On the Effect of Information Asymmetry in Human-AI Teams

Information asymmetry is key to effective human-AI teamwork: The research demonstrates that humans and AI can achieve better combined performance than either could alone, but only when humans have access to contextual information that the AI lacks. This suggests that successful AI deployment requires identifying what unique knowledge humans bring to the table.

Explainability alone isn't enough: While many organizations focus on making AI systems more transparent and interpretable, this paper shows that explainability is only useful when there's actual complementarity potential between human and AI capabilities. Simply explaining AI decisions won't improve outcomes if humans don't have relevant information to contribute.

Humans need the right information to override AI: The experiment confirms that when humans have contextual knowledge the AI doesn't possess, they can meaningfully adjust AI recommendations to improve final decisions. This validates the practical value of human oversight in AI systems, but only when humans are positioned to catch what the AI misses.

Real-world AI teams require different expertise: The findings highlight that effective human-AI collaboration isn't about humans becoming better at understanding AI—it's about leveraging the different types of information each party can access. Organizations should design AI systems with the understanding that humans will contribute contextual and situational knowledge rather than just validating algorithmic outputs.