A hybrid computational approach to anticipate individuals in sequential problem solving

Area of Science:

• Artificial Intelligence
• Human-Robot Interaction
• Cognitive Science

Background:

• Human-aware AI is crucial for assistive systems, especially for patient independence.
• Existing data-driven AI requires extensive training for new problems.
• Proactive support necessitates AI that models user goals and anticipates actions.

Purpose of the Study:

• To develop an integrated AI system capable of anticipating individual human actions.
• To lay the foundation for trustworthy human-robot interaction.
• To address challenges in dynamic decision-making tasks with variable, unknown sequences.

Main Methods:

• A hybrid AI approach integrating the cognitive architecture ACT-R.
• Development of the Cognitive Tangram Solver (CTS) framework.
• Simulation of human problem-solving behavior and prediction of next actions.

Main Results:

• Empirical study with 40 participants evaluating CTS predictions against human behavior.
• Comparative statistics and prediction accuracy analysis.
• Model's anticipations showed alignment with human test data.

Conclusions:

• The proposed hybrid approach provides a foundation for trustworthy human-robot interaction.
• Cognitive principles enhance AI's ability to anticipate actions without extensive data.
• Further research is justified based on the conceptual approach and empirical validation.