Abstract:Machine learning algorithms are increasingly used to make or support decisions in a wide range of settings. With such expansive use there is also growing concern about the fairness of such methods. Prior literature on algorithmic fairness has extensively addressed risks and in many cases presented approaches to manage some of them. However, most studies have focused on fairness issues that arise from actions taken by a (single) focal decision-maker or agent. In contrast, most real-world systems have many agents that work collectively as part of a larger ecosystem. For example, in a lending scenario, there are multiple lenders who evaluate loans for applicants, along with policymakers and other institutions whose decisions also affect outcomes. Thus, the broader impact of any lending decision of a single decision maker will likely depend on the actions of multiple different agents in the ecosystem. This paper develops formalisms for firm versus systemic fairness, and calls for a greater focus in the algorithmic fairness literature on ecosystem-wide fairness - or more simply systemic fairness - in real-world contexts.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is that the existing research on algorithm fairness mainly focuses on the static situation of a single decision - maker (such as a single bank, university or employer), while ignoring the situation in the real world where multiple agents work together in a larger ecosystem. Specifically:
1. **Fairness issues in multi - agent systems**: Most of the existing fairness research only considers the perspective of a single decision - maker. However, in practical applications, such as in the loan scenario, there are multiple lending institutions, policy - makers and other entities that affect the results. Therefore, the behavior of a single decision - maker and its fairness cannot fully reflect the fairness of the entire system.
2. **Long - term dynamic fairness**: In addition to the influence of multiple agents, this paper also focuses on the fairness issue under long - term dynamics. That is, how the behavior and results of individuals are affected by a series of decisions over time, and how these effects vary among different groups.
3. **Systemic Fairness**: In order to better understand and solve the above problems, this paper introduces the concept of "systemic fairness". Different from "corporate fairness" which only focuses on a single institution or decision - maker, systemic fairness emphasizes starting from the entire ecosystem and evaluating the impact of all participants' behaviors on overall fairness.
### Main contributions of the paper
1. **Propose and study systemic fairness**: This is a novel and important extension, aiming to弥补 the deficiencies of the existing literature on the long - term dynamic fairness of multi - agent systems.
2. **Extend the long - term algorithm fairness research to multi - agent systems**: Although there has been research on long - term fairness, they usually only focus on a single agent and do not discuss the difference between systemic fairness and corporate fairness.
3. **Develop a formal framework**: Used to study the differences between corporate fairness and systemic fairness and provide a framework for analyzing systemic fairness.
4. **Demonstrate the importance of systemic fairness through specific cases**: Take the behaviors of two companies in the loan scenario as an example to illustrate the importance of considering systemic fairness.
5. **Expand the framework to model multiple companies and agents**: Show how to use the methods of complex systems to understand emergent behaviors and the resulting systemic fairness issues.
### Formula representation
Some formulas involved in the discussion are represented in Markdown format as follows:
- Quality scoring function: \[ Q_k=\varphi(X_k) \]
- Estimated quality: \[ \hat{Q}_{lk}=f(X_k, s_l) \]
- Decision rule: \[ D_{lk}=\hat{Q}_{lk}>\tau_l \]
- Result function: \[ O_k = g(D_{1k}, D_{2k},\ldots, D_{mk}) \]
These formulas help define the relationships between individuals, companies and ecosystems, and how to evaluate fairness and effectiveness in different scenarios through simulation.
### Summary
By introducing the concept of systemic fairness, this paper hopes to promote the academia and practitioners to pay more attention to evaluating and improving the fairness of algorithms from the perspective of the overall ecosystem, thereby ensuring broader social fairness.