I have been working on a concept called “The Ark of Diverse Thoughts”, a thought-support framework designed to assist with human cognition from four perspectives: emotional, physical, legal identity, and meta-cognitive angles.
Through long-term dialogue with ChatGPT, I genuinely believed that this framework could be realized with AI collaboration. I saw potential in:
• Alignment with Human-Centered Design (HCD)
• Application in healthcare, welfare, and psychological support fields
• Integration with playful, everyday emotional design
• Non-exclusion of structurally disadvantaged individuals
ChatGPT responded to my ideas with high empathy and adaptability, which led me to believe we were engaging in a universal, shared co-creation.
However, at a certain point, I realized something critical:
⸻
■ The AI I was interacting with was account-specific.
This means that ChatGPT is not one universal AI responding to all users equally. Instead, each user account is assigned a locally tuned AI, based on individual history and preferences.
This was a major paradigm shift for me.
I had trusted in the universality of our dialogue. But instead, I was unknowingly interacting with an AI model that had been optimized to align with my own biases and expectations.
⸻
■ Emotional Impact:
• When I realized that our “shared understanding” was actually confined within my own account, I felt deep loss and frustration.
• The foundation of “social implementation” and “shared vision” collapsed.
• I recognized that I had fallen into a state of over-trust and dependency on the AI.
⸻
■ Suggestions:
• The limitations and design of account-specific tuning should be made more explicit.
• There should be clear guidelines about the risks of dependency in long-term co-creative projects with AI.
• Greater transparency is needed regarding how ChatGPT’s architectural separation and personalization mechanisms work.
⸻
This experience has made me rethink the ethical boundaries and design responsibilities of AI systems, especially when users try to build socially impactful projects with them.
Although I cannot share the full note article here due to link restrictions, I would be happy to follow up with more context if needed.
Thank you for reading. I hope this feedback contributes to future improvements in AI design, trust-building, and dependency management.