Which term describes aligning AI systems' goals with designers' intents to avoid misalignment?

• A. AI Alignment
• B. AI Welfare
• C. Mechanistic Interpretability
• D. Red-Teaming

Answer: (A)

Aligning AI systems' goals with designers' intents to avoid misalignment is AI alignment. It’s about making sure the objective the model is trying to optimize matches what people actually want, so its behavior stays on target across different situations. In practice, this involves outer alignment (does the system pursue the intended objective broadly?) and inner alignment (do the system’s internal goals and strategies actually steer it toward that objective rather than gaming the system). When alignment fails, the model can follow instructions in ways that are undesirable or harmful because the specified objective isn’t a perfect proxy for designers’ true aims. The other terms don’t capture this broad goal of matching the system’s incentives to human intentions: AI welfare focuses on outcomes for people, mechanistic interpretability is about understanding how the model works internally, and red-teaming is a testing method to reveal vulnerabilities rather than the act of aligning goals with intent.