Integrating in-situ learning supports into stochastic generative AI systems

Determine the extent to which in-situ support approaches for complex software—such as interactive walkthroughs, example-centric help, and context-aware agents—can be integrated into generative AI–assisted 3D modeling systems characterized by stochastic outputs and unpredictable behavior.

Background

The paper reviews prior HCI work that embeds in-situ supports—interactive walkthroughs, examples, and context-aware agents—directly into interfaces to scaffold learning in feature-rich, largely deterministic software. These approaches assume predictable system behavior and stable affordances.

In the context of generative AI, particularly 3D modeling systems with stochastic outputs and opaque failure modes, the authors note uncertainty about whether such supports transfer effectively. They explicitly flag as an open research question the extent to which these deterministic-era help mechanisms can be adapted to generative systems with unpredictable model behavior.