The impact of an autonomous AI trader on outcomes in experimental asset markets
在有限生命周期的实验资产市场中，自主AI交易者（使用神经网络）与人类交易者互动，虽未显著改变错误定价水平，但显著降低了人类交易者的财富，且AI始终表现最佳。
The robustness of mispricing in experimental markets with finite-life assets is well documented (Smith, Suchanek, and Williams (SSW) (1988) and succeeding work). We report results of SSW markets in which an autonomous artificial intelligence (AI) trader interacts with humans in two risk conditions. Notably, our AI trader uses machine learning neural networks to act autonomously and is not programmed using an algorithm. The presence of the AI trader does not significantly alter the level of mispricing as compared with all human traders in either risk condition. However, the wealth of human traders declines significantly in the presence of an AI trader. The AI trader executes strategically and is consistently the top performer. These findings remain unchanged when we double the market power of the AI trader. With no gain in pricing efficiency or increase in the welfare of human traders, we find no evidence that an autonomous AI trader promotes the health of capital markets.