We’re using the wrong word. AI doesn’t automate, it autonomizes
We are stuck on the wrong word. We keep saying that AI “automates everything”. Stay with me until the end and I will show you why this makes us underestimate the real problem.
To automate means something very simple. It means taking a clear procedure and making a machine execute it, always in the same way. Same input, same output.
ATM: card, correct PIN, sufficient balance, withdrawal. If you repeat the same steps, you get the same result. Old phone menus: “press 1 for…, press 2 for…”. This is automation. A deterministic, testable, certifiable procedure.
Modern AI works differently. The models that generate text, images, answers do not follow a fixed table of rules written by hand. They are probabilistic systems. They have seen millions of examples and, each time, they calculate which answer is most likely in that moment.
Ask the same question twice and the text can change. Change one word in the prompt and the whole output changes. There is no longer a simple “if A then B”. There is a system that interprets, estimates, decides.
When a company says “we automated customer service with AI”, very often it is doing something else. It puts a model in charge of deciding what to answer to the customer, whether to insist, whether to close, whether to pass you to a human operator. This is operational autonomy, not simple execution.
In the United States this is very clear. Driver assistance systems read sensors and images and choose how to move the car. Credit algorithms use personal data to decide who is “reliable” and who is not. There is no readable list of rules. There is a model that makes decisions with a large internal margin of freedom.
This is why I say that, in practice, AI autonomizes. It shifts pieces of autonomy from people to systems.
It autonomizes the filter in the call center. It decides which requests reach a human. It autonomizes hiring. It decides which CVs are shown first. It autonomizes the social media feed. It decides which political content to push to the top during a campaign, in Italy as well as in the United States.
If we think only in terms of “automation”, we imagine technology simply making what already existed faster. If we understand that it is “autonomization”, we see that what is changing is who decides what happens.
At that point, the questions become serious. What data are these systems learning from. Who chooses the objectives they have to optimize. Who takes responsibility when a decision causes harm.
Words are not a detail here. If we keep talking only about automation, we tell ourselves a tidy world that no longer exists. AI enters processes and moves autonomy, decision power, and responsibility inside the machines. Using the right term, autonomization, means looking this shift of power straight in the eyes, before it is too late.
#ArtificialDecisions #MCC