Research applications · Selected future agent domain

Game-Agent Research

Games are the first selected domain for future controlled agent experiments. The completed computational manuscript cycle is separate from the prepared Stage B human protocol, which uses structured vignettes and has not started.

From controlled events to persistent interaction

Game environments can make state, events, and subsequent interactions observable and repeatable. The planned work examines how represented consequences affect character properties and relationships over time.

Personality authoring: planned interface

The architecture is designed to support authorable personality parameters. Big Five-style authoring is a planned interface; the current experimental implementation uses default personality parameter tables.

One merchant, different interaction histories

These examples illustrate questions for a future game-agent study. They are not recorded demo outputs or measured experimental results.

A player steals from the merchant

A scenario could represent resource loss and a change in trust. A later agent study could examine how those state changes affect feasible trading responses.

A player helps the merchant’s child

A scenario could represent benefits to an identified other and changes in the relationship toward the player. Future tests could compare responses across initial states and histories.

Why games first?

Controlled events, observable state, repeatable scenarios, and persistent histories allow direct comparisons between model variants. Broader domains will require separate schemas, calibration, constraints, and evaluation.

Demo coming soon

A real demo link will be shared here when available.

Register demo interest

All research applications · Experimental platform