Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
It is increasingly common for algorithms rather than people to take complex decisions. Many of those algorithms are however written by people and the information available to them when developing an algorithm will influence the algorithm they write. We implement such a setting in a controlled environment where participants program an explicit strategy to play the minority game on their behalf in a multi-round strategy tournament. The minority game is a stylized example of the large and important class of games with strategic substitutes such as market-entry and congestion games. Given the large strategy space and multiplicity of equilibria of the minority game, developing a successful strategy is no easy task. Over three experiments we vary the information available to participants to study whether more or better information can help people to improve their strategies over time. Providing participants with the strategies played by others turns out to reduce overall performance, while reducing the noise in feedback about the potential performance of a strategy under consideration increases efficiency. We argue that these different information structures put different types of strategies in a good light. The increased popularity of the favored strategies in turn drives overall efficiency.