What Appears and What Can Be Explained
As AI systems become more complex, they can exhibit behaviors that go beyond what programmers explicitly specified. This section separates the problem of observing new capabilities from the problem of explaining them.
Capabilities that arise from complexity rather than from an isolated programmed feature.
The effort to explain model behavior by tracing internal structures and representational pathways.
A philosophical concept concerning awareness and subjective experience, not a direct inference from model complexity.
When an AI system reaches a certain level of sophistication, it can display what are called . These are abilities that seem to arise from the interaction of many components rather than from one directly coded feature.
To move beyond observing outputs, researchers pursue . This effort looks inside the model to understand what structures and circuits are producing particular behaviors.
The distinction is critical: are what we observe from the outside, while is the attempt to explain those behaviors from the inside.
It is equally important not to collapse technical questions into philosophical ones. Concepts like belong to a different level of analysis and should not be inferred from complex outputs alone.
