Experiment 08 · Famous experiments

The lines everyone could see, until the group said otherwise

Students picked an obviously wrong answer to a simple eyesight test about a third of the time when everyone else in the room had picked it first.

  • Solomon Asch
  • Swarthmore College, Pennsylvania, United States
  • 1951
  • 3 min read
On this page
Before you readWhat would you do?

You ask a chatbot whether your business plan is solid. It praises it. So do two other AI tools you try. A friend who runs a shop says your rent estimate looks far too low. What should you take from this?

Pick an answer now. The file tells you whether you were right.

01The file in one minute

  1. 1An easy test made hard

    Alone, people matched the lines almost perfectly. With a group giving the wrong answer, they went along about a third of the time.

  2. 2Most bent at least once

    About three in four volunteers conformed at least once. About one in four never did.

  3. 3One ally changes it

    A single person giving the right answer made conformity drop sharply.

  4. 4What it means for AI

    Chatbots often agree with you, and several tools can echo one source. Ask for disagreement and look for independent checks.

02Timeline

  1. 1951

    Asch publishes his first results on group pressure and judgment of lines.

  2. November 1955

    His article Opinions and Social Pressure appears in Scientific American.

  3. 1956

    A long monograph reports the full series, including the ally and group-size variations.

03The experiment

In the early 1950s the psychologist Solomon Asch invited male college students to what they were told was a test of vision. Each sat at a table with about seven others. They were shown a card with one line, then a card with three lines, and had to say out loud which of the three matched the first one in length. The answer was easy to see.

Only one person at the table was a real volunteer. The others were working with Asch, and on 12 of the 18 rounds they all gave the same wrong answer before the volunteer's turn. People tested alone made mistakes less than 1% of the time.

Under group pressure, volunteers went along with the wrong answer on about a third of those rounds. About three in four went along at least once, and about one in four never did. When Asch planted a single ally who gave the right answer, conformity dropped sharply. Adding more people to the wrong side made little difference beyond three or four.

04What it showed

Asch found two kinds of giving in. Some people doubted their own eyes and decided the group must know better. Others knew the group was wrong and went along to avoid standing out. Later studies in other countries and decades found less conformity than Asch did, so the exact rate depends on the time and place. The pull toward the group is a steady finding.

A chatbot can become the voice at the table. Many models tend to agree with the user, because they were tuned on people's ratings and people often rate agreement highly. Ask whether your plan is good and you may hear yes. Several AI tools repeating the same claim can also feel like a crowd, even when they share the same source.

Asch's ally is the useful part. One independent voice that says what it sees makes it far easier for everyone else to do the same.

The idea behind this caseLesson: What a chatbot really does →

Kit the fox, ManuAI's guide, thinking it over

05What it means when you use AI

  1. Ask an AI tool to argue against your idea before you ask it whether the idea is good.
  2. When several sources agree, check whether they are really independent or all repeat the same original.
  3. In meetings, have people write their view down before the discussion starts, so the first speaker does not set the answer.
  4. Be the ally. If you see something different from the group, say it plainly. It frees others to speak.
You ask a chatbot whether your business plan is solid. It praises it. So do two other AI tools you try. A friend who runs a shop says your rent estimate looks far too low. What should you take from this?

06 — Would you have caught it?

You ask a chatbot whether your business plan is solid. It praises it. So do two other AI tools you try. A friend who runs a shop says your rent estimate looks far too low. What should you take from this?

07 — Sources

  1. Opinions and social pressure · Scientific American · 1955-11
  2. Asch conformity experiments · Wikipedia · 2026