Irrationality-Aware Human Machine Collaboration: Mitigating Alterfactual Irrationality in Copy Trading

The data generated by human decisions inherit human irrationality, yet artificial intelligence (AI) algorithms often operate under the implicit assumption that the data they train on is rational. We challenge this convenient assumption and develop an irrationality-aware human-machine collaboration (IA-HMC) framework to address it. Within this framework, we propose a new concept of alterfactual irrationality, which identifies irrational human decisions influenced by irrelevant alternative inputs. We then develop irrationality-aware machine learning methods to account for this irrationality and augment human-machine collaboration via the contraction method. We apply the framework to mitigate the irrationality present in the decision-making process of copy trading, a practice that enables layperson investors (followers) to assess and replicate expert traders’ trades. We specifically identify two main sources of alterfactual irrationality prevalent among followers’ decisions: herding behavior and identity bias. Our results demonstrate that our proposed method outperforms both the original follower decisions and the contraction method, improving the win ratio by 49.0% and 10.2%, respectively. This study marks the first attempt to design machine learning algorithms to overcome human decision irrationality, suggesting that aligning AI with human preferences needs to account for human irrational behaviors.

History: Ahmed Abbasi, Senior Editor; Xiao Fang, Associate Editor.

Supplemental Material: The online appendix is available at https://doi.org/10.1287/isre.2023.0591.