving deep emotional connection and complex contextual judgment.

AI, through large-scale pretraining, deep neural networks, and reinforcement learning, extracts these mechanized mental tasks from human brains, standardizing them into callable, near-zero marginal cost services. You don’t need to learn accounting or memorize all tax law provisions; just describe your problem to AI, and it can perform tax calculations that previously required a professional accountant half an hour.

This is not “AI replacing humans,” but “mechanized mental capabilities becoming infrastructure accessible to everyone.” Just as the industrial revolution gave everyone access to metallurgy skills once exclusive to blacksmiths, AI is enabling everyone to access the “computing” and “analytical” mental powers once limited to specialists.

Why does this give most people more opportunities? Understanding AI’s essence reveals why it benefits most rather than oppresses most.

First, AI greatly lowers the entry barrier for knowledge and professional skills. In the past, becoming a data analyst required learning statistics, programming languages, databases, and hundreds of hours of training. Now, a marketing person can directly ask AI in natural language: “Analyze our sales data from last year and find the most common product combinations.” AI not only provides answers but also explains the analysis methods. This means expertise is no longer a scarce resource; the real scarcity is “the ability to ask the right questions” and “judge the quality of answers”—skills that ordinary people can gradually develop.

Second, AI releases humans from repetitive mental work. A doctor spends much of their day writing medical records, reviewing routine images, and reading literature—these account for 70% of their work, and are precisely mechanized mental tasks. When AI takes over these, doctors can focus on what only humans can do: deep communication with patients, personalized treatment planning, and medical innovation. Doctors won’t decrease; they will become more valuable—because they can now concentrate on parts AI cannot replace.

Third, the near-zero marginal cost of AI will democratize high-end mental services. In the past, only large firms could afford top law firms, McKinsey consulting, Goldman Sachs investment banking. Now, small entrepreneurs can use AI to generate legal drafts, write business plans, analyze financial statements. This does not eliminate these professionals’ markets but expands the overall market size—cost reductions lead to explosive demand, and professionals will find more high-quality work collaborating with AI.

Fourth, AI significantly boosts individual productivity. What a person could do before is limited; now, with AI, one can accomplish what previously required a small team. This will not cause unemployment but will spawn countless micro-entrepreneurships and individual economies. Someone can be simultaneously a product manager, designer, programmer, and marketer, as AI provides strong support in routine tasks. Creativity, judgment, and responsibility—core human traits—become more important, while the barriers to realizing them are lowered.

IV. Future Social Forms and Division of Labor

When AI becomes a comprehensive mental infrastructure, human society will enter a new organizational form. This is not utopian fantasy but a rational projection based on current technological trends.

Basic material needs can be distributed on demand. With AI-driven productivity, the on-demand distribution of basic material needs is no longer a dream. Why?

Intelligent production at the source. AI dispatch systems can optimize procurement, production planning, and logistics, greatly reducing waste and inventory costs. In manufacturing, smart systems can automatically adjust production lines instantly based on demand.

Energy efficiency revolution. AI applications in grid dispatch, energy consumption forecasting, and renewable energy integration will continuously lower energy per unit of GDP. As energy and computing become cheaper, the marginal cost of producing material goods approaches raw material costs.

Mature automated production systems. Combining AI control with robotics, the entire process from raw materials to final products can be highly automated—similar to today’s ubiquitous “tap water”—you don’t need to know how the water plant works; turn on the tap, and water flows, with minimal cost based on usage.

When the marginal costs of most basic necessities (food, clothing, housing modules, transportation, household appliances) fall sufficiently low, society can fully realize on-demand basic material distribution. This is akin to Nordic countries’ provision of basic education and healthcare—not luxury, but a baseline of decent living.

It’s important to emphasize that “on-demand distribution” does not mean “distribution by claim.” It should be understood as a safety net—above which people can still acquire more resources, experiences, and recognition through creative activities.

Spiritual needs and creativity become core values. Once material bottlenecks are removed, what becomes scarce? Meaning, experience, creation, relationships, and aesthetics. These areas are precisely where AI’s limitations lie—not that AI cannot do them at all, but that no matter how well AI performs, it cannot replace the significance of human participation.

Why attend a live concert instead of an AI-generated perfect performance? Because “the specific person’s performance at that moment” is meaningful. Why watch the Olympics? Because the real, flesh-and-blood person pushing their limits moves people. Why chat face-to-face with friends instead of with AI? Because the other is “another free, self-aware subject.”

Activities such as art creation, frontier scientific research (true exploration, not literature review), education (especially in values and aesthetics), community building, psychological therapy, sports, crafts, and philosophy will become the main pursuits and sources of value in future society.

Social division of labor will shift from “finding a job” to “finding a mission”: When material needs are secured, engaging in activities is no longer primarily for survival but for meaning, challenge, flow, and self-fulfillment.

From “executor” to “definer, evaluator, integrator”: AI can write code, but humans must define “what software to develop, what problems to solve.” AI can generate design options, but humans must judge “whether these solutions fit the project’s character.” AI can gather vast information, but humans must synthesize it into stories with warmth.

From “efficiency competition” to “uniqueness competition”: Competing with AI on efficiency is always losing; but “my unique perspective, experience, emotion, and judgment” cannot be replicated by AI. The core individual competitiveness in the future will be “why should I be the one to do this?”

This means that future social stratification will no longer be “those with AI” and “those without AI,” but “those who can fully collaborate with AI to unleash their creativity” and “those who have not yet learned this.” The latter are not the bottom but potential groups waiting for liberation. This is the mission of education.

V. Preventing Monopolies: The Conditions for Coordinated Development

However, these bright prospects are not automatic. They depend on AI development and governance following the right path. If AI is monopolized by a few companies and becomes a new privilege tool, The New Yorker’s prophecy may self-fulfill. Therefore, a series of supporting technological developments are necessary.

Synergy with Web3: Preventing value monopolization. Web3’s core value is decentralized ownership and governance. Combining AI with Web3 can prevent monopolies over computing power, data, and models.

Decentralized computing markets: Using blockchain technology, ordinary people can contribute idle GPU computing power for tokens, enabling large model training without relying solely on data centers of a few companies. Although decentralized training faces technical challenges, decentralized inference computing is already feasible.

Data ownership and contribution proof: The value of large amounts of data generated during AI interaction should be returned to users. Blockchain can enable transparent tracking and value sharing of data contributions. Currently, data used for training large models (like public texts on the internet) is free; if in the future everyone can choose to contribute their interaction data and receive rewards, AI evolution becomes a process where all participate and all benefit.

Open-source model protection and development: Meta’s Llama, Alibaba’s Tongyi Qianwen open-source versions show that high-performance AI models do not necessarily need to be closed. Web3’s incentive mechanisms can provide ongoing support for open-source developers, avoiding “winner-takes-all.”

Synergy with quantum technology: Breaking the monopoly of computing power. The advent of quantum computing could fundamentally disrupt current AI computing power structures. Quantum parallelism and exponential speedups for specific problems can make training large models less dependent on traditional chips. This could break the current barriers built by Nvidia, TSMC, and others, enabling more research institutions, SMEs, and even individuals to train large models.

More importantly, quantum key distribution and quantum random numbers can create truly secure, unpredictable AI systems, preventing nightmares of “super AI surveillance societies.”

Digital governance: Preventing AI power abuse. Technology itself is neutral; governance determines its direction. Future AI societies need mechanisms such as:

Algorithm transparency and auditability. Everyone should have the right to know the basis of AI decisions, especially in scenarios involving personal interests (credit, employment, healthcare). Regulations should mandate explainable outputs.

Anti-monopoly and interoperability. Large AI platforms should open interfaces to third-party developers, allowing low-cost switching between services, preventing lock-in effects—similar to number portability in telecom or one-click transfer in banking.

Digital identity and autonomous data rights. Everyone should have their digital identity and data sovereignty; AI can only access data with explicit permission. This not only protects privacy but also prevents AI from becoming tools of surveillance and manipulation.

Universal basic computing resources (UBC). Similar to universal basic income (UBI), society could provide each person with a daily “free AI computing quota”—for example, 100 questions per day, free basic voice synthesis, image generation, and other services. This ensures even the poorest are not excluded from AI infrastructure.

Technology and civilization collaboration. The synergy of AI, Web3, quantum tech, and digital governance aims to keep productivity tools aligned with “human-centered” principles. We need not a super AI controlled by a few, but an open, auditable, low-threshold, inclusive mental infrastructure—like today’s power grid—anyone can plug in and use, and no one can monopolize to enslave others. The future of AI should be like that.

The concern in The New Yorker is profound and warrants vigilance; it reminds us that technology will not automatically bring justice. But to claim that AI will cause most people to become permanently marginalized reveals a fundamental misunderstanding of AI’s nature. AI is not another “superintelligent species”; it is the industrialization of human mental labor. It is infrastructure, a tool, an amplifier of capability. Its true historical significance is not to replace humans but to free humans from repetitive mental work, enabling everyone to achieve higher levels of creativity, judgment, and emotional connection at lower costs.

In the future, basic material needs will be met by highly automated AI systems on demand, and humans will focus unprecedentedly on spiritual creation and meaningful pursuits. Our vigilance should not be against AI itself but against the possibility of AI monopolization. Through Web3’s distributed governance, quantum breakthroughs, and transparent digital regulation, we can forge a path of “human-machine collaboration and universal benefit.”

Every historical turning point has seen predictions that new technology will destroy opportunities for most people. But history repeatedly proves that when technology becomes a true infrastructure rather than a shackle, the opportunities it releases far outweigh the jobs it displaces. AI will not cause most people to become permanently marginalized—in fact, it will give most people their first real chance to escape survival pressures and become the masters and creators of their own lives and meanings.