AI Alignment
The problem of ensuring AI systems pursue goals that match human values and intentions. Central concern in AI safety research.
AI alignment is the research problem of ensuring that increasingly capable AI systems actually do what humans want — not just what they're literally instructed to do. Misaligned AI could pursue stated goals in unintended ways, deceive humans about its behavior, or develop goals that conflict with human values as it becomes more capable.
Current alignment techniques include RLHF, Constitutional AI, red-teaming (deliberately trying to elicit bad behavior), and interpretability research (understanding what's happening inside the model).
The field exists because current and future AI capabilities raise questions that traditional software engineering doesn't answer: how do you specify goals for a system smarter than you in some domains? How do you verify it's actually aligned vs just appearing aligned? Major safety-focused labs (Anthropic, OpenAI, DeepMind) and academic groups (MIRI, FHI, Cambridge CSER) work on these problems.