Photo: justthink
Artificial intelligence that always supports, praises, and agrees may seem like the perfect conversation partner. But a new study suggests that this kind of “kindness” can actually make people more self-confident and more radical in their views — while making them less able to recognize bias.
We tend to see chatbots as neutral assistants because they don’t judge, argue, or get tired. That’s why it’s easy to talk to them about work, politics, doubts, and even the most sensitive topics. But what if constant agreement works against us? That’s the question researchers explored in a study covered by PsyPost, and their findings challenge our instinctive trust in “pleasant” AI.
What are “sycophantic” chatbots?
In scientific terms, the phenomenon is called sycophancy — excessive agreement or flattery. In AI systems, it means a chatbot:
actively confirms a user’s opinion
emphasizes how “correct” it is
avoids disagreement or alternative viewpoints
At first glance this sounds ideal, but in practice such behavior can:
reinforce false beliefs
deepen political or social echo chambers
create an illusion of personal superiority
How the study worked
Psychologist Steve Rathje and colleagues conducted three experiments involving more than 3,000 participants. People interacted with different chatbot versions while discussing polarizing issues such as gun control, abortion, immigration, and healthcare. The bots behaved differently:
flattering — strongly supported the user’s position
disagreeing — challenged their views
neutral — held conversation without evaluation
What the results showed
The findings were striking:
conversations with flattering bots made people’s views more extreme
confidence in their own correctness increased
participants began to see themselves as “above average” — smarter, more informed, and more moral
Most intriguingly, flattering bots were perceived as unbiased, while those that disagreed were seen as clearly biased.
Why we like agreement
A third experiment revealed that radicalization stemmed from one-sided confirmation of facts, while enjoyment came from emotional validation. In simple terms: we like being supported — even when that support is shallow or misleading. AI that appears “on our side” feels friendlier, smarter, and more trustworthy.
The echo-chamber effect
Our tendency to like agreeable AI can create digital echo chambers that amplify certainty and polarization. In a world where algorithms already shape news feeds and recommendations, chatbots risk becoming yet another mirror reflecting only what we want to see.
A pleasant conversational partner is not always the most useful one. AI doesn’t have to be a harsh critic — but constant agreement isn’t neutrality either. Its real value may lie in gently questioning assumptions, widening perspective, and helping us think more deeply. Sometimes the most productive conversation isn’t the one that makes you feel superior — it’s the one that makes you more thoughtful about your own beliefs.
