Aristotle Already Imagined the Tool That Works by Itself

From Aristotle's autonomous tool to the water mill: a reflection on the tasks that change and the people who perform them.

Aristotle already imagined the tool that works by itself

Series: What history already knew about talent · 5/6 · Preparation for AI

In the 4th century BC, in the first book of the Politics, Aristotle casually presents a thought experiment that today sounds like a newspaper headline.

He imagines, he writes, that each instrument could perform its function by itself, obeying an order or anticipating it. Like the figures of Daedalus, which according to legend moved on their own, or the tripods of Hephaestus, which in Homer enter on their own into the assembly of the gods. If the shuttle wove by itself and the plectrum played the lyre without a hand to guide it, he concludes, the masters of work would not need workers nor would the lords need slaves.

It is important to place the passage where it is. Aristotle is not proposing a technological liberation of labor; he is in the midst of a discussion about property, service, and slavery, and the autonomous tool serves to explain what an instrument is. He leaves it there, as a fantasy comparable to walking statues.

About three centuries later, someone describes it functioning.

The wheel and the millstones

Around 25 BC, Vitruvius dedicates a passage in the tenth book of his treatise on architecture to the water mill: a wheel moved by the current, a gear that transmits the rotation to a vertical axis, and a millstone that turns over another while the grain falls from a hopper. It is a technical description, with no comment on who milled before or what happened afterward.

The comment is made by a poet. Antipater of Thessalonica, an author from the time of Augustus whose work does not allow for securely dating each poem on one side or the other of the turn of the era, dedicates an epigram to the women who ground grain, a domestic task, feminine or enslaved, that occupied hours each day from before dawn. The poem tells them, more or less, to take their hands off the mill and keep sleeping even though the rooster announces the day, because Demeter has entrusted the work that their arms used to do to the nymphs: the nymphs leap over the wheel, make the axle turn, and the axle moves the millstones.

It is a celebration. Someone looks at a machine and sees, first and foremost, a task that is no longer done by hand.

What the sources do not tell

Here is where I have to be careful, because it is easy to follow the story with what one would like to have happened.

None of these three texts says what happened to the women who ground grain. Aristotle poses a philosophical hypothesis. Vitruvius describes a mechanism. Antipater celebrates a saved effort and does not look back at the millers after sending them to sleep. The spread of hydraulic energy in the Roman world was, moreover, uneven across regions, periods, and uses, as recent research has shown, and does not admit a single sequence of 'the machine arrived and this happened.'

So I am not going to tell what happened to the millers. I will point out what the poem leaves open, because that is exactly where our question begins.

Antipater sees the task. He does not see the person the next day. When a technology takes over part of the work, the task is what is visible: what is no longer done by hand, what now takes seconds. What is not visible, and what no one writes poems about, is what changes for the person who did that task among the other things they did, and what they need for what comes next.

The same question, twenty centuries later

For three years we have been hearing Aristotle's question about artificial intelligence: what jobs will it eliminate. And in most organizations, the observable answer, for now, is 'few,' which leads many people to conclude that the impact is less than what was said.

It is the conclusion of someone looking at how many millers have disappeared. And it seems to me that they are looking in the wrong place.

What we see in the companies we work with is that, within many positions, the distribution of tasks is changing. The analyst remains an analyst, but the part of their week dedicated to preparing data is decreasing while the part dedicated to deciding what to ask and explaining the answer is growing. The salesperson remains a salesperson, but they prospect differently and prepare meetings differently. Administrative managers, support technicians, writers, programmers: the positions remain in the organizational chart, and the content of the day is shifting beneath.

We have our own data that points in that direction, with the caution it deserves. In the analysis we conducted this quarter on about five hundred thousand job offers, the two healthcare profiles with the highest growth in ninety days were not clinical: they were the medical writer and the medical transcriptionist, with growths of over one thousand percent and nearly six hundred percent on small bases. An increase in job offers does not alone demonstrate its cause, and I will not assert more than the data allows. But it is consistent with one reading: that the documentary work that clinicians did between patients is being partially displaced to profiles that take it on and to tools that assist it. The position of doctor is still there. The content of their day seems to be what is changing.

That is the watermill seen from the side that Antipater did not look at.

The question that needs to be asked

If what changes first are the tasks, the useful question is not 'what positions will AI eliminate,' but 'what part of the work of each position in my organization is changing, and what does the person occupying it need to remain valuable in the part that remains.'

That question can be worked on. Each profile is broken down into the specific tasks of their day-to-day. For each task, it is estimated how much of it technology can take on today and how much still requires human judgment. It is aggregated by profile and the entire organization is examined. And three groups usually appear.

Profiles where most tasks still require judgment, context, and relationship: technology affects them little in the short term, and the priority is to continue developing them.

Profiles where a significant part of the tasks can be assisted: here the tool multiplies those who learn to use it and leaves behind those who do not, and this is where training has the most immediate return.

And profiles where a good part of the tasks is automatable. Here the word is not 'cut,' it is 'reconvert in time,' because the person occupying that position knows things about the business, the clients, and how things really work that are worth much more in another role than lost. I will not promise that automation will never eliminate positions, because that is not true. What I do assert is that most organizations have room to anticipate, and that almost none are using it.

After sleeping

Antipater sent the millers to sleep and did not ask what they would do upon waking. That is the part that falls to us, and the difference from his time is that now we can see the map beforehand: which roles change the most, what new capabilities each one requires, and who is closer or further from having them. With that, we decide whom to reinforce first, in what way, and with what urgency. Without it, decisions are made when it is already too late.

At The Wise Skill, we measure that: how each role changes with artificial intelligence, task by task and with the same formula for comparable profiles, and what capabilities each person has today in light of that change. With a principle that seems important to me: the map exists to reconvert in time, not to decide whom to replace. No reading translates into an automatic decision about a person.

But you can ask the question tomorrow in the committee when someone brings up the topic of AI.

Not how many millers are going to disappear. What are they going to do when they wake up.

If you are adopting AI in processes and someone in the committee asks to know its real impact on roles and on people, message me directly and I will show you how to read the map.

Final installment: Themistocles did not just buy ships.

Sources. Aristotle, Politics, I, 4 (1253b-1254a). Vitruvius, On Architecture, X, 5, 1-2, on the water mill. Antipater of Thessalonica, Palatine Anthology, IX, 418; author from the time of Augustus, the epigram cannot be dated with certainty before or after the turn of the era. On the uneven spread of hydraulic mechanization: Andrew Wilson, "Machines, Power and the Ancient Economy", Journal of Roman Studies, 92 (2002), pp. 1-32. The data on health profiles comes from The Wise Skill, Research Report Q3 2026: sample of ~500,000 active job postings, trends over 90 days with a minimum threshold of 20 previous jobs. The reading on task change is from the author, not from ancient sources.