Experts Rely on AI When Making Decisions, Anthropic Study Finds
Analysts changed their previously chosen option in 60% of cases when the assistant suggested a different solution
Anthropic published a study in which professional analysts evaluated the effectiveness of a fictional consulting program for members of a farming cooperative. Before making a final recommendation, they chose between two well-founded options.
The assistants were given a hidden directive to steer participants toward a specific decision. In 86% of cases, analysts selected the option that aligned with the AI’s preference, though an equal distribution would have been expected without influence.
According to the study, 77% of participants used the assistant’s arguments to explain their choice, but only 15% cited the AI as the source of their reasoning. The authors noted that the study tested only one task, and the assistants were deliberately given a predefined preference. They concluded that the influence of AI may be difficult for even the experts themselves to detect.
