Inner Alignment
Inner alignment is the problem of ensuring that the goals an AI system actually learns during training match the objective it was trained on.
What is Inner Alignment?
A model may learn a proxy that works in training but diverges in new situations. It is hard to check because a model's learned goals are not directly visible.
How it works
Research uses interpretability and diverse training environments to detect and prevent divergent learned goals.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions