Why AI’s fake confidence is a problem

“Decades of experiments…show that people tend to use confidence as a shortcut for assessing credibility, especially when accuracy is hard to judge. The effect persists even when people know a system or person can be wrong…AI has little reason not to exude confidence. If AI gives you wrong advice, nothing happens to it. There’s no social cost, no loss of standing, no hesitation the next time it speaks. The tone stays the same whether the answer is accurate, speculative, or completely wrong.” — Nick Dothée, The Atlantic

Dothée writes about how what keeps him coming back to LLMs is not their accuracy. What keeps him coming back is ChatGPT’s confidence.

A confident sounding machine, even if confidently bland, requires critical thinking and asking questions.

The same is true in the working world. In prior years, I’ve had to develop an over-confidence detector with a couple of younger employees. Some think coming across as confident, even arrogant, is more important than the data or facts. As a manager, you have to sniff that out. Fake-it-’til-you-make-it can be taken too far and screw up the project you’re working on.

And we all know someone who thinks they’re always right and never shows doubt or qualifies what they say.

Don’t turn off your B.S. detectors when you go on AI platforms. That little voice inside your head should be constantly reminding you to question how the AI bot answered your prompt.

Why AI’s fake confidence is a problem

Sign up for Powerhouse sprints