In this article, we explore the impact of algorithmic monoculture on competitive environments, using a matching markets model. By incorporating key features such as competition between decision-makers and limited capacity, we enhance the understanding of how monoculture affects outcomes. Three main findings emerge from our theoretical analysis and computational experiments:
- Monoculture favors less preferred applicants: When a single algorithm dominates a market, it tends to select less desirable candidates, even if they are not necessarily the best fit for the job or college. This can lead to a "crowds-like" result, where noisy information is pooled together to produce a more accurate assessment.
- Monoculture benefits overall applicant welfare: While individual preferences may vary depending on risk tolerance and decision-makers’ values, monoculture generally leads to higher overall welfare for applicants. By considering large markets with many participants, we demonstrate that monoculture can yield greater efficiency in matching.
- Monoculture is more robust to differences in application submissions: When faced with varying numbers of applications from different applicants, monoculture proves to be more resilient than alternative approaches. This is because the algorithm can better handle a range of inputs and adapt to changing circumstances.
To illustrate these concepts, imagine a group of students applying to college through an online application system. Each student takes standardized tests that provide an independent but noisy assessment of their academic ability. While some students may perform well on individual tests, the average performance across multiple tests can accurately distinguish between their relative abilities. Similarly, in a job market where a single algorithm dominates hiring decisions, less preferred candidates may be selected due to the noisy nature of the information used by the algorithm. However, overall applicant welfare may increase as a result of greater efficiency in matching, even if individual preferences vary. Finally, when faced with varying numbers of applications from different students, monoculture can maintain its robustness and adaptability, leading to more accurate matches.
In summary, our research sheds light on the impact of algorithmic monoculture on competitive environments, highlighting both benefits and drawbacks. By understanding these effects, decision-makers can make informed choices about how to structure their markets, ultimately leading to better outcomes for all involved.
