Why chatbots make things up: AI was built to surprise

Why would anyone make a computer unpredictable?

Computers are good at exact answers. In the early 1950s, some engineers wanted something else.

In an interview published by MIT News on 23 October 2026, Benjamin Lindquist (an assistant professor in MIT’s History Section) explains why. He says many people then believed chance plays a part in discovery and creativity. If a machine only follows fixed steps, it can never leap to a new idea.

His example is John Pierce, an engineer at Bell Labs. Pierce thought maths could add a burst of the unexpected to art and break clichés.

What did that have to do with neural networks?

Lindquist says one way to add irregularity was through artificial neural networks. Many early ones were built by psychologists. They hoped to model the brain, mistakes included.

He says the early fans valued these systems for the opposite reason from ours. Today we want chatbots to be right. They wanted a system that sometimes gave an answer nobody expected.

Where do made-up answers come from, then?

In his new paper, Lindquist traces the origin of AI hallucinations to this early work. He says there is a link between creative answers and wrong ones.

He also says companies now “sand down” the strangeness for legal and money reasons. In his words, it is still there at the core, but largely held back.

What’s the catch?

• This is one historian’s reading, given in an interview. We read the interview, not the paper.
• Lindquist does not say every chatbot error is caused by deliberate randomness. He says the history shows a trade-off.
• His lesson: push hard against surprises and you may lose something else, such as creativity.

Try it

This step is our idea, not Lindquist’s. Ask a chatbot an open question twice. Try “Give me a name for a robot that waters plants.” Then ask a factual one twice. Try “How many days are in a leap year?” Which answers changed? Which did you want to change?

You decide what you want from the machine: a surprise or a fact.

Sources

• Q&A: How AI came to model the “messy” human mind, Adam Zewe, MIT News, 23 October 2026
• The Irrational Computer: Randomness and Unreason in the History of Artificial Intelligence, Benjamin Lindquist, Past and Present (not opened in full)