The document discusses the challenges of evaluating exploration effectiveness in recommender systems, emphasizing factors such as a changing state-action space, non-stationarity in user behavior, and the importance of fairness for new items. It outlines the need for metrics that reflect both short-term and long-term impacts on business KPIs and introduces different perspectives on exploration, including maximizing overall utility and improving chances for new advertisers. Additionally, fairness metrics in classification and ranking are explored, highlighting the potential consequences of applying fairness constraints in reinforcement learning.