Objective alignment focuses on whether a system interprets and pursues a stated objective as intended. It considers the target, the measurements used to reward progress, and whether the system selects actions that genuinely serve that target.
A system can be aligned with a narrow objective without being aligned with broader human values. If the objective omits important constraints, strong optimization can produce a technically successful result that people still consider unacceptable.
ELI5
Objective alignment asks whether an AI helper is aiming at the exact target it was given. A target can be clear and measurable while still leaving out important rules.
For example, a delivery helper told only to be fast might choose a risky route. Adding rules about safety changes what counts as successfully meeting the objective.
