The Architecture of Power

A delivery rider works the week he has a fever. He is ill; he knows he is ill, and he takes every job anyway.

Nobody ordered him to. No algorithm flagged him. He changed his own behaviour because of what the system might do, and the system never had to do anything at all.

That is the most powerful thing happening in artificial intelligence right now, and almost nobody is measuring it.

Dr Iris Efthymiou, a behavioural economist, calls it anticipatory compliance. The behaviour change that happens before the algorithm moves. The nudge nobody sent. The punishment nobody issued.

You have probably already done it.

You softened an email because a system might read it. You did not apply for the loan, the job, the flat, because you were fairly sure what the answer would be. You stayed quiet when speaking up would have cost you a rating you cannot see and cannot appeal.

Nothing happened to you. That is exactly the point.

Here is what makes this book different from everything else on the shelf. Every law we have written to control these systems waits for an event: a decision, a rejection, a wrong answer. The EU AI Act sorts systems by the risk their decisions carry. Data protection law gives you rights over automated decisions.

But there is no decision to appeal when you never applied. A system that scares away the people it would have treated worst will score better on fairness than one that treats them badly, and it will be congratulated for it.

We have built an entire machinery of oversight that cannot see the biggest harm.

What you will find inside:

• Why nudge theory stops working when the scoring never stops, and what replaces it
• The Dutch childcare benefits scandal, where a secret fraud model wrongly destroyed tens of thousands of families and brought down a government
• Australia's Robodebt, which a Royal Commission called crude, cruel and unlawful
• Why Amazon quietly killed its own AI recruiting engine
• How credit scores became permanent identity, and who that traps
• What AI is really costing in energy, water and human exhaustion, and who is paying
• A detailed agenda for governments, companies and citizens that does not depend on anyone's goodwill

The last third of this book refuses despair. Public AI infrastructure. Data commons. Taxing capital instead of labour. Algorithmic accountability with actual enforcement power. Shorter working hours. A social contract that puts rights before efficiency.

If you read one book about AI this year, make it the one that explains what it is doing to you, not just what it can do.