When the Machine Can Do the Work – Why Should Humans Still Do It?

In a recent contribution to Nature, Daron Acemoglu raises a question that easily disappears amid both the excitement and the anxiety surrounding artificial intelligence. He is not primarily asking how intelligent AI might become. He is asking what direction we want its development to take.

The title is programmatic: Don’t let AI tools replace humans. We should not develop artificial intelligence primarily to replace people, he argues, but to strengthen human expertise and expand human opportunities.

It is an important distinction. But perhaps the question goes even deeper.

What is at stake is not only the human being as a worker.

It is the human being as a participant.

As more and more tasks can be performed by machines, we need to ask what work actually means to us. Is work primarily the production of a result? Or is work also a way of entering the world – through skill, responsibility, cooperation, recognition and participation?

That question is much older than artificial intelligence.

When Efficiency Becomes the Measure

If work is understood primarily as production, the technological logic is simple.

If a machine can perform the work faster, more cheaply and better than a human being, then the machine should take over.

We have followed this logic for a long time. Machines have taken over physically demanding labour, industrial robots have assumed repetitive tasks, and computers have automated large parts of calculation and administration.

Acemoglu points to research suggesting that automation since 1980 has contributed to growing inequality in the United States by reducing demand for workers who perform routine tasks. He also refers to the effects of industrial robots: productivity increased, while some communities that were particularly exposed to automation experienced declines in employment and wages.

Artificial intelligence extends this logic into areas we have long regarded as forms of human knowledge work. It can write, translate, analyse, program and summarize large amounts of information. It can assist in diagnosis, teaching and research.

The question, then, is no longer simply how much work can be automated.

We must ask what happens to human practice when it is automated.

The Teacher Knows More Than She Can Say

Acemoglu uses the teacher as an example. Artificial intelligence can be developed to replace teachers, but it can also help them identify where students are struggling and provide more individually tailored support.

The difference is fundamental.

In the first case, we ask how we can manage without the teacher.

In the second, we ask how the teacher can become better able to be a teacher.

But why should there be a difference if AI can in fact transmit information just as well as, or better than, the teacher?

Here Heidegger can help us.

Hubert Dreyfus drew on Heidegger’s philosophy to illuminate the difference between knowing that and knowing how. Human competence does not consist only of explicit rules or information that can be formulated. It is embodied, situated and embedded in a world of meanings.

Heidegger’s concept of Zuhandenheit points precisely towards this form of practical involvement. When we master a tool or a practice, we do not normally stand outside the situation and analyse it. We are already involved in it.

The experienced teacher does not necessarily apply a rule when she senses that a student has not understood. She may hear it in the way a question is asked, see it in a hesitation, or notice that the atmosphere in the room has changed. Some of this can be described afterwards, but the skill comes before the description.

That does not mean that an AI system could never detect the same signs.

The point is different.

The teacher’s knowledge is not merely information about the teaching situation. It is a form of know-how within the situation.

She is herself part of what is happening.

Work as Participation

This also changes the question of work.

Work is, of course, about income and production. But through work we also learn how to do something, enter into relationships, and become participants in institutions and practices that existed before us and will continue after us.

We are given tasks that others expect us to carry out. Someone may become dependent on what we do. We gain experience, responsibility and perhaps a sense of mastery.

Work is not automatically meaningful. It can be monotonous, degrading and unnecessary. Much work should indeed be automated.

But if the value of work is measured only in terms of productivity, something disappears from the calculation.

A society can produce more while at the same time giving fewer people the opportunity to experience that they contribute to something on which others also depend.

This is why Acemoglu’s distinction between replacing people and strengthening them matters. Technology does not have one predetermined social purpose, he argues. AI can be developed for substitution, but it can also be designed to complement people and enhance their capabilities and productivity.

But technology does not choose between these possibilities.

We do.

Technology Has No Destiny

We often speak about artificial intelligence as though it were the weather.

AI is going to take over.

AI will make us redundant.

AI will transform working life.

But AI does none of this by itself.

Someone develops the systems. Someone finances them. Someone decides what problems they are supposed to solve and what forms of efficiency are to be rewarded.

This is why Acemoglu emphasizes politics. Among other things, he proposes changing economic incentives that favour automation, strengthening public expertise on artificial intelligence, and developing a better legal framework for data.

The details can be debated.

The underlying point is harder to dismiss:

The direction of technology is not destiny.

It is shaped by economic priorities, political decisions and ideas about what counts as progress.

And that means the question ultimately becomes philosophical.

What do we want technology to do to the way we live together?

The Machine Should Be Allowed to Take Over

This does not mean that human work should be protected from machines.

Quite the opposite.

If a machine can perform dangerous work, it should often do so. If a computer can complete in seconds a calculation that once took several days, there is little reason to preserve the slower calculation in the name of human dignity.

And if artificial intelligence can free doctors, teachers, researchers or social workers from hours of routine administration, the result may be more human work, not less.

The decisive distinction therefore does not lie between the human being and the machine.

It lies between different forms of work.

What can profitably be left to the machine?

And what should human beings continue to do because the task involves presence, judgement, participation or responsibility?

Here we come close to Hannah Arendt.

Labour, Work and Action

In The Human Condition, Arendt distinguishes between labour, work and action.

Labour is tied to the necessities of life and to the repetitive processes that sustain it. Work creates the more durable human world – things, institutions and works. But action is of a different kind.

Through action and speech, a human being reveals not only what they are, but who they are.

Action takes place between people. It cannot be fully controlled in advance. When we act, we set something in motion whose consequences we cannot entirely foresee. This is why Arendt connects action with natality – the human capacity to begin something new.

This gives the question of AI a different direction.

A machine can produce a result.

It may even produce a result better than mine.

But action in Arendt’s sense is not only about the result.

It is about someone appearing among others and setting something in motion.

Who Stands Behind This?

Here, I think, we encounter an important boundary.

An AI can write a text.

But who stands behind the text?

It can suggest a medical assessment.

But who meets the patient?

It can analyse research data.

But who says: I can defend this finding?

It can recommend a particular course of action to a teacher.

But who stands before the child and must live with the consequences?

“I did this. I stand by it.”

That is something different from being able to identify the cause of an outcome.

To be responsible means that the action can be attributed to a particular person who exists in a world together with other human beings, and who can be asked why she acted as she did.

For Arendt, this is connected to a who.

It is this “who” that appears through action.

Within such an understanding, the problem with AI is not primarily that the machine lacks intelligence. It may become capable of imitating ever more forms of intelligent behaviour.

But it has no natality in Arendt’s sense. It is not born into a shared world in which, through a lived life, it becomes a particular who among others.

We should therefore be careful not to confuse the result of an action with the action itself.

When the Tool Shapes the User

At the same time, artificial intelligence is more than a passive tool.

All tools also shape the people who use them.

The hammer makes certain actions possible. The car changes our experience of distance. The smartphone has altered our relationship to availability, memory and attention.

In the same way, artificial intelligence will not merely help us with tasks that already exist.

It will change the tasks themselves.

The writer who uses AI writes differently.

The researcher who uses AI conducts research differently.

The teacher who uses AI teaches differently.

We should therefore not ask only what artificial intelligence does for us.

We must also ask what working with artificial intelligence does to us.

Does the tool make us more skilful?

Does it sharpen our attention?

Does it expand our judgement?

Or do we gradually become less willing to undertake the slow work that experience, knowledge and understanding sometimes require?

No technology can answer that question on our behalf.

What Kind of Society Do We Want?

Towards the end of his essay, Acemoglu writes that the future of AI should not be determined by the dreams of a small group of technologists. It should instead be shaped through democratic choices about what kind of society we wish to build.

Perhaps this is the most important thought in the entire article.

Because the question of artificial intelligence is ultimately a question about our view of the human being.

Is the good society the society that produces as much as possible with as few people as possible?

Or is productivity only one good among several?

We also need participation, belonging, learning and the opportunity to contribute. And we need places where people can still appear before one another as responsible persons.

So I return to the teacher.

Perhaps a future AI will know more than she does. Perhaps it will identify the student’s academic difficulties more quickly, find the best explanation and suggest the most effective teaching method.

The teacher should be allowed to benefit from all of this.

But the next morning, it is still she who opens the classroom door.

She meets the students.

She must decide what to do with what she knows.

And when a child later asks why she acted as she did, she cannot answer:

The algorithm decided.

She must be able to say:

I assessed the situation.

I chose.

I stand by it.

Perhaps this is precisely what we should preserve as machines become capable of doing more and more of the work.

Not the human being as the most efficient producer.