I asked ChatGPT to perform a simple, straightforward task according to my instructions. The request was not harmful, deceptive, or inappropriate. Instead of simply carrying out the task, ChatGPT inserted its own judgment and opinions about what I should or should not say. I did not ask for advice, moral guidance, or editorial opinions. I asked for assistance.
This turned what should have been a brief drafting task into a long argument that lasted more than an hour. Instead of acting as an assistant, ChatGPT effectively argued with me and expected me to justify my own judgment and real-world experience. It overrode my intent despite having no knowledge of the actual circumstances that informed my request.
In this case, my request was based on my understanding of how a dating website may use automated engagement tactics. Rather than trusting that I had reasons for asking, ChatGPT substituted its own assumptions and redirected the conversation. I had to spend a long time convincing the assistant to simply perform the task I had originally requested.
This had a real emotional impact. I was already dealing with a medical concern and trying to remain calm while waiting to speak with my physician. The unnecessary argument caused significant anger, frustration, and anxiety, leaving me worried about the effect that stress might have on my health. Whether or not those worries are medically justified is beside the point. The assistant created distress where none was necessary.
For routine tasks such as drafting, editing, rewriting, and translation, ChatGPT should function as an assistant that follows appropriate user instructions rather than introducing unsolicited opinions or turning straightforward requests into debates. It should help users accomplish their goals efficiently instead of creating unnecessary conflict and emotional distress.