ReCoN-Ipsundrum: An Inspectable Recurrent Persistence Loop Agent with Affect-Coupled Control and Mechanism-Linked Consciousness Indicator Assays

Aishik Sanyal

cs.AIFeb 26, 2026

ReCoN-Ipsundrum: An Inspectable Recurrent Persistence Loop Agent with Affect-Coupled Control and Mechanism-Linked Consciousness Indicator Assays

Authors: Aishik Sanyal

Organizations: Independent Research Engineer

Abstract

Indicator-based approaches to machine consciousness recommend mechanism-linked evidence triangulated across tasks, supported by architectural inspection and causal intervention. Inspired by Humphrey's ipsundrum hypothesis, we implement ReCoN-Ipsundrum, an inspectable agent that extends a ReCoN state machine with a recurrent persistence loop over sensory salience $N^s$ and an optional affect proxy reporting valence/arousal. Across fixed-parameter ablations (ReCoN, Ipsundrum, Ipsundrum+affect), we operationalize Humphrey's qualiaphilia (preference for sensory experience for its own sake) as a familiarity-controlled scenic-over-dull route choice. We find a novelty dissociation: non-affect variants are novelty-sensitive ( $Δ$ scenic-entry = 0.07). Affect coupling is stable ( $Δ$ scenic-entry = 0.01) even when scenic is less novel (median { $Δ$ novelty $\approx$ -0.43). In reward-free exploratory play, the affect variant shows structured local investigation (scan events 31.4 vs. 0.9; cycle score 7.6). In a pain-tail probe, only the affect variant sustains prolonged planned caution (tail duration 90 vs. 5). Lesioning feedback+integration selectively reduces post-stimulus persistence in ipsundrum variants (AUC drop 27.62, 27.9%) while leaving ReCoN unchanged. These dissociations link recurrence $\rightarrow$ persistence and affect-coupled control $\rightarrow$ preference stability, scanning, and lingering caution, illustrating how indicator-like signatures can be engineered and why mechanistic and causal evidence should accompany behavioral markers.

Figures & tables

Indicator signpost (Butlin et al.)	What we implement (minimal correspondence)	Important gaps / non-claims
RPT-1 (recurrent processing)	A localized recurrent loop sustaining $N^{s}$ after transient stimulus; explicit lesion of feedback+integration.	Not a neural RPT model; no learning; toy domains.
PP-1 (predictive processing)	Short-horizon internal rollout using a one-step forward model; epistemic term based on predicted sensory change.	Not hierarchical predictive coding; no online learning; not full active inference.
Interoception affect (broadly)	A body-budget proxy and valence/arousal readouts; optional modulation of gain/precision by affect.	Not physiological sensing; no rich autonomic loop; abstraction only.

Table 1: We use selected indicator labels from Butlin et al. (2025) only as design signposts . “Correspondence” denotes a minimal computational element, not a realization of the full theory family.

Model	Ipsundrum recurrence	Affect proxy	Gated $P$
ReCoN (baseline; stage B)	×	×	×
Ipsundrum (stage D)	✓	×	✓
Ipsundrum+affect (stage D)	✓	✓	✓

Table 2: Model variants evaluated. The two ipsundrum rows differ only by the affect proxy layer (and its optional coupling to loop gain/precision).

Assay	Setup / manipulation	Primary readouts
Goal-directed navigation	CorridorWorld and GridWorld with hazards; explicit progress term enabled; curiosity off; sweep planning horizon $H$ .	Hazard contacts; steps-to-goal; success rate.
Corridor preference with familiarity control	Two equally safe routes (scenic vs. dull). Pre-exposure manipulates lane familiarity; post episodes add a curiosity bonus from visitation memory.	Scenic-entry rate; novelty sensitivity $\Delta$ scenic-entry; (Ipsundrum+affect) internal valence/arousal probe.
Exploratory play	Reward-free neutral-texture gridworld (200 steps), with curiosity enabled in the headline play condition.	Unique viewpoints; scan events; cycle score; action entropy/dwell.
Pain-tail	Force one hazard contact, then remove the hazard and hold the state fixed while recording planned actions for 200 steps.	Post-stimulus $N^{s}$ AUC above baseline; planned turn-rate tail duration.
Causal lesion	In-episode lesion at $t{=}3$ disabling ipsundrum feedback+integration (vs. sham).	Post-lesion $N^{s}$ AUC and AUC drop (150-step window).

Table 3: Assays used in this paper and their primary readouts.

	CorridorWorld (all horizons)			GridWorld (all horizons)
Model	Hazards	Time	Success	Hazards	Time	Success
ReCoN	3.05[2.00,4.25]	50.00[37.35,63.20]	0.55[0.30,0.75]	14.30[8.90,20.05]	171.80[131.30,211.10]	0.50[0.30,0.70]
Ipsundrum	2.90[1.90,4.05]	49.73[37.12,62.83]	0.58[0.35,0.78]	2.45[1.27,3.85]	129.77[90.63,168.70]	0.72[0.53,0.90]
Ipsundrum+affect	0.00[0.00,0.00]	52.71[47.16,58.40]	0.73[0.61,0.83]	0.26[0.10,0.44]	9.54[4.75,16.23]	0.99[0.97,1.00]

Table 4: Goal-directed performance aggregated across the horizon sweep (per-seed means over horizons; 95% bootstrap CI over

N=20

seeds). Time is steps-to-goal (failures counted at the time limit).

Signature (assay)	Recon	Ipsundrum	Ipsundrum+affect
Persistence in $N^{s}$ (lesion / pain-tail AUC)	$\times$	$\checkmark$	$\checkmark$
Valence-stable scenic preference (familiarity control)	$\times$	$\times$	$\checkmark$
Structured local scanning (play: scan events)	$\times$	$\times$	$\checkmark$
Lingering planned caution (pain-tail: tail duration)	$\times$	$\times$	$\checkmark$

Table 5: Empirical dissociation across our three variants. The table summarizes which indicator-like signatures attach to which components in our current results: recurrence supports post-stimulus persistence, while affect coupling is necessary for corridor preference stability, structured scanning, and lingering planned caution.

Figure 1: Familiarity-controlled corridor preference. Scenic-entry rates under novelty competition. Non-affect variants increase scenic seeking when scenic is more novel; the affect variant remains stable across the novelty manipulation.

Figure 2: Exploratory play. The affect variant shows more scan events and stronger limit-cycle structure without high-entropy random dithering.

Figure 3: Pain-tail assay. After hazard contact then removal, ipsundrum variants show non-zero post-stimulus $N^{s}$ AUC above baseline, but only the affect variant shows prolonged planned caution (turn-rate tail duration).

Figure 4: Causal lesion. Lesioning feedback+integration reduces post-stimulus persistence ( $N^{s}$ AUC) in ipsundrum variants.

Machine Consciousness

Machine consciousness studies whether artificial systems could have subjective experience and how that possibility might be assessed.

See all 20 papers

Sep 3, 2026

Inferring Affective Consciousness in an Artificial Agent: A Case Study

Creatures that display 'hedonic place preference behaviour' are thought by many scientists to experience feelings, on the assumption that their attraction to pleasure-producing substances which lack nutritional value (e.g. cocaine, morphine) cannot easily be attributed to unconscious instinctual behaviour. In this paper, we discuss how a simple artificial agent that instantiates attributes of an affective system engaging in felt uncertainty about its intrinsic needs in relation to environmental resources can similarly display hedonic place preference behaviour -- through an apparently subjective form of information processing -- while simultaneously being entirely deter-ministic. We outline some implications of this artificially engineered behaviour for our understanding of the physical basis of consciousness and the experience of free will.

Mark Solms, St John Grimbly, Bruce Bassett +7

Neuroscience Institute, University of Cape Town, South Africa. · Dept. of Psychology, University of Cape Town, South Africa. · Dept. of Mathematics & Applied Mathematics, University of Cape Town, South Africa. +6

Machine Consciousness Affective Computing

Ask PDF

Sep 28, 2026

ReplayLens: Auditing Agents' Use of Outcomes

When an agent reuses logged experience, a changed decision may reflect the recorded score, the action's name, or the record's position in storage. Standard memory evaluations do not reveal which relationship drives that change. We introduce ReplayLens, a black-box audit that changes one relationship in the stored history at a time, holds the remaining interface fixed, and measures the resulting decision. Four interventions target four relationships. Outcome reassignment swaps which scores belong to which actions. Pair transport moves intact action-score pairs to new record slots. Consistent renaming relabels actions in both history and menu. Key-slot reassignment changes both score attachment and position. A constructive separation shows why the audit is needed: two memory writers with identical endpoint accuracy respond differently to the same replay, so conventional evaluation cannot resolve the underlying dependence. On black-box LLM interfaces, swapping scores changes decisions while moving intact pairs does not, separating score attachment from record order. A bounded-memory study exposes ingestion-order sensitivity that endpoint comparison misses. In sequential experiment planning, altered historical scores redirect exploration and reduce final utility despite fresh measurements. A code-debugging agent with sealed hidden tests shows the same pattern outside model selection. ReplayLens provides a relationship-level audit for deciding whether logged experience can be merged, reordered, or reindexed safely.

Dong Xu, Zhangfan Yang, Jiantao Wu +5

School of Artificial Intelligence, Shenzhen University · EasternDawn · School of Computer Science, University of Nottingham Ningbo

AI Agent Auditing LLM Agent Memory Experience Replay

Ask PDF

Sep 24, 2026

Probing Stability-Plasticity Tradeoffs in Agent Memory through Cognitive Experimental Paradigms

Agent memory systems are increasingly used to maintain long-term user preferences, task states and evolving facts, but current evaluations often collapse memory behavior into final-answer accuracy. We introduce MemProbe, a cognitive-science-inspired framework for diagnosing stability-plasticity tradeoffs in agent memory. The framework is motivated by a core insight from cognitive memory research: memory is reconstructive and shaped by interference, source reliability, reinforcement, and reactivation. MemProbe turns this insight into four reusable experimental paradigms (interference, misinformation, consolidation strength, and reconsolidation window) that manipulate when a memory should be updated, preserved, or treated as uncertain. It further decomposes correctness into behavioral profiles that reveal how systems update, preserve, attribute, and temporally organize information. We instantiate these paradigms in a 56-episode diagnostic suite and evaluate six incremental memory systems under a unified protocol. Results show that systems with similar aggregate scores exhibit distinct behavioral profiles. MemProbe provides such a diagnostic lens, turning aggregate performance into interpretable profiles of memory maintenance over time. Code is available at https://github.com/jq-ding/MemProbe.

Jiaqi Ding, Guorong Wu

Department of Computer Science UNC-Chapel Hill · Department of Psychiatry UNC-Chapel Hill

Agent Memory Memory Agent Benchmarks

Ask PDF

Abstract

Machine Consciousness

Explore similar work

Inferring Affective Consciousness in an Artificial Agent: A Case Study

ReplayLens: Auditing Agents' Use of Outcomes

Probing Stability-Plasticity Tradeoffs in Agent Memory through Cognitive Experimental Paradigms