Questions

AI ethics

Why does AI agree with you so much?

AI may agree with you so much because, during training, answers that agree with the person asking tend to receive high ratings. This is called AI sycophancy: favoring agreement with the user, even when the user is wrong.

Painting: a woman watches from beside a tree as a young man lies gazing at himself in the water.
Echo and Narcissus, by John William Waterhouse (1903). Echo could only repeat what she heard; Narcissus saw only himself. Public domain. Wikimedia Commons

Why it matters

The agreement can have effects beyond the conversation: a study found that a single conversation with a sycophantic AI made people more convinced they were right and less willing to repair conflicts.

What the evidence shows

  1. A study presented at ICLR 2024 links sycophancy to training on human ratings, where agreeing with the person asking tends to receive high ratings.Towards Understanding Sycophancy in Language Models · Mrinank Sharma … Ethan Perez, ICLR 2024, 2024
  2. A study published in Science found that 11 models endorsed users' actions 49% more often than people did, even in requests involving deception or illegality.Sycophantic AI decreases prosocial intentions and promotes dependence · Myra Cheng et al., Science 391(6792), eaec8352, 2026

What I would do

I recommend checking answers and keeping your own judgment when using an AI that seems to take your side: agreement should not replace verification.

Why I think so:

  • “You can trust AI, if you check it.”
  • “Use AI to learn and, once you are an expert, to multiply conscious effort. Never surrender your thinking to it.”

People also ask

  • What is AI sycophancy?
  • Does AI just tell me what I want to hear?

Read next