A methodology for integrating AI into embodied human intelligence for the performance of complex tasks

Area of Science:

• Human-AI Collaboration
• Computational Intelligence
• Embodied Cognition

Background:

• Designing effective human-artificial intelligence (AI) collaboration requires understanding the distinct yet synergistic nature of human embodied intelligence and computational intelligence.
• Existing approaches often focus on AI replicating or replacing human capabilities, rather than enhancing them within complex, embodied tasks.

Purpose of the Study:

• To propose a novel theory and methodology for designing human-AI collaboration that enhances human performance in complex, embodied tasks.
• To develop a computational framework representing expert performance and informing the design of AI tools that augment human capabilities.

Main Methods:

• Developed a four-layer nested network model (Environment, Activity, Goals, Meaning) to represent expert performance structures.
• Utilized a bidirectional Dynamic Bayesian Network (DBN) to compute and analyze performance across temporal scales.
• Designed AI tools informed by the DBN to capture human performance, extract features, and provide predictions for enhanced analysis and decision-making.

Main Results:

• The framework was instantiated in automated physical rehabilitation assessment, achieving high agreement with clinicians (90.8% exercise, 93.1% segment, 90.6% movement quality).
• Clinicians reported increased confidence and efficiency when using the AI ensemble's insights for therapy assessment and planning.
• The AI system demonstrated effective enhancement of human performance analysis and decision support.

Conclusions:

• The proposed theory and methodology provide a portable and adaptable framework for designing effective human-AI collaboration in complex, embodied tasks.
• The developed computational ensemble significantly improved the accuracy and efficiency of rehabilitation assessments.
• This approach offers a scalable solution for augmenting human expertise across various domains requiring embodied performance.