Feedback Loops and Label Bias Questions

Understand how model decisions and product actions can influence the data generation process and labels, creating feedback loops and bias. Identify selection and measurement biases, how logging and instrumentation choices censor data, and how product interventions change user behavior. Discuss detection techniques and mitigation strategies such as randomized exploration, counterfactual logging, inverse propensity weighting, off policy evaluation, and causal analysis to reduce leakage and unintended amplification of biases.