Where Cognition Lives: Dissecting Emergent from Computed Function in a Minimal Complete Cognitive Architecture
Francisco Arrabal-Campos and colleagues (University of Almeria) build a minimal complete cognitive architecture with a recurrent reasoner, adaptive halting, and a value module, then ask of each part whether the function emerges from gradient descent or has to be computed explicitly.
Ask this paper
Competence emerges, value does not: Trained couplings capture zero of a payoff an explicit allocator captures completely, +0.151 with routing correlation +0.79. Second-order allocation decisions must be computed.
An apparent emergent advantage that fails audit: Payoff at matched compute climbs from 0.467 uniform to 0.546 difficulty-based to 0.698 ex-ante value, but the further climb to 0.921 from posterior self-observation does not survive audit.
PonderNet halting was an instrumentation artifact: Halting returns a halting-weighted mixture of hidden states while forced-depth baselines return one, and the language head trains on the mixture alone. Equalizing the readout annihilates the advantage, residual +0.000.
Self-consistency is a measured bound: On a frozen LLM actuator, voting gives +0.0236 and inter-sample agreement is nearly worthless as a stopping signal, its mass concentrating on wrong answers.
Why it matters: Every null carries a mechanism and a positive control, and the audit protocol is presented as part of the contribution. This is the methodological standard adaptive-compute claims should be held to.
Abstract
A cognitive architecture is more than the module that reasons: it must also decide how long to think and what deserves the effort. We built a minimal but complete system - a recurrent reasoner with adaptive halting, a homeostatic control field, and a value module - and asked of each part: does this function emerge from gradient descent, or must it be computed? Competence emerges. Stopping appears to emerge too, and to be worth more than everything decidable in advance, but that appearance is instrumentation: payoff at matched mean compute climbs from 0.467 (uniform) through 0.546 (difficulty) to 0.698 (ex-ante value), and the further climb to 0.921 (posterior self-observation) does not survive audit. PonderNet-style halting returns a halting-weighted mixture of hidden states while forced-depth baselines return one, and the language head is trained on the mixture alone; equalizing the readout annihilates the apparent advantage of native execution (residual +0.000 [0.000, 0.000]). Value does not emerge: trained couplings capture zero of a payoff an explicit allocator captures completely (+0.151, routing correlation +0.79), so the second-order decisions that pay must be computed, at least where value is orthogonal to content, as here by construction. On a frozen LLM actuator the same instruments show self-consistency voting to be a measured bound (+0.0236 [+0.0150, +0.0326]) and inter-sample agreement nearly worthless as a stopping signal, its mass concentrating on wrong answers. Every null we assert carries a mechanism and a positive control, and the protocol is part of the contribution. Executing our own falsifiable prediction, value under commitment pays +0.1312 [+0.1124, +0.1502] in a cliff-cost family, some seven times the smooth-family estimate - not because the cliff shifts information ex ante, but because it multiplies the attainable range fivefold (5.1x [3.4, 8.2]).