AI Is Reshaping Human Value: We Are Both Contributors to Civilization and Experiencers of Life

Shou-De Lin, Professor, Department of Computer Science and Information Engineering, National Taiwan University

In recent years, large language models and generative AI have advanced faster than many people expected. Working at the intersection of research, education, and entrepreneurship in information technology, my most immediate impression is that many cognitive tasks we once took for granted as requiring human effort are gradually being broken apart and redistributed. Some are already being handed over to AI.

“Will AI take our jobs?” is, of course, a question that concerns almost everyone. But in recent years, I have increasingly felt that if AI continues to develop along its current trajectory, we will eventually confront a more fundamental question:

If one day the amount of human labor required to sustain social production and the functioning of civilization falls dramatically, should we still treat “work” as the central mechanism for distributing resources, determining social status, and defining the value of a human life?

Broadly speaking, the technology world offers two different visions of this future.

One is relatively optimistic. Although Jensen Huang, Yann LeCun, and Andrew Ng do not make exactly the same arguments, all have emphasized, in different ways, the complementarity between AI and humans. Put simply, humans and AI can work together to make the pie bigger. AI changes the task structure of jobs and raises human productivity; lower costs can then stimulate new demand, ultimately creating products, services, and even jobs that did not previously exist.

The other vision is closer to the concerns or expectations expressed by Elon Musk and Geoffrey Hinton. If future AI is no longer merely a tool for improving efficiency, but reaches or exceeds human performance across an increasingly broad range of cognitive tasks—and is then combined with highly autonomous robots—the experience of automation over the past two centuries may no longer apply directly to the next stage. Musk has even made the bold prediction that in a society where AI and robotics become sufficiently advanced, work may one day cease to be a necessity for survival and instead become a personal choice, more like sports, entertainment, or games.

These two views may appear contradictory. I suspect that both could turn out to be true, just at different times and on different scales.

At least in the short term and within particular domains, I largely agree with Huang’s emphasis on demand expansion and human-AI collaboration. But if we extend the horizon to ten or twenty years—or even longer—I lean toward another possibility: not that “jobs suddenly disappear,” but that the amount of human time required to sustain the same level of economic activity and civilizational output gradually declines.

Yet “less human time is required” does not mean that “everyone will naturally work fewer hours.” The former is a technological and economic possibility; the latter depends on how wealth is distributed, how working hours are organized, and how society is designed. AI can create the conditions for reducing necessary labor, but it will not automatically create a society in which everyone shares the productivity dividend.

Before discussing that long-term direction, however, we need to clarify one thing: different kinds of work may be automated at very different speeds.

The strongest evidence for technological optimism is history itself. The major technological revolutions of the past two centuries did replace enormous amounts of human labor, yet they did not produce permanent mass unemployment.

Steam engines replaced human and animal power. Computers replaced manual calculation. Software such as Excel dramatically reduced the labor required for calculation and tabulation. Many tasks and occupations disappeared as a result, but humans were not pushed out of economic activity altogether.

One important reason is that what we call a “job” is actually a bundle of many different tasks.

Andrew Ng once summarized this idea in a simple phrase: AI automates tasks rather than jobs.

A physician does more than interpret medical images. An engineer does more than write code. A teacher does more than transmit knowledge. When some tasks are handed over to AI, the remaining work can be recombined, while humans gradually shift toward judgment, communication, design, integration, and responsibility.

Excel is a good example. It automated enormous amounts of manual calculation and spreadsheet work, but companies did not stop needing financial analysis. Instead, the center of gravity shifted from “calculating the numbers” to “building models, interpreting results, and using information to make decisions.”

Another frequently discussed concept is the Jevons Paradox: when technology makes a service much cheaper, demand for that service may increase.

Suppose the cost of developing customized software falls by 90 percent. Products that were once affordable only to large corporations might become accessible to small businesses or even individuals, causing the overall software market to expand dramatically. Drug design, animation, personalized education, and legal analysis could experience similar effects.

And what expands may not be quantity alone. As production costs fall, we may begin to demand levels of quality, complexity, and personalization that were previously unaffordable. A company that once built a single software system might someday create hundreds of highly customized versions for different customers.

So after AI raises productivity, humanity may not simply maintain the same level of output. We may use the newly available capacity to do things that were previously too expensive or difficult to attempt.

The real question, then, is this:

Can the expansion of demand, quality, and complexity keep pace with the productivity gains created by AI?

Suppose AI makes one software engineer twenty times more productive, while total demand for software increases only fivefold. Even though the market expands dramatically, the total number of human work hours needed to produce that software could still fall. Conversely, if demand increases one hundredfold while productivity rises only twentyfold, the amount of human labor required could actually increase.

I believe this race is the key divide between the two competing visions of the future.

Historical analogies are useful, but there is something about the current wave of AI that deserves particular caution.

Past automation technologies often targeted relatively well-defined abilities or parts of a workflow. Cars move faster than humans. Calculators calculate faster than humans. Industrial robots weld with greater precision than humans. Their impact has been enormous, but the capabilities themselves generally had relatively clear boundaries.

What is especially noteworthy about the current wave of AI is not that it already possesses complete, human-like “general problem-solving ability”—it is still too early to make that claim—but that the same underlying AI capabilities are beginning to cross boundaries between occupations and automate large numbers of cognitive subtasks shared across many professions.

Today, AI can already write code, analyze data, read documents, generate designs, translate languages, propose hypotheses, plan workflows, and call other tools to carry out multi-step tasks. These activities were once distributed among engineers, analysts, designers, translators, researchers, and administrative staff. Increasingly, the same class of models can perform all of them.

In other words, AI is not merely automating one occupational skill. It is automating a set of cognitive capabilities that can transfer across jobs.

More importantly, digital intelligence can be replicated rapidly.

When one person learns a skill, everyone else does not automatically acquire it; every human brain must spend time learning for itself. But once a model acquires a new capability, an improved version of that model can be rapidly deployed to vast numbers of users and machines.

An excellent engineer cannot instantly copy twenty years of experience into ten thousand junior engineers. AI may eventually approximate something surprisingly close to that effect. This is a fundamental difference between digital and biological intelligence.

Therefore, when most of a job can be represented as information input, information processing, and information output, the long-term possibility of human labor substitution deserves serious attention.

This does not mean that programmers, lawyers, accountants, or analysts will suddenly disappear one day. A more plausible path is that work once requiring ten people can later be done by five, and eventually by three—or that the same ten people remain, but accomplish what once required fifty.

Which outcome ultimately occurs will still depend on the race between demand and productivity. For that reason, I am reluctant to assume that because every previous technological revolution eventually created new jobs, AI must necessarily follow exactly the same path.

If the story ended here, it would be easy to conclude that AI will rapidly replace every kind of work, because AI is evolving faster than humans are.

But the world we live in is not made of information alone.

Robotics and Physical AI have long confronted the enormous difference between bits and atoms. Information can be copied and transmitted rapidly. The physical world, by contrast, is constrained by materials, energy, space, time, friction, wear, and safety.

The marginal cost of copying a piece of software one million times can be extremely low. Manufacturing one million robots is entirely different. Robots require motors, sensors, gearboxes, batteries, and mechanical structures. Behind them are minerals, factories, logistics, electricity, maintenance, and safety certification.

Even more importantly, the real world is not a simulation.

When software fails, we can rerun it. When a robotic arm hits a person, we cannot simply press reset and pretend nothing happened. An AI system can experience millions of rounds of trial and error in simulation. An aircraft, a nuclear power plant, or a medical robot obviously cannot learn through the same kind of experimentation in the real world.

For this reason, I expect physical labor to be replaced much more slowly than purely digital work. Manufacturing, energy, healthcare, transportation, construction, infrastructure, experimental science, and other domains deeply entangled with the physical world may retain human-AI collaboration for a long time.

AI can advance rapidly, but turning intelligence into safe, reliable, large-scale physical action ultimately requires dealing with the physical world.

Beyond the physical world lies another constraint that the technology sector often underestimates: institutions.

The fact that AI is technically capable of doing something does not mean society will allow it to do so autonomously.

Healthcare, aviation, energy, finance, public governance, and high-risk industries all involve safety certification, legal liability, insurance, ethics, and social trust. Even if AI someday becomes more accurate than human physicians on average, that does not mean we will allow AI to make all treatment decisions autonomously the next morning.

Putting these factors together, I would divide the pace of automation into three layers:

• Bits: cognitive capability — Can AI actually do the task?
• Atoms: physical deployment — Even if AI knows what to do, can machines perform it safely and reliably in the real world?
• Institutions: institutional acceptance — Once the technology and machines are ready, are the law and society willing to let them operate autonomously?

These three clocks may move at completely different speeds. AI's cognitive capabilities can change noticeably within months or a year or two. Large-scale physical infrastructure often takes many years to deploy. Laws, ethics, and social institutions may move even more slowly.

I therefore expect the world of the next ten to twenty years to be highly uneven. In the digital world, the human time required to produce the same output may fall rapidly. In the physical world and in domains involving high levels of responsibility, human-AI collaboration may persist for much longer.

Only when Physical AI, robotics, and institutions gradually cross these thresholds might the “necessary human labor time” required to sustain civilization begin to decline across the board.

At this point, a natural question arises.

If sustaining society in the future requires only 20 percent of today's human work hours, does that mean only 20 percent of people will need to work while the remaining 80 percent lose their jobs? Not necessarily.

One possibility is that 20 percent of people continue working full-time—and perhaps earn even more because they control AI and capital—while the other 80 percent lose both employment and income.

Another possibility is to redistribute necessary labor so that most people continue to participate in civilization, but each person needs to devote only 20 percent as much time to it.

These two forms of distribution could produce completely different societies.

Whether AI can reduce necessary human work hours is a technological and economic question. Who ultimately receives the time that AI saves is an institutional question.

If we simply leave the outcome to unconstrained market competition, returns to capital and technology may become highly concentrated. The result may not be “everyone works less.” It could instead be that a small number of people command extraordinary productive capacity while the market value of everyone else's labor declines sharply.

That is why I increasingly prefer to think of 20/80 in terms of the distribution of time—not as a prediction, but as a call for an institutional direction worth pursuing.

If society can use tax policy, working-time reform, and redesigned mechanisms of resource distribution to return the enormous productive capacity released by AI to a broader population, perhaps the future will not be one in which “20 percent work and 80 percent are unemployed.” Instead, everyone might devote only a portion of life to necessary labor.

A person might spend part of the week contributing to system maintenance, public governance, education and care, or local services, while using the rest of the time to be with family, create, exercise, study history, participate in the community, or explore something they are genuinely fascinated by.

Thinking about the future this way changes the question that interests me.

Instead of repeatedly asking:

“What jobs will remain that only humans can do?”

I would rather ask:

“As less and less human labor is required to sustain civilization, what roles do we still want humans to play within it?”

In the examples that follow, I am not claiming that “AI will never be able to do these things.” If AI continues to improve, it may eventually perform many of them faster and more accurately than humans.

The real question is:

Even if AI becomes capable of doing them, which activities do we still want humans to participate in? Which decisions are we unwilling to hand over completely?

I can think of at least four roles.

In the future, most routine failures and standardized procedures may well be diagnosed by AI and even repaired by robots. The harder cases are “open-world exceptions”—long-tail situations that have never been fully defined in advance.

These do not occur only in high-tech facilities. They appear constantly in ordinary life: allocating supplies when a community loses water and electricity during extreme weather; responding to a sudden emotional breakdown or interpersonal conflict involving a resident with dementia in a care facility; resolving family privacy disputes that emerge while digitizing local historical archives.

Today, human advantages in such situations often come from understanding the immediate context, making judgments across domains, and accepting the consequences of decisions when no complete set of rules exists.

AI may become increasingly capable of handling such problems. But even if it can propose better solutions, society may still want humans involved in situations with no standard answer—to exercise judgment, understand the circumstances of those affected, and bear ultimate responsibility.

An engineer diagnosing an unprecedented failure is contributing to civilization. So is a frontline caregiver calming an elderly resident, or a community volunteer resolving a neighborhood dispute.

People sometimes joke, “AI can't go to jail.” The line is an oversimplification, but it points to an important issue: as AI becomes increasingly capable of making decisions, how much decision-making authority are humans willing to hand over?

Which tasks can AI perform autonomously? How much risk are we willing to accept? When something goes wrong, who bears the cost? Which values should not be sacrificed simply for efficiency?

There is rarely a universally correct mathematical optimum between efficiency and fairness, convenience and privacy, or innovation and safety. Human participation matters not necessarily because AI “cannot calculate the answer,” but because these questions do not have a single objectively correct answer in the first place.

The boundaries of medical AI require participation from physicians, patients, and society. The direction of AI in education requires discussion among teachers, students, and parents. AI in public governance involves every citizen's choices about fairness, freedom, and risk.

Choosing values is itself a form of participation in civilization.

We used to say, “AI is good at interpolation; humans are good at creativity.” Looking back, I think we said that too soon.

AI can already propose hypotheses, design molecules, search new algorithmic spaces, and generate solutions that humans did not explicitly specify. I am no longer comfortable treating “creativity” as a fortress that AI will never breach.

But another question remains:

In a world without a clearly defined reward function, what problems do we actually want to solve?

Why explore Mars rather than the deep ocean? Why is one mathematical problem worth a generation of effort while another attracts little attention? Which diseases deserve more resources? Which disappearing languages, cultures, or crafts should be preserved?

Eventually, all of these questions encounter the same word: what is “worth” doing?

AI may certainly recommend research directions in the future and may even predict more accurately which ones are most likely to succeed. But “most likely to succeed” and “what kind of civilization do we want to become?” are not the same question.

So the future role of humans in science and civilization may not depend on our remaining permanently better than AI at solving problems. The question of “where do we want to go next?” may become more important instead.

And this is not just a question for scientists. Artists choose which experiences are worth expressing. Historians choose which memories are worth preserving. A curious child may ask a question that no one has seriously considered before. All are participating in the same process:

Deciding where civilization should look next.

If finance, energy, logistics, transportation, and information systems are someday operated largely by interconnected autonomous AI agents, efficiency may become extraordinarily high. But deep interconnection also creates new forms of fragility. A local error could propagate through an entire system faster than humans can react.

AI alignment, safety guardrails, multi-layer fail-safes, independent redundancy, human override mechanisms, and system recovery under extreme conditions will therefore remain important.

The best anomaly detection, security monitoring, and even incident response systems of the future may themselves be AI systems. The crucial human role may therefore not be personally watching every anomalous signal. Instead, humans may need to decide:

How far should we allow systems to optimize? How much systemic risk are we willing to tolerate in exchange for efficiency? Which critical systems must retain redundancy? Under what circumstances must a human retain the authority to say, “Stop”?

Civilization has always done two things at once: expanded the boundary of “what we can do,” while also deciding which things we choose not to do, even when we can.

If the thought experiment above ever comes close to reality, I believe it will bring about a change more interesting than simply “the disappearance of work.”

We may need to separate two kinds of value that have long been conflated.

The first is a person's instrumental value: What can I produce for society? What problems can I solve? How much economic output can I create?

The second is a person's intrinsic value: What does it mean for a person simply to live, feel, form relationships, explore the world, and experience life?

Industrial society and capitalism make it easy to conflate the two. Work determines income; income shapes our range of choices; occupation often influences social status. Over time, we can easily internalize an unspoken equation:

A person's value is roughly what the market is willing to pay for that person's labor or contribution.

Yet many of the most important things in life have no market price at all.

Caring for your children. Spending time with aging parents. Writing a poem no one will buy. Playing basketball without an audience. Researching the history of your hometown. Staying beside a friend through a difficult period in life. These activities may contribute almost nothing to GDP, but few of us would therefore conclude that they have no value.

If AI truly reduces the labor required to sustain material life, large amounts of “non-utilitarian time”—something historically available mainly to a minority—might, for the first time, become accessible to much of society.

Some people may use it to make art. Others may study philosophy, exercise, travel, care for family, participate in their communities, or spend ten years investigating a question with no commercial value simply because it fascinates them.

These activities may add little to GDP. But there is no reason the purpose of civilization should be identical to maximizing GDP.

Nor do we need to keep searching for “something AI will never be able to do” in order to prove that human activity has value.

Even if AI someday composes music better than I do, I can still play an instrument. Even if AI writes better than I do, I may continue writing because I enjoy thinking and writing. Cars have long been faster than humans, yet we have not stopped running marathons. Computers have long surpassed humans at chess, yet people have not stopped playing chess.

Some things are worth doing not because they prove that we are better than machines.

We simply want to experience them ourselves.

The meaning of a piece of music does not exist only in the sound waves. It also exists in the person who hears them. A sporting event is meaningful not merely because a collection of physical objects moves according to the laws of physics, but because players and spectators jointly give it meaning.

I increasingly suspect that we have placed too much emphasis on the human role as “producer.”

Humans are also participants, experiencers, interpreters, and creators of value. If no one in a civilization experiences wonder, sadness, love, achievement, or beauty, then no matter how extraordinary its productive efficiency becomes, I am not sure what that efficiency is ultimately for.

This may be one of the most interesting reversals brought about by AI.

As machines gradually take over more of “production,” human value does not necessarily shrink. Instead, we may rediscover dimensions of human life that were always there but were long obscured by economic production:

Through our actions, we contribute to civilization. Through our lives, we experience, interpret, and give meaning to civilization.

This is also a question I am increasingly asked in many different settings.

The most immediate anxiety among students is: “What should I learn so that AI won't replace me?”

It is a reasonable question. But if even part of the argument above turns out to be correct, perhaps two other questions are even more important:

How can I continue to make meaningful contributions to the world? And do I know how to live?

The first includes cross-disciplinary understanding, experience with physical systems, open-world problem solving, system security, AI alignment, risk judgment, and the ability to make decisions when information is incomplete or values are in conflict.

For computer science students, I would not recommend building your sense of value around “I can code better than AI” or “I can solve algorithm problems better than AI.”

Programming will certainly remain important. But its future importance may increasingly resemble that of mathematics. We do not study mathematics because society needs every person to perform every calculation by hand. We study it to understand structure, abstract problems, and verify results.

Programming may gradually move in the same direction. The importance of personally writing every line of code may decline, while understanding systems, defining problems, judging whether results are correct, and being capable of building new things become more important.

The second question is one that education has historically discussed far less:

When a person no longer needs to spend most of the time working, does that person know how to live?

Aesthetic appreciation, curiosity, sports, art, relationships, empathy, community, and finding something one genuinely wants to pursue over the long term may all become more important than we currently imagine.

Our education systems devote enormous effort to teaching children how to become good producers: how to calculate, how to program, how to solve problems, how to enter good schools, and how to obtain professional qualifications.

But if technology someday makes “necessary work” occupy far less of a human life, education may have another mission that we have long underestimated:

Teaching people how to use their freedom.

So if I were to offer students some advice today, I would put it this way:

Becoming someone whom AI finds difficult to replace is certainly valuable.

But perhaps what matters even more is becoming someone who, while living alongside AI, still knows how to contribute—and also knows how to live.

No one knows what AI's ultimate impact on employment will be. I also do not believe we currently have enough evidence to predict with confidence how many jobs will disappear ten or twenty years from now.

Technological optimists remind us that when AI lowers costs, new demand emerges, and when some tasks are automated, jobs are recombined. This has happened many times throughout history, and it will certainly continue to happen.

At the same time, AI is crossing occupational boundaries and entering an increasing number of cognitive tasks shared across professions. Digital intelligence can also be replicated rapidly and deployed at low marginal cost. For that reason, I do not think we can simply extrapolate from previous technological revolutions and assume that this one must produce the same outcome.

Nor will these changes happen everywhere at the same speed. The purely digital world may move quickly. Once AI enters the physical world, materials, energy, safety, and deployment constraints will slow it down. When it moves further into medicine, law, finance, or public governance, responsibility, trust, and institutions will slow it down again.

The future I imagine, therefore, is not one in which we suddenly jump from “everyone works” to “no one works.” It is more likely to be a long and uneven transition.

If I had to identify one long-term direction about which I feel relatively confident, it would be this:

For a given level of civilizational output, technological progress will reduce the amount of human time required to sustain that output.

But whether we will actually work less as a result is a different question.

We may use the capacity we save to create more products, higher quality, and more complex services. Or the productivity created by AI may become concentrated in the hands of a small number of people, leaving us not with a world in which everyone works less, but one in which some remain extremely busy while the market value of others declines rapidly.

Between technological progress and a world in which “humans can work less” lies an entire layer of institutional design.

I now think of the issue in three layers.

• Technological possibility: Can AI reduce the amount of human labor required to sustain civilization?
• Institutional choice: If it can, are we willing to convert that productivity dividend into more free time for more people, rather than greater wealth concentration or unemployment?
• The value of life: If one day we truly no longer need to exchange most of our lives for the resources required to survive, what will we do with that time?

AI may help us solve the first problem.

The second must be answered collectively by institutions and society.

The third, ultimately, each of us may have to answer for ourselves.

For thousands of years, humans have continually invented tools, machines, and computers to increase productivity. But for what?

If the final purpose is merely to produce more, and then use the time we save to produce still more, the process seems to have no endpoint.

I hope for another possibility.

What technological progress ultimately gives back to us may be time.

Less of our lives spent on things we have no choice but to do, and more of our lives reserved for things we would still choose to do even if no one paid us.

If AI ultimately takes us there, then what it changes will be far more than work.

It will also force us to revisit a very old question:

Where does the value of a human life come from?

My answer has at least two parts.

We are contributors to civilization, and we are experiencers of life.