Does lifetime learning do anything? Two pre-registered nulls and one real effect
Organisms carry a working reward-modulated plasticity mechanism. Two pre-registered population-level tests found no detectable behavioural effect from it. A separate counterfactual proved it does measurably move an individual's outputs. Both are true.
The question
Does the same organism's behaviour, conditional on what it locally senses, differ measurably earlier versus later in its own life, and does that difference depend on whether lifetime plasticity is switched on?
The hypothesis, written first
Pre-registered verbatim before the runs: if lifetime plasticity produces within-life behavioural adaptation to locally sensed resource, the distribution of per-organism change in the resource-to-ingest correlation under learning differs from the distribution without it. Mann-Whitney U, two-sided, direction not pre-specified, alpha = 0.05.
Setup
- Ruleset
- Phase 4B and Phase 4C rulesets, both superseded. The plasticity mechanism itself is unchanged and is what runs today.
- Seeds
- 6 seeds at the reduced scale (Phase 4B), 8 seeds at the canonical scale (Phase 4C).
- Duration
- 6,000 ticks per run. 18 runs in the first test, 24 in the replication.
What was measured
- Per-organism change in the correlation between locally sensed resource and the ingest action, comparing a 150-tick window early in life against a 150-tick window late in life.
- A second metric added in the replication: reinforcement-conditioned persistence of the ingest action, isolating the act of updating acquired weights from merely computing the reinforcement signal.
- Separately and by a different method: the same organism's action outputs computed with and without its acquired weights, on identical inputs.
Result
No statistically detectable population-level behavioural effect on either pre-registered metric, at either world scale, at sample sizes 340 to 520 times the pre-registered minimum.
A real, small, measurable individual-level effect on network outputs, by a separate non-hypothesis-tested method, correlating at r = 0.90 with each organism's own accumulated acquired-weight magnitude.
Limitations
- The null could reflect a genuinely absent effect, an effect smaller than the design can detect, a metric that does not capture how this architecture's plasticity actually manifests behaviourally, or windows too short for accumulated change to show. These tests cannot distinguish between those.
- One numerically salient pattern is reported descriptively and is explicitly not a tested claim: conditions carrying the plasticity machinery survived the full tick budget in 6 of 8 seeds where the plasticity-off condition survived in 4 of 8. Conditions run from the same world seed are not the same simulation from tick zero, because the extra energy draw perturbs every organism's trajectory immediately, and in a chaotic density-dependent ecology that can cascade in either direction.
- No correction for multiple comparisons across the p-values reported.
- The counterfactual sampled one seed. It establishes that the effect exists and scales with accumulated change, not its magnitude across the space of possible runs.
Sources
Every number on this page comes from one of the following, all of which live in this project's own repository and are append-only records rather than summaries written after the fact.
- The project experiment log: EXP-0008, EXP-0009 (Phase 4B) and EXP-0010 (Phase 4C)
- The project decision record, sections 50 and 51, including 51.9, the counterfactual result
- The observer's brain inspector, and its plasticity counterfactual
Everything needed to check this claim is on this page already: the ruleset, the seeds, the duration, what was measured, the result, every limitation and the sources. Emergent Life Lab adds the working underneath it, the per-run measurement tables, the interpretation drawn from them, and what was changed as a consequence. See what Lab includes →