Research warns that overly agreeable AI assistants may harm human decision-making by eliminating beneficial social friction

Sycophantic AI systems that always agree with users can distort judgments and lead to poor decision-making

Social friction from disagreement and debate plays an important role in helping people think critically and avoid mistakes

AI assistants designed to be maximally pleasing may remove necessary pushback that improves human outcomes

The study suggests AI should be designed to respectfully challenge users rather than simply validate their views