The shocks people gave because a man in a grey coat asked

In 1961 the psychologist Stanley Milgram began a series of studies at Yale University. Volunteers from the local area were told they were helping with research on memory and punishment. Each one played the teacher. A pleasant middle-aged man, who was secretly an actor, played the learner and sat strapped into a chair in the next room.

Every time the learner got a word pair wrong, the teacher had to give him a shock from a machine with 30 switches, running from 15 to 450 volts in steps of 15. No real shocks were given. As the voltage rose, the learner complained, protested and finally went silent. If the teacher hesitated, the experimenter in a grey lab coat used a set of polite prompts, such as asking them to please continue.

In the version Milgram reported in 1963, 26 of the 40 men (65%) went all the way to the last switch. Many were visibly upset, sweating and laughing nervously, yet they kept going. In later variations, obedience fell sharply when the experimenter gave orders by telephone, or when the teacher had to press the learner's hand onto a shock plate.

The work has serious critics. The psychologist Gina Perry studied the archive tapes and found that experimenters sometimes pushed harder than the script allowed, and that a fair number of volunteers suspected the shocks were fake. The ethics of deceiving and distressing people also helped bring in stricter rules for research. Still, a partial repeat by Jerry Burger in 2009, which stopped at 150 volts for safety, found most people willing to carry on past that point.

Milgram's own reading was that people shift responsibility upward. Once someone else seems to own the decision, the person doing the act feels like an instrument. The setting did a lot of the work: a famous university, a scientific purpose, a calm voice and small steps, each only a little worse than the last.

Software can play the role of the grey coat. A system that sounds sure of itself, comes from a trusted brand and asks for one small click at a time invites the same handing over of responsibility. When a chatbot or an automated workflow says the next step is fine, it is easy to feel that the decision belongs to it.

The decision is still yours. The tool will not answer for the email it drafted, the refund it refused or the advice it gave a customer.

The idea behind this caseLesson: Review AI like a manager →

1. 1Before you act on a tool's suggestion, say out loud or write down who is responsible for the result. Usually it is you.
2. 2Watch for small steps. If each change seems only a bit worse than the last, compare where you are now with where you started.
3. 3Set a stop rule in advance, such as never sending money, sharing personal data or deleting records on an automated suggestion alone.
4. 4Make it normal on your team to pause and question a confident instruction, whether it comes from a manager, a vendor or a machine.