Models of Intervention: Helping Agents and Human Users Avoid Undesirable Outcomes

Area of Science:

• Human-Computer Interaction
• Artificial Intelligence
• Cognitive Science

Background:

• Assisting users in unfamiliar digital environments is crucial for preventing frustration and undesirable outcomes.
• The Intervention Problem, distinct from Plan Recognition, involves deciding the optimal timing for user assistance.
• Existing Plan Recognition methods may not adequately address the dynamic nature of user assistance.

Purpose of the Study:

• To formalize and solve a family of Intervention Problems.
• To develop and evaluate a Human-Aware Intervention system for real-time user guidance.
• To improve the prediction of user behavior and potential undesirable outcomes.

Main Methods:

• Formalizing Intervention Problems and solving them with a combination of Plan Recognition and classification algorithms.
• Generalizing to Human-Aware Intervention by developing a learned model using a revised feature set.
• Conducting a human-subject study to evaluate the Human-Aware Intervention model's performance.

Main Results:

• Classification algorithms outperformed three recent Plan Recognition approaches in benchmark tests.
• The revised model for Human-Aware Intervention demonstrated superior performance in predicting when human users might trigger undesirable outcomes.
• The human-subject study confirmed the effectiveness of the developed Human-Aware Intervention system.

Conclusions:

• A novel approach combining Plan Recognition and classification effectively addresses the Intervention Problem.
• Human-Aware Intervention systems can be successfully developed and deployed to guide users in real-time.
• The developed models show significant promise for enhancing user experience and task completion in complex digital environments.