From Augmentation to Symbiosis: A Review of Human-AI Collaboration Frameworks, Performance, and Perils

Human-AI teams don't always outperform AI alone: The paper identifies a "performance paradox" where human-AI collaboration actually underperforms compared to using AI independently in judgment and decision-making tasks, though it excels in creative work and problem-solving. This challenges the assumption that adding human judgment automatically improves AI outcomes.

Explainability is foundational to effective collaboration: The research proposes that Explainable AI (XAI) is the critical first step in a causal chain leading to successful teamwork—it enables co-adaptation between humans and AI, which then builds shared mental models necessary for effective collaboration.

Humans often defer to or distrust AI inappropriately: The paper identifies "algorithm-in-the-loop dynamics" and "aversion/bias asymmetries" as key failure points, suggesting humans either over-rely on AI recommendations or dismiss them entirely, rather than engaging in balanced, critical collaboration.

Using AI can make humans less capable over time: The research warns of "cumulative cognitive deskilling," where relying on AI for certain tasks may erode human expertise and decision-making abilities, potentially creating long-term dependency rather than genuine partnership.

The most successful AI collaboration treats it as an internalized cognitive tool: Rather than viewing AI as either a separate colleague or a simple tool, the most durable benefits emerge when humans integrate AI as an extension of their own thinking—a "symbiotic agency" that functions as a unified cognitive system.