Pittsburgh Learning Classifier Systems for Explainable Reinforcement Learning: Comparing with XCS

Rule-based AI systems offer better explainability than deep learning: The paper demonstrates that Learning Classifier Systems (LCSs), which operate through interpretable rules rather than neural networks, provide a more transparent alternative to the "black box" deep learning approaches commonly used in reinforcement learning today.

Simpler, more interpretable solutions are possible with Pittsburgh LCSs: The novel Pittsburgh systems (PPL-DL and PPL-ST) developed in this research produced more compact and interpretable rulesets compared to Michigan-style systems like XCS, meaning the AI's decision-making logic can be more easily understood by humans, even if this comes at higher computational cost.

Explainable AI can match or exceed performance of opaque systems: PPL-ST achieved equal or better performance than XCS, particularly in uncertain environments, suggesting that choosing interpretability doesn't necessarily mean sacrificing effectiveness—a key consideration for responsible AI deployment.

Trade-offs exist between explainability and computational efficiency: While the Pittsburgh systems produced more understandable policies, they required more computational resources than Michigan systems, highlighting that organizations must weigh the benefits of transparency against practical implementation costs.

Symbolic AI deserves renewed attention for high-stakes applications: The research suggests that rule-based evolutionary systems warrant reconsideration for reinforcement learning domains where explainability is critical, particularly in uncertain real-world scenarios where understanding why an AI made a decision matters.