SYNTHETIC HUMANITIES LAB / EXPERIMENT 02
How the encounter unfolds.
Different deliveries can leave different temporal traces, even when their average effects resemble one another.
Does earlier sound change reception of an identical passage? ↗
Loading measured responses…
Eight new repetitions under paired background conditions. Each response is compared with silence of the same duration, then with the reference response. Seeds reproduce simulator variability; they are not eight biological flies.
RESPONSE / measured differences
| Population | Temporal difference | Background variability | Agreement between halves | Declared criterion |
|---|
Temporal difference: RMS of the average paired trace. Variability: RMS of its across-seed standard deviation, both Hz per neuron. Agreement: cosine similarity between the first four and last four repetitions. The criterion requires difference > variability and agreement > 0.5. This rule was fixed before these runs; it is a descriptive check, not a statistical significance test.
Where the responses separate
Direct JON partners; complete 100 ms bins. Line: mean paired difference. Band: observed seed range, not confidence interval. Positive means a larger rate change in the comparison. Elapsed-time alignment compares different passages when performances have different pacing.
| Selected episode | Mean difference | Direction across repetitions |
|---|
The three half-second episodes are selected by strongest absolute mean separation, with at least one second between starts. Their selection is exploratory; these are not independently confirmed events.
FUNCTIONAL IMPLICATION / what does this mean for the modeled fly?
Direct JON partners are cells connected immediately downstream of the injected auditory neurons. A difference here means the disturbance has propagated beyond the cells we directly stimulate. Descending neurons provide a separate, broad measure of activity associated with output toward the body. A rate difference alone does not specify a movement or feeling.
READING / an affective interpretation
The interpreter receives only response measurements and temporal positions. It has no poem text or performer identity. Sound, sensory mapping, neural model, measurement and prose each make a different part of the encounter available.
ONE PAUSE / the next resumption
We moved 180 ms of verified silence from after source line 8 to after line 4. Every speech sample, total silence and duration remain unchanged. Source line 5 is blank; line 6 is the first displaced spoken onset.
For each rendition, compare 300 ms after its own line-6 onset with 300 ms before, subtract matched silence, then subtract the reference contrast. The injection values in these aligned windows are identical; earlier histories and absolute times differ. This is an onset-associated response measure, not a measured causal latency.
AFTER THE SOUND / five seconds of observation
| Time after each ending | Mean downstream difference | Signs + / − / 0 |
|---|
Relative to each recording’s own ending. These are between-condition differences after subtracting silence. A nonzero bin does not by itself establish sustained persistence.
Exploratory named output circuits
DNa02, DNp01, MDN and pIP10 retain their annotated identities. Passing this numerical criterion does not establish steering, escape, backward walking or song. These small populations are exploratory, not the primary test.
| Population | Difference / variability | Half agreement | Criterion |
|---|
Method, translations and downloadable evidence
The same frozen MaleCNS / fly.ai model receives the existing amplitude-based JON input. No antennal mechanics or frequency sensitivity was added. The change is experimental: new paired runs, controlled temporal interventions, declared measurements, and a longer observation after sound.
Distinguishing rearranged input does not by itself demonstrate memory or complex temporal processing: a response that follows the changing drive can also pass this check. The localized pause probes a narrower contrast. We have not calibrated these rates to a living fly’s actions.
Full spike arrays remain in the local archive. Public counts include all named groups at every simulation step and phase totals for annotated cell types. The original recorded listening chamber is a different, earlier experiment.