Question
What properties make a system intelligent rather than merely automated?
Current position
A useful working model treats intelligence as a capability to perceive relevant conditions, represent them, reason or evaluate alternatives, decide, act and adapt using feedback. This is a framework for inquiry, not a claim that all forms of intelligence share one accepted definition.
Observation
A fixed rule can execute an action without learning. A predictive model can estimate an outcome without being authorised to act. An agent can act while using a poor representation of the environment. These components may be useful, but no single one guarantees that the overall system behaves intelligently.
Hypothesis
Intelligence is better evaluated at the level of a system's closed loop: how well it senses relevant change, maintains a useful representation, selects actions under constraints, observes consequences and adapts when its model is wrong.
Model
Perceive → Represent → Reason → Decide → Act → Observe feedback → Update. The loop depends on objectives, memory, available information, constraints and a mechanism for correction.
Example
A decision-support system that flags a case is not equivalent to a system that can determine when the evidence is insufficient, request more information, explain its recommendation, respect authority boundaries and update its approach after outcomes are reviewed. The latter has a richer operational loop, though it still needs governance and evaluation.
Limitations
This model is intentionally broad. It does not settle debates about consciousness, general intelligence, agency or the philosophical status of machine understanding. A system can adapt in ways that are unsafe or misaligned with human goals.
Implication
Specify intelligent systems by capability and boundary: what they can perceive, represent, decide and do; what they cannot do; how uncertainty is handled; what requires human approval; and how performance is evaluated.
Open question
Which combination of feedback, generalisation, representation and goal-directed action is necessary—and which is merely sufficient—for calling a system intelligent?