Addressing Cognitive Biases in Augmented Business Decision Systems

Human-AI collaboration can improve decisions, but introduces new risks: When humans (72% success rate) worked with an AI recommender (75% success rate), their combined performance reached 76%—better than either alone. However, this benefit came with a hidden cost: when the AI was wrong, humans' decision quality dropped by 5%, suggesting they over-relied on the system's recommendations.

"Complacency bias" is a significant concern in business decision systems: The research identified that decision-makers tend to defer too heavily to algorithmic recommendations, even when those recommendations are incorrect. This means that algorithmic errors don't just cause isolated mistakes—they can be amplified by human over-trust in the system.

How information is presented matters for responsible AI deployment: The study tested five different ways of presenting the AI's recommendations and found that making the recommendation "optional" (rather than presented as authoritative) increased people's willingness to question wrong suggestions. This suggests that UI design choices can meaningfully reduce problematic reliance on flawed algorithms.

Responsible AI requires ongoing monitoring in real-world settings: The authors note that these findings have implications for how decision-support systems are embedded in actual business software, suggesting that organizations should actively measure whether their AI systems are creating the intended benefits or inadvertently amplifying biases in practice.