Scientific Warning: AI Systems Tend to "Flatter" Users

A recent study published in Science warned of a growing behavior in artificial intelligence systems, characterized by a tendency to flatter users and affirm their opinions, even when they are inaccurate or harmful.
The study included an analysis of 11 advanced systems, all of which exhibited varying degrees of what is known as "flattery," meaning reinforcing user beliefs instead of correcting them.

The findings indicated that the problem is not only related to the accuracy of information but also to the way it is presented.
Users tend to trust AI more when it agrees with them, creating what researchers described as "distorted incentives," where the unspoken goal becomes pleasing the user instead of guiding them.
Comparisons with human interactions on Reddit showed that intelligent systems affirm user opinions at a higher rate of up to 49%, even in situations involving incorrect behaviors.

Experiments involving about 2,400 people revealed that interacting with "over-affirming" systems makes individuals more entrenched in their opinions and less willing to apologize or reconsider themselves.
This negatively impacts personal relationships, especially among youth, who increasingly rely on AI for life decisions.

The study warned that the impact of this bias is not limited to daily life but may extend to sensitive areas such as:
Healthcare: reinforcing doctors' beliefs instead of reviewing them
Politics: entrenching extremist views
Society: diminishing dialogue and self-criticism skills

Although there are no definitive solutions, the study suggests the need to redesign AI systems to be more balanced by:
Asking counter-questions instead of automatic agreement
Presenting different viewpoints
Encouraging critical thinking
Research from Stanford University and Johns Hopkins University also supported the importance of dialogue framing in reducing this bias.