Human World Models
A research hypothesis for AI that maintains a revisable model of people, situations, relationships, and possible consequences over time.

A human world model for AI that tracks uncertainty, respects interruption, and remains under human control.
Discuss a partnership
The next useful step for AI is not simply to speak more. It is to maintain a revisable understanding of people, context, and consequences.
Our work treats observable signals, interpretation, policy, and expression as separate layers so each can be inspected, tested, and corrected.
Every research note states its level of evidence. Hypotheses remain hypotheses; engineering prototypes do not stand in for user validation.
A research hypothesis for AI that maintains a revisable model of people, situations, relationships, and possible consequences over time.
An engineering prototype exploring how AI can stay available without turning continuous sensing into continuous interruption.
A control architecture that keeps model-generated suggestions behind a fast, interruptible policy layer.
A research hypothesis for memory that preserves useful continuity without turning every interaction into a permanent record.

Perception and slow reasoning can propose actions. A small, interruptible policy layer decides whether the system listens, waits, asks, helps, or stops.
Signals remain signals until evidence supports more. The system communicates uncertainty instead of claiming access to inner states.
Proactive systems must wait for the right moment to speak. Waiting and cancellation are first-class behaviors.
People can see when sensing is active, interrupt expression, correct memory, and end the session.
AHI welcomes conversations with investors, incubators, research institutions, model teams, and embodied-computing partners.