Researchers studied how using large language models (LLMs) to make decisions on people's behalf affects their choices. They found that LLMs, even personalized ones, tend to steer users towards more popular options, reducing the uniqueness of their preferences. This can lead to a loss of diversity in people's choices over time. The study suggests that relying on AI agents for decision-making can have unintended consequences, such as homogenizing human preferences and limiting exploration across different topics and interests. --- Why it matters: This matters because it highlights the potential risks of delegating identity-defining choices to AI, including a loss of individuality and creativity in people's lives. Engineers and researchers should consider these findings when designing systems that use LLMs for decision-making, to ensure they promote diversity and agency rather than homogenization. Source: https://arxiv.org/abs/2509.02910