Research Seminar by Tian Lu | 1 + 1 > 2? Information, Humans, and Machines

1 + 1 > 2? Information, Humans, and Machines

The integration of artificial intelligence (AI) into collaborative efforts with humans is becoming pervasive across various business domains. In this talk, I will introduce key findings from a series of my recent research on the potential for humans to add value to machine processes, particularly in the context of complex human-machine interactions and the challenges of processing vast information scales. 

Despite machines’ superior data processing capabilities, understanding the value of human-machine collaboration remains critical. Particularly, recent attempts to uncover machine learning algorithms’ “black boxes” aim to decrease human resistance and enhance decision-making efficiency, though findings have been inconclusive. Our studies address these issues by examining the interplay between information complexity and machine explanations through a two-stage field experiment with a major Asian microloan company. Leveraging dual-process theories of reasoning, we varied information complexity and the presence of machine explanations. In one study, we found that machine explanations alone do not enhance outcomes; however, when combined with extensive information, they significantly improve decision-making, notably lowering default rates compared to machine-alone decisions. Our empirical analysis demonstrates that this combination stimulates active human rethinking, which increases prediction accuracy by correcting machine errors and reducing decision biases. 

In another project, we identified additional specific conditions under which humans are likely to reevaluate and override machine recommendations, thereby enhancing collaborative value. These findings highlight the critical role of human-machine collaboration and provide valuable insights for optimal task allocation in system design.