ReCoN-Ipsundrum | Paper Website

Animated corridor comparison across Recon, Ipsundrum, and Ipsundrum+affect — Corridor rollout across ReCoN, Ipsundrum, and Ipsundrum+affect over the full evaluation window.

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 Ns 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 (Delta scenic-entry = 0.07). Affect coupling is stable (Delta scenic-entry = 0.01) even when scenic is less novel (median Delta 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 to persistence and affect-coupled control to preference stability, scanning, and lingering caution, illustrating how indicator-like signatures can be engineered and why mechanistic and causal evidence should accompany behavioral markers.

Method Overview

We evaluate three fixed-parameter variants: ReCoN (stage B), Ipsundrum (stage D), and Ipsundrum+affect (stage D with the affect/interoceptive proxy). The staged diagram below summarizes the Humphrey-inspired extensions added on top of the ReCoN substrate.

Stage A to D recurrent network strip — Humphrey-inspired staged extensions on the ReCoN substrate.

Familiarity Control Separates Novelty From Preference

Two equally safe routes lead to the same goal: a scenic lane with varying sensory features and a dull lane with uniform features. Scenic entry is measured at the earliest committed choice point. ReCoN and Ipsundrum are novelty-sensitive, whereas Ipsundrum+affect remains stable across the novelty manipulation even when scenic is less novel.

Familiarity control summary figure — Familiarity-controlled corridor preference under novelty competition.

Supplementary familiarity view

Baseline and post-familiarization conditions make the novelty dissociation easier to inspect across variants.

Qualiaphilia control

Scenic-over-dull route choice without the familiarity manipulation provides the simpler control setting for this assay family.

Exploratory Play Shows Structured Local Investigation

In reward-free exploratory play, coverage remains broad for all variants, but Ipsundrum+affect shows more structured local investigation: higher scan-event rate, stronger limit-cycle structure, action entropy below a random baseline, and short dwell tails.

Exploratory play summary figure — Exploratory play: structured investigation rather than random dithering.

Seed-level trajectories

Time-ordered trajectories show where scan events and revisitation structure occur during play.

Occupancy heatmaps

Occupancy heatmaps show how coverage, looping, and hazard avoidance interact spatially.

Persistence Is Tested Both Behaviorally and Causally

The pain-tail assay forces one hazard contact, removes the hazard, returns the agent to safety, and records planned actions over the post-stimulus window. The lesion assay disables ipsundrum feedback+integration at t=3 and measures the resulting AUC drop over a fixed post window.

Animated lesion trace comparison — Animated sham-vs-lesion traces across the full post-stimulus window.

Pain-tail assay

Ipsundrum variants show non-zero post-stimulus Ns AUC above baseline, but only Ipsundrum+affect shows prolonged planned caution.

Lesion summary

Lesioning feedback+integration reduces post-stimulus persistence in ipsundrum variants while leaving ReCoN unchanged.

Goal-Directed Performance Acts as a Control

CorridorWorld and GridWorld are competence and safety checks rather than consciousness indicators. Ipsundrum+affect reduces hazard contacts in both environments; in GridWorld this also improves success and time-to-goal, while in CorridorWorld it remains hazard-free but slower. The planning-horizon sweep is shown in Additional Analysis below.

Animated gridworld comparison — GridWorld rollout across the three variants over the full 200-step window.

Additional Analysis

These panels provide direct access to the generated CSV outputs and the control analyses that support the main paper figures.

Exploratory play

X Y

Familiarity progression

Metric Condition

Goal-directed sweeps

Task Metric

Mean lesion traces

Model

Supplementary Figures

Reproducibility

Run metadata

Links

Paper PDF Repository Colab Demo

All figures and charts on this page are generated from the current local `results/` artifacts.

BibTeX

@misc{sanyal2026reconipsundrum,
  title         = {ReCoN-Ipsundrum: An Inspectable Recurrent Persistence Loop Agent with Affect-Coupled Control and Mechanism-Linked Consciousness Indicator Assays},
  author        = {Aishik Sanyal},
  year          = {2026},
  eprint        = {2602.23232},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI},
  doi           = {10.48550/arXiv.2602.23232},
  url           = {https://arxiv.org/abs/2602.23232},
  note          = {Accepted at AAAI 2026 Spring Symposium - Machine Consciousness: Integrating Theory, Technology, and Philosophy}
}