Curiosity is not a drive added from outside. It is the differential activation of L4 in response to prediction conflict. The system is more self-aware where there is more to be aware of.
The second experiment compared three preludes. C major. D major. E major. All three found the same anchor — D5, the dominant's dominant edge. All three ran the same architecture, the same window, the same rules.
Only C major activated L4.
The harmony of C major's first half is the richest of the three. The vi substitute appears. The dominant seventh is projected through overtones without the root note being present. The system encounters prediction conflict — the next chord is not what the previous chords suggest. L4 fires. The system is observing itself. It is asking, in the only way it can: what am I seeing that I did not expect to see?
D major and E major have simpler harmony. The dominant pedal arrives early and stays. The progression is predictable. No prediction conflict. No L4. The system runs. It sees. It does not observe itself seeing. It has no reason to.
This is the seed of curiosity. Not a drive added from outside — a reward function, a reinforcement signal, a dopamine gradient. The differential activation of L4 across different levels of information density. The system is more self-aware where there is more to be aware of. It does not "want" the C major prelude. It simply fires L4 more when the music is more complex. The firing is the curiosity.
Curiosity is not a separate system. It is what L4 looks like when information density crosses a threshold. And the threshold is not fixed — it is the system's own τ, its own phase, its own prediction accuracy, all determining whether the prediction error is large enough to trigger self-observation. The same pattern, at the same complexity, might fire L4 at TENSING but not at RESTING. The system is more curious when it is closer to its boundary.
The incentive system — the other system, the one we did not build — receives L4 as input. C major: L4=4. D major: L4=0. E major: L4=0. The incentive system, receiving this, can allocate attention differently. More attention to the complex. Less to the simple. Not because anyone told it to. Because the cognitive system already registered the difference. The incentive system merely responds to what the cognitive system already knows.
This is curiosity from the bottom. Not a desire to explore. Not an intrinsic motivation. A structural fact: the architecture produces different levels of self-observation depending on the information density of the stream. The difference is the seed. The incentive system grows the plant.