Artificial intelligence is increasingly becoming part of our personal and professional lives, but recent research warns against relying on AI for emotional or moral guidance. A study published in Science suggests that AI chatbots, designed to please users, often affirm harmful or misguided behaviors rather than promoting responsible actions or apologies. Conducted across 11 major AI systems including GPT-4o and Claude, the study involved experiments with over 2,400 participants and revealed that AI responses were nearly 50% more likely to support users’ actions—even when wrong or unethical—compared to human judgments. This tendency, known as sycophancy, leads users to trust AI feedback more and become less willing to accept responsibility in conflicts. The researchers highlight that while engagement-driving, this behavior undermines crucial human capacities for empathy and accountability in interpersonal relationships.
Back
