Our long-standing imagination of an AI society has been heavily influenced by science fiction: thousands of humanoid robots in work uniforms welding, transporting, and assembling on assembly lines, as if simply putting a human-like shell on a mechanical arm would complete the “machine replacement” of human labor. However, the reality is completely different. A highly automated AI society does not require so many physical robots; what it needs is a complete set of invisible, self-sustaining automated processes. What is truly consumed are energy, computing power, data, and the attention of a very small number of human decision-makers. This is the real material basis of future productivity. 1. Productivity: From ‘Robot’ to ‘Process Itself’
1. Physical robots are merely marginal players. In chip factories, Amazon warehouses, and Tesla factories, mechanical arms do exist, but they are the clumsiest, most expensive, and highest maintenance components of the entire automation system. What really does the work is the software-defined processes: photolithography machines self-calibrate, self-change targets, and self-optimize parameters; logistics systems predict demand, allocate containers, and plan drone routes autonomously; even agriculture has transformed into “algorithms + nozzles + satellite imagery,” with no humanoid robots seen in the fields.
2. Production materials are completely “dematerialized.” In the traditional industrial era, production materials were land, factories, and machines; in the AI era, the core production materials have become:
- Electricity (sustainable, cheap, green power)
- Computing power (data centers, photonic chips, distributed inference networks)
- Data (continuously flowing, updated training/inference data)
- Attention (the very few humans who define the objective functions)
Note the last point: robots do not need salaries, but large models require humans to tell them “what we want to optimize today.” This “goal-setting power” will become the scarcest production material.
3. The true source of productivity explosion is not because machines are faster than human hands, but because AI has reduced “coordination costs” to nearly zero. The upper limit of human organization is determined by communication costs (Dunbar’s number, bureaucratic overhead), while AI systems can instantly coordinate millions of sub-tasks with zero friction, zero misunderstanding, and zero office politics. This is where productivity truly takes off, not because a mechanical arm lifted ten times more boxes than a human.
2. Production Relations: Who Owns the ‘Process’ Owns Everything
As the core of productivity shifts from “things” to “processes,” production relations also undergo a dramatic change.
1. Capitalists → Process Owners. Traditional capitalists own factories, machines, and land; future capitalists will own:
- The optimal objective function (that tells the AI system “what to do”)
- The largest data feedback loop (the more users, the smarter the system, leading to winner-takes-all)
- The cheapest energy and computing power channels
You will see a new type of “super capital” emerge: neither traditional energy giants nor chip manufacturers, but those who can feed billions of user behaviors to models in real-time and adjust incentive structures instantly.
2. Proletariat → “Data Proletarians” and “Attention Proletarians.” Most people will no longer earn income through “labor” but will survive by being “optimized.”
- Your search history, consumption records, and even heartbeat data contribute to the next optimization of a model; you are free raw material.
- Your attention generated while scrolling through short videos, playing games, or even dating has become the most valuable scarce resource, being precisely harvested.
3. New class differentiation: those who can define objectives vs. those who are defined by objectives. The most brutal divide in future society will be:
- A very small number of people (0.001%) retain the authority to “give instructions to AI”; they decide what the entire society optimizes (profit? happiness? carbon emissions? birth rates?)
- The remaining 99.999% can only seek their optimal solutions within the established objective functions, gently placed by processes like NPCs.
This is not a conspiracy theory but an inevitable outcome of technological logic: any complex AI system ultimately requires a final loss function, and whoever has the authority to modify this function holds the actual sovereignty of this civilization.
3. Possible Future Scenarios
1. Extreme abundance + extreme boredom. Goods and services tend toward being free or nearly free (abundant energy, perfected automated processes), but most people lose the feeling of being “needed.” Universal Basic Income (UBI) will arrive, but it addresses survival issues, not meaning issues.
2. “Anti-automation” becomes the new luxury. Handcrafted coffee, handmade furniture, human companionship, and human creation will become the ultimate way for super-rich individuals to show off, just as we flaunt private jets today. “This cup of coffee was grown by a real person, ground by a real person, and brewed by a real person, not optimized by AI for the best taste.”
3. The ultimate political question. When productivity no longer requires mass participation, what legitimacy does democracy have? “One person, one vote” was meaningful in the industrial era because everyone was a potential laborer; but when 99% of people contribute nothing to productivity, should voting rights still be equal? This may be the most dangerous political philosophy question of the second half of the 21st century.
Conclusion
We always fantasize that an AI society will have countless robots working for us, but the more realistic picture is that there will be fewer and fewer robots because they are fundamentally redundant. Instead, there will be a global, invisible, yet omnipresent web of automated processes, and most of us will find ourselves living as nodes on this web, becoming the fuel needed for its self-optimization. At that time, the real scarcity will not be labor but the willingness and ability of someone to answer the simplest yet most critical question: “What do we want this machine to optimize for us next?”