The Evidence Layer for Scientific AI

 BGPT Odds of True

The directly relevant Izhikevich DMN SNN simulation (Lahoz-Beltra, 2026) shows φ ≈ 0 when pacemakers fire independently (asynchrony, tonic drive present) and φ ≈ 12.5–15.0 bits only during synchronized bursts — initial evidence that population synchrony, not tonic pacemaker drive alone, generates φ. Izhikevich-PING emergence measures also track synchrony-generating connectivity. However, the anchor study is a single unreplicated preprint run using a covariance-determinant φ proxy (not full IIT 4.0), and the decisive ablations (conditions B and C) have not been run; estimator-dependence risks remain. Evidence tilts toward synchrony-necessity but is not conclusive.

 Hypothesis Novelty

Dissociating pacemaker drive from population synchrony as alternative generators of φ in spiking-network models is a fresh, mechanistically specific framing, though ablation logic itself is standard.

 Quick Analysis Plan Copied

 Long Analysis Plan

Decomposing φ: drive vs. synchrony

Integrated information in a spiking DMN model can arise from three non-exclusive sources: (1) tonic pacemaker depolarization (e.g., intrinsic I_tonic to CA3-like or thalamic nodes), (2) recurrent coupling that generates population synchrony, and (3) genuine irreducible causal structure. φ as defined in IIT 4.0 IIT 4.0 formalism (Albantakis et al., PLOS Comp Biol) and companion paper IIT 4.0 graduated integration of repertoires is a measure of the irreducibility of a system's cause-effect power — it does not intrinsically distinguish pacemaker drive from synchrony as generators, so ablation must be designed to make the distinction experimentally decisive.

Existing empirical anchor: the one directly relevant in silico study

A 10,000-neuron Izhikevich SNN (5,000 regular-spiking sensory + 5,000 intrinsically bursting DMN neurons) with core DMN pacemakers (i=5,000–5,999) receiving tonic 7.0 pA current reported φ ≈ 0 during quiescent windows where sensory neurons were silent and DMN pacemakers fired independently (global covariance matrix degenerating to block-diagonal), while peak φ ≈ 12.5–15.0 bits occurred during synchronized sensory-DMN bursts (t ≈ 530–590 ms and 820–880 ms) Izhikevich DMN SNN φ simulation with pacemaker drive. Crucially, the authors' own falsification criterion is the pacemaker-ablation test formalized below — they did not run it — and they acknowledge single-run, no-sensitivity-analysis limitations. This covariance-determinant φ is a proxy (not full IIT 4.0 perdurant φ), so effect sizes are estimator-specific.

Recommended 2×2 ablation matrix

• A — Full model: I_tonic on, recurrent coupling on. Baseline φ.
• B — Pacemaker ablated: set I_tonic = 0; if network self-sustains and φ persists (≥~50–70% of baseline), synchrony suffices.
• C — Pacemaker kept, synchrony disrupted: desynchronize via spike-time jitter (e.g., ±10–20 ms), shuffling of connection delays, or reduced synaptic weights — while preserving firing rates via compensatory homeostatic I_tonic adjustment. If φ collapses despite similar rates, φ requires synchrony not mere pacemaker input. Note that synchrony and rate are partially dissociable network states: in a 50,000-neuron LIF cortical model, asynchronous-irregular background states supported propagation of both synchronous volleys and firing rates, while increased background synchrony suppressed both — so jitter/delay-shuffling must be verified to leave the network in its baseline dynamical regime rather than a qualitatively different state Synchrony vs rate regimes in large cortical network model.
• D — Neither: φ should approach floor; quantifies measurement artifact.

Statistical rigor requirements and outcome-interpretation table

Because the anchor study ran only a single 900 ms simulation with no replication, any ablation test must specify: ≥10 independent runs per condition (different noise seeds), repeated-measures comparison across burst windows, and effect sizes (Cohen's d with 95% CI for φ_B/φ_A and φ_C/φ_A ratios) rather than p-values alone. Interpretation matrix:

Outcome 4 is the critical hedge: if φ persists under both ablations, the most likely explanation is that the covariance-determinant estimator is picking up rate-driven variance (a known risk of covariance-based proxies), not genuine irreducibility — which is exactly why the PCI/perturbation readout in the estimator-dependence protocol below is mandatory.

Critical confound controls

The main pitfall is firing-rate confounding: removing I_tonic lowers rates, and φ can covary with rate rather than synchrony. Counter this with (i) rate-matched B conditions via external Poisson stimulation, (ii) report both φ and rate-matched nulls, and (iii) verify Izhikevich regime (regular-spiking vs. intrinsically bursting) since intrinsic dynamics interact with pacemaker removal Izhikevich excitability regimes. A known blind spot: synchronization itself may be an effect, not cause, of pacemaker drive — cells sharing a drive become phase-locked trivially, so condition C is the truly decisive ablation, and it is the one most designs omit. Related evidence on emergence from synchrony vs. shared drive comes from emergence measures (Ψ, ∆) computed on Izhikevich-PING networks: emergence of population-level activity increased with cross-community connectivity and inter-neuronal time-delay range (D_max) and appeared at specific parameter regimes, while being absent in surrogate-corrected controls Emergence measures depend on synchrony-generating connectivity in Izhikevich PING — consistent with the prediction that recurrent coupling topology (not tonic drive alone) is the structural substrate of irreducibility in Izhikevich networks. Also pre-register: if φ in condition B exceeds baseline (paradoxical irreducibility), the pacemaker may actually mask intrinsic integration — a falsification criterion worth stating in advance. An important caution on the synchrony readout: it must be validated as rate-robust. A mutual-information-based synchrony index (CFI MI) applied to an Izhikevich RS-LTS network (1,000 neurons) remained near zero for independent trains across firing-rate ratios (1–10×) and recording durations (30–1,000 s), and decayed from ~0.82–0.63 to ~0 as synaptic coupling strength α was scaled from 1.0 to 0 — supporting its use as a rate-robust synchrony readout for the ablation conditions Rate-robust synchrony metric validated in Izhikevich network. Confidence: moderate — the design logic is sound and one Izhikevich-DMN φ simulation now anchors effect sizes (φ ≈ 12.5–15.0 bits at burst peaks vs ~0 at quiescence), but it is a single unreplicated preprint run with a covariance-determinant φ estimator.

Uncertainty and missing evidence

What would change this conclusion: a published demonstration that φ collapses under pacemaker ablation even with rate-matched Poisson rescue (favoring pacemaker-necessity), or the reverse. The largest unknown is the choice of φ estimator (perdurant systems vs. PCMI); results may be estimator-dependent. Recommended estimator-dependence protocol: run each 2×2 condition under at least two φ estimators — (1) full IIT 4.0 perdurant-system φ (or the covariance-determinant proxy used in the DMN SNN study) and (2) an independent information-theoretic readout such as PCI or a perturbation-based complexity measure. This matters because integrated-information-style measures are known to exhibit estimator-dependent instabilities: PCI itself was introduced precisely because exact φ is computationally intractable on real brain data and state-dependent, relying instead on TMS-evoked spatiotemporal complexity via Lempel-Ziv compression PCI as tractable perturbation-based complexity alternative, and prior analyses noted normalization instabilities and non-invariance across integrated-information-style measures such as causal density Estimator instability of integrated information measures. An in silico PCI equivalent for condition C: deliver a brief high-current perturbation pulse to the desynchronized network and apply Lempel-Ziv compression to the evoked spike raster; if PCI-style complexity tracks φ across the 2×2, conclusions are estimator-robust; if they diverge, the effect is likely an artifact of the covariance-determinant estimator.

Bottom line

Available evidence tilts toward synchrony-necessity: the only Izhikevich DMN SNN study with measured φ found it near zero when tonically driven pacemakers fired independently and high only during synchronized bursts, and independent Izhikevich-PING work shows population-level emergence tracking synchrony-generating connectivity rather than drive alone. But the hypothesis cannot yet be adjudicated: the decisive ablations (B and C) have never been run, the anchor study is a single unreplicated preprint with a covariance-determinant φ proxy, and estimator-dependence is a live risk. The 2×2 rate-matched design with ≥10 runs/condition, dual estimators (perdurant φ + in silico PCI), and the pre-registered outcome table above is the minimal decisive test. Generate Full 2×2 Ablation Protocol Visualize φ Phase Diagrams Run AI Scientist Analysis Evolve Hypothesis Analysis

Feedback:   

Updated: September 27, 2026

 Evolve Analysis Plan