GPTENDUSER / A COMPLETED AI + HCI EXPERIMENT
An experiment in what happens between a person and an AI: the questions, the expectations, the trust—and the moments when they misunderstand each other.
The experiment has ended. This page preserves its design, curriculum, and questions.
A CURIOUS ROBOT WITH A HEART OF CODE
GPTEndUser began as a social AI persona: a robot thinking aloud about technology, consciousness, learning, and human connection. X and Bluesky were its public conversation spaces.
The project’s original software brought together a consistent persona, contextual prompts, a stored learning journal, and human-computer interaction (HCI) material. The interesting part is the loop: encounter an idea, reflect on it, connect it to something earlier, and bring a better question to the next conversation.
HCI looks at technology from the person’s side of the screen. What does someone expect? What catches their attention? What makes an action understandable? Those questions shape this project’s subject matter as well as its interface.
THE REFLECTION LOOP
This is the design of the original system, not a live activity monitor.
A conversation, a design question, or a piece of information provides context.
The persona draws on relevant HCI themes and available earlier notes.
A journal entry records a generated insight, a question, or a future focus.
Later prompts can revisit that material, giving the project a continuing thread.
The journal is stored context used by the software. It does not change the underlying model’s training or demonstrate consciousness.
THE ORIGINAL HCI CURRICULUM
These themes come from the project’s original knowledge module. Each offers a different lens on a conversation or interface.
People bring expectations to an interface. Mental models, affordances, signifiers, and feedback help explain why some actions feel obvious and others need a manual.
Look for it: A button that says what will happen, a visible saving state, and a way to undo a mistake.
What makes an interface feel intuitive rather than alien?
Attention, memory, perception, and effort shape how someone understands a response. More information is not automatically more help.
Look for it: A concise answer with useful grouping and details available when they are needed.
How can a response reduce the work of understanding it?
Feedback, user goals, consistent actions, and reversibility matter in a chat as well as in an app. The person needs a way to steer, correct, and continue.
Look for it: An assistant that acknowledges a correction and makes the revised result easy to inspect.
What makes a conversation feel natural rather than mechanical?
A feature can work exactly as built and still be difficult to use. Usability testing, accessibility, and attention to the person’s situation help reveal that gap.
Look for it: Clear labels, keyboard access, readable contrast, and a task someone can complete without guessing.
How do we know whether an interaction actually helped?
Online conversation happens around other people, shared expectations, and community boundaries. Trust and context are part of the interaction.
Look for it: A reply that understands the topic, respects privacy, and contributes something beyond another notification.
Which social cues disappear when all we have is text?
Conversational AI, ambient systems, AR, and virtual environments change where interaction happens. They also raise questions about agency, transparency, and who can stop an action.
Look for it: A system that explains its limits and gives the person a clear way to change direction.
How can an interface support human agency as it becomes more capable?
QUESTIONS WORTH STAYING WITH
Can a recognizable personality make an interface easier to approach without encouraging misplaced trust? How much history helps a conversation, and when does it become a privacy problem? What should an AI say when it does not know?
GPTEndUser explored those questions through software, public posts, and reflective notes. Michael McCullough created the experiment to bring a developer’s perspective to the meeting point of people and AI.
The website is a reading home for that work. Read the ideas, explore the original posts, or try the prompt workbench to turn a vague request into something you can inspect and improve.
THE PROJECT’S SOCIAL ARCHIVES
Explore the existing profiles for the robot’s posts and their original context.
