Objectives: In occupational therapy, documentary requirements include a comprehensive report manually written or encoded post-client-session, which lengthens daily working hours. With increasing interest in using artificial intelligence to improve productivity in practice, we evaluate an early version of a tool that automatically generates a report from mixed-language (Filipino-English) conversations between practitioners and their clients during a session.
Methods: We simulate two initial evaluation sessions, each with a report written by designated occupational therapists (OTs) and a tool-generated report. For each session, two other OTs not present during the simulation listened to the audio recording, and either (1) wrote a report or (2) revised the generated report. We compare all reports using automatic evaluation metrics and, recorded changes in documentation time, with reports’ quality validated by other OTs.
Impact/Results: The generated reports had high semantic similarity (cosine similarity: 0.81 for Case 1, 0.82 for Case 2) to the actual reports, comparable to those manually written from recordings, with at least 85% reduction in documentation time. Validation results also revealed the human tendency to rely on generated outputs.
Conclusion/Contribution to the occupational therapy profession: The documentation tool exhibits potential in improving OT productivity, but the risk of overdependence must be carefully considered. An OT’s inputs in AI tool development is vital in creating culturally relevant and clinically valid tools compliant with legal and ethical standards for practice, education and research.