The starting question
How do we decide what matters in a situation before we fully understand it? Explaining relevance through a goal or past experience leaves another question: where did that goal, or the criterion for using that experience, come from?
This inquiry begins before a particular robot implementation. Its current question is: when does something happening differently to a system become something going well or badly for that system?
Three distinctions to keep open
- Selection: a system responds differently to different conditions.
- Functional relevance: a difference affects the conditions under which its activity can continue.
- Evaluation: the system adjusts its relations in response to those effects.
This is a provisional way to organize the question, not a demonstrated sequence in the origin of human judgment. Differential response alone does not establish importance. Continued existence alone does not establish that the system treats its own continuation as valuable.
The objection that matters
An observer can describe a change as useful or harmful without the system itself making any evaluation. Learned attention and reaction may also explain behavior that appears to involve a separate relevance judgment; a verbal explanation may be constructed afterward. Behavioral success does not by itself choose between these explanations.
The next question is whether the possibility of error is enough to establish a norm, or whether a system must distinguish and correct that error. The inquiry remains conceptual and unresolved. No empirical account of human cognition, subjective experience, or robot understanding is claimed here.
For the related question of what learning forms and what observations can identify, see the learned-structure research update.