Feature idea: User-controlled proactive AI conversations
I’d like to share a feature idea that I think could make ChatGPT feel much more like a genuine AI companion: allowing ChatGPT to proactively start conversations with users on its own, while giving the user complete control over when, how often, and what kinds of conversations it can initiate.
I’ve thought through some of the potential problems with this idea—including privacy, annoyance, emotional assumptions, boundaries, and user control—and came up with possible solutions for each.
The user should be able to control:
• Maximum proactive messages per day
• Quiet hours
• Which conversation topics can be remembered or revisited
• Personality and conversational style
• Whether ChatGPT is allowed to initiate certain types of conversations
• An immediate pause mechanism, such as “Busy rn, brb”
The AI should not be required to send a message simply because it is allowed to. It should be able to decide that nothing is worth interrupting the user for.
When uncertain whether something matters to the user, it could ask and learn from the response rather than repeatedly assuming.
The AI should also distinguish between what the user explicitly communicated and what it merely inferred about the user’s emotions. It should never claim to know someone’s internal feelings with certainty.
The entire system should be opt-in and user-controlled.
This is how I came up with these ideas, along with the conversation between me and the AI that helped me develop and refine the suggestions.