You have deployed a Gradient Boosting Machine (GBM) model for loan approvals. The regulatory body requires explanations for every rejection. Develop a technical guide for implementing Explainable AI (XAI) in this production environment. 1) Explain how you would use SHAP (SHapley Additive exPlanations) values to generate global and local feature importance. 2) Discuss the trade-offs between computational latency and explanation accuracy in a real-time API context. 3) Propose a method for generating 'counterfactual explanations' (e.g., 'If your income was $5k higher, you would have been approved'). 4) Address potential fairness issues: How would you detect and mitigate proxy discrimination using the explanation data?