Home › Claude › aifoundations › what_is_ai_intro › The value alignment problem, as AIMA frames it, is
The value alignment problem, as AIMA frames it, is
AMismatch between training and deployment distributions
BWeights drifting after months of fine-tuning
CCost difference between training and inference
DMismatch between the agent's objective and ours
Answer & Solution
Correct answer: D. Mismatch between the agent's objective and ours
Specifying our true preferences as a machine objective is hard. When the objective put into the machine diverges from what we actually want, more-capable systems cause more harm.
Related questions
The rational-agent view subsumes the laws-of-thought view becauseThe 'right thing' for an AI agent to do is defined byWhy is cognitive science conceptually distinct from AI per AIMA?AIMA's 'standard model' across AI, control, OR, statistics and economics isWhy does AIMA argue we must move beyond the standard model of fixed objectives?Limited rationality, per AIMA, refers toThe 'laws of thought' approach in AI traces back toWhy has the rational-agent view prevailed in AI?