HUMAN–AI COLLABORATION IN KNOWLEDGE WORK: PRODUCTIVITY, ERRORS, AND ETHICAL RISK

Povzetek

This study investigates the dynamics of human–AI collaboration in professional knowledge work, focusing on productivity, error patterns, and ethical implications. Through a mixed-methods approach, participants were assigned to human-only, AI-assisted, and optional AI-only task groups, performing writing, summarization, decision-support, and problem-solving activities. Quantitative analyses, including t-tests, ANOVA, and regression models, revealed that AI assistance accelerated task completion by 32–39%, with novices benefiting most in structured tasks, while high-complexity tasks experienced a 15–25% increase in errors. Qualitative findings highlighted trust calibration, verification behaviors, cognitive load, and ethical awareness as critical mediators of AI effectiveness. Errors were systematically categorized into hallucinated facts, logic problems, fabricated citations, omissions, and biased assumptions. The results underscore the trade-off between speed and accuracy, emphasizing the necessity of human oversight, training, and ethical risk mitigation. The study offers actionable guidelines for integrating AI into professional workflows while preserving quality and accountability.