RESPONSIBLE AI, COMPLIANCE, AND RISK

REGULATIONS IN BRIEF Rules about artificial intelligence are developing in many places, and details change. Some regions have broad laws that sort systems by risk and set duties for the riskier ones, while others rely on existing rules about privacy, consumer protection, and discrimination. You do not need to memorize any law as a beginner, but you should know that such rules exist, learn which ones apply where your system operates, and involve legal experts early. Treat the standards and frameworks described in this chapter as practical guides that help you build the habits regulators tend to ask for, such as documentation, testing, human oversight, and incident handling. FAIRNESS CHECKS IN PRACTICE To check fairness in a simple way, run your evaluation cases in pairs that differ in only one detail that should not matter, such as a name or an address, and compare the results. If the agent treats the pair differently, you have found a problem to investigate. Keep a record of what was tested and what was found. No test proves fairness completely, but the habit of looking makes unfair behavior far less likely to hide.