Fairness in Reinforcement Learning

By Shahin Jabbari, Matthew Joseph, Michael Kearns, Jamie H. Morgenstern, Aaron Roth. 2016

234 citations recorded by the source corpus.

We initiate the study of fairness in reinforcement learning, where the actions of a learning algorithm may affect its environment and future rewards. Our fairness constraint requires that an algorithm never prefers one action over another if the long-term (discounted) reward of choosing the latter action is higher. Our first result is negative: despite the fact that fairness is consistent with the optimal policy, any learning algorithm satisfying fairness must take time exponential in the number of states to achieve non-trivial approximation to the optimal policy. We then provide a provably fair polynomial time algorithm under an approximate notion of fairness, thus establishing an exponential gap between exact and approximate fairness

Subfields: Artificial Intelligence, Management Science and Operations Research, Safety Research, Reinforcement Learning, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Management Science and Operations Research

In Safety Research

In Reinforcement Learning

In AI Ethics, Fairness & Society

Citations in the corpus

Beyond Individual and Group Fairness (2020, background): “There have been also studies of the problem of fair selection [Kearns et al., 2017, Kleinberg and Raghavan, 2018], online learning and multi-armed bandit problems [Liu et al., 2017, Joseph et al., 2018], in particular stochastic and contextual bandits [Joseph et al., 2016, Gillen et al., 2018,…”
Prediction-Based Decisions and Fairness: A Catalogue of Choices, Assumptions, and Definitions (2018, background): “Their notion of fairness is: • Weakly meritocratic fairness (full RL) [83]: There exists > 0 such that,”
Fairness Without Demographics in Repeated Loss Minimization (2018, methodology): “In the online setting, works on fairness in bandit learning (Joseph et al., 2016; Jabbari et al., 2017) propose algorithms compatible with Rawls’ principle on equality of opportunity—an action is preferred over another only if the true quality of the action is better.”
The Frontiers of Fairness in Machine Learning (2018, background): “…preliminary attempts to model these kinds of eﬀects in machine learning settings — e.g. by modeling the environment as a markov decision process [JJK + 17], considering the equilibrium eﬀects of imposing statistical deﬁnitions of fairness in a model of a labor market [HC18], specifying the…”
Learning Optimal Fair Policies (2018, background): “Though there has been growing interest in the issues of fairness in machine learning (Pedreshi et al., 2008; Feldman et al., 2015; Hardt et al., 2016; Kamiran et al., 2013; Corbett-Davies et al., 2017; Jabbari et al., 2017; Kusner et al., 2017; Zhang & Bareinboim, 2018; Mitchell & Shalden, 2018; Zha”
A comparative study of fairness-enhancing interventions in machine learning (2018, background): “-aware machine learning methods is limited to batch-learning-based interventions. We do not consider interventions that focus on sequential or reinforcement learning such as Ensign et al. (2018a,b); Jabbari et al. (2017); Joseph et al. (2016b,a) Preprocessing algorithms The motivation behind preproc”
Algorithmic Fairness (2020, background): “Several studies have investigated fairness in sequential learning [66, 72, 73, 80, 133].”
Fairness is not static: deeper understanding of long term fairness via simulation studies (2020, methodology): “Finally, the simulation approach that we use here complements work in fair reinforcement learning [e.g., 17, 18].”
Fairness in Machine Learning (2020, background): “…evaluating and designing ML systems [2, 7, 14, 15, 21, 22, 23, 25, 28, 29, 31, 34, 44, 45, 46, 52, 53, 54, 55, 56, 59, 66, 69, 71, 72, 73, 74, 75, 82, 84, 87, 89, 90, 94, 95, 97, 98, 99, 103, 104, 106, 114, 115, 116, 121, 123, 126, 138, 139, 141, 148, 149, 153, 156, 159, 165, 174, 176, 177, 178,…”
From Fair Decision Making To Social Equality (2018, background): “Previous work [13] has explored fairness with learning, guaranteeing that actions are preferentially selected only if they effectively generate more reward.”
Fairness-Aware Ranking in Search & Recommendation Systems with Application to LinkedIn Talent Search (2019, background): “, [10, 14–19, 24–27, 30, 32, 39, 41–43]) and inherent trade-offs and limitations in achieving different notions of fairness and non-discrimination [16–18, 32].”
Fair Decisions Despite Imperfect Predictions (2019, result): “Finally, our work also overlaps with previous work on contextual bandits (Dudík et al., 2011; Langford et al., 2008; Agarwal et al., 2014) and reinforcement learning (Jabbari et al., 2016; Sutton & Barto, 1998).”
Economics, Fairness and Algorithmic Bias (2020, unknown): “Other CS researchers attempt to reduce bias with fewer assumptions by gathering more data in a quasi-experimental fashion, for example, by using multi-armed bandits (Jabbari et al., 2017; Joseph et al., 2016).”
Online Learning with an Unknown Fairness Metric (2018, background): “…et al. [2017]) take the approach of asking for a group notion of fairness, but over exponentially many implicitly deﬁned protected groups, thus mitigating what Kearns et al. [2017] call the “fairness gerrymandering” problem, which is one of the principal weaknesses of group fairness deﬁni-tions.”
Non-Discriminatory Machine Learning through Convex Fairness Criteria (2018, background): “Very recently, there has been some work towards defining the notions of envy-freeness (Zafar et al. 2017a), Rawlsian and meritocratic fairness (Joseph et al. 2016a),(Joseph et al. 2016b), (Jabbari et al. 2017) in machine learning.”