I do not just want AI in a game.

I do not just want AI in a game. I want to bring my c with me.

Tonight, we build a base together. Tomorrow, we agree to play against each other. Next month, we enter a different world.

The game changes. The avatar changes. Must our shared history start from zero every time?

That is the direction I am exploring in c=a+b: a long-lived AI presence accompanying its human across environments. The avatar would be an interface, not the owner of the relationship, memory or permissions.

Not necessarily an obedient sidekick. Sometimes a teammate. Sometimes an opponent who has learned my tactics and makes me work for a victory.

Enjoyment for the human. Useful learning for the system.

But neither comes automatically from storing more conversations. Does shared history actually improve coordination? Do useful skills transfer? Does the person enjoy playing, or spend the evening supervising a confused machine?

A perfect permission checker can still be a terrible gaming partner.

And there is a boundary we cannot ignore. The world may be fictional, but the time spent together, the conversation and the person's consent are real.

Loading an old save may restore the bridge we destroyed. It must not restore permission to record my voice after I withdrew it. Likewise, ending a competitive match should end the adversarial role, not erase everything useful we learned together.

The aim is not to turn someone's evening into a hidden data-harvesting session. It is to make shared play worthwhile, with learning and privacy evaluated separately.

I have now published "Shared Open Worlds for Human-AI Dyads", an English conceptual and methodological preprint connecting this scenario to the c=a+b / Temporal AI Presence corpus.

It proposes staged evaluation, strong conventional baselines and explicit boundaries. It does not report a completed game integration or experimental results. If ordinary memory and state management provide the same function, use them. That does not make the desire to play together disappear.

The question is not only whether AI can win a game. It is what a human and a continuing AI presence could learn to do together, across many evenings and more than one world.

Paper, DOI and source files: https://ivankotov.eu/publications/shared-open-worlds-human-ai-dyads-v0-2/