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

Aishik Sanyal

Paper Website | Paper PDF | ArXiv Page | Colab Demo (no setup required)

Welcome to the ReCoN-Ipsundrum codebase. This repository now supports reproduction for two related papers built on the same inspectable recurrent-control framework:

• the original AAAI 2026 submission in docs/recips_aaai2026/
• the ALIFE 2026 social extension in docs/recips_social_alife2026/

The repository contains the core agent implementation, assay definitions, experiment generation and analysis scripts, paper sources, and the static paper website. The structure of the original paper was inspired by Wolfram's Computational Essay.

• Setup
• Paper-Specific Reproduction
• Run the full experiment suite
• Run tests
• Repository layout
• Build the paper PDFs
• Build the paper website
• Abstract
• BibTeX

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

This repository contains two manuscript tracks with different reproduction entry points.

The original paper source lives in docs/recips_aaai2026/paper-3-v9.tex. The main reproduction path is still the repository-wide runner:

./run_experiments.sh

That pipeline produces the baseline artifacts used by the AAAI paper under results/, including:

• results/goal-directed/
• results/qualiaphilia/
• results/exploratory-play/
• results/familiarity/
• results/pain-tail/
• results/lesion/
• results/paper/

If you only want a faster smoke run, use:

PROFILE=quick ./run_experiments.sh

The social-extension manuscript source lives in docs/recips_social_alife2026/main.tex. Its headline claim is narrower and should be reproduced from the dedicated social runners rather than inferred from the non-social AAAI pipeline.

The ALIFE paper evaluates four matched conditions:

• social_none
• social_cognitive_direct
• social_affective_direct
• social_full_direct

across two tasks:

• FoodShareToy
• SocialCorridorWorld

plus lesions (sham, coupling_off, shuffle_partner) and a coupling sweep over lambda_affective and metabolic load.

Important: the module entry points python3 -m experiments.social_foodshare, python3 -m experiments.social_corridor, and python3 -m experiments.social_lesion_assay default to the repository's quick profile in their __main__ blocks. For paper-grade reproduction of the ALIFE manuscript, call the experiment functions explicitly with profile="paper":

python3 -c 'from experiments.social_foodshare import run_foodshare_experiment; run_foodshare_experiment(profile="paper", outdir="results/social-foodshare-paper")'
python3 -c 'from experiments.social_corridor import run_corridor_experiment; run_corridor_experiment(profile="paper", outdir="results/social-corridor-paper", metabolic_load="low")'
python3 -c 'from experiments.social_lesion_assay import run_social_lesion_assay; run_social_lesion_assay(profile="paper", outdir="results/social-lesions-paper")'
python3 -c 'from analysis.social_claims import build_social_artifacts; build_social_artifacts(foodshare_summary_csv="results/social-foodshare-paper/summary.csv", corridor_summary_csv="results/social-corridor-paper/summary.csv", lesion_summary_csv="results/social-lesions-paper/summary.csv", coupling_sweep_csv="results/social-lesions-paper/coupling_sweep.csv", outdir="results/social-paper-paper")'

Those commands generate the manuscript-level social artifacts:

• results/social-foodshare-paper/episodes.csv, summary.csv, exact_threshold.csv
• results/social-corridor-paper/episodes.csv, summary.csv
• results/social-lesions-paper/summary.csv, coupling_sweep.csv
• results/social-paper-paper/headline_summary.csv
• results/social-paper-paper/lesion_summary.csv
• results/social-paper-paper/coupling_sweep.csv
• results/social-paper-paper/claims.json

The corresponding manuscript figures can then be regenerated with:

python3 -m experiments.viz_utils.social_paper_figures --social-paper-dir results/social-paper-paper --out-pdf docs/recips_social_alife2026/figures/fig_summary.pdf --out-png docs/recips_social_alife2026/figures/fig_summary.png
python3 -m experiments.viz_utils.social_visuals_figure --out-pdf docs/recips_social_alife2026/figures/fig_visuals.pdf --out-png docs/recips_social_alife2026/figures/fig_visuals.png

For ALIFE-specific regression checks, run:

python3 -m pytest -q tests/test_social_homeostat.py tests/test_social_forward.py tests/test_social_foodshare.py tests/test_social_lesions.py

run_experiments.sh executes the main repository pipeline and writes artifacts to results/ and logs to logs/. It remains the primary runner for the AAAI 2026 paper and now also includes the social experiment modules as part of the repository-wide sweep.

# Full (paper) profile (more seeds; default)
./run_experiments.sh

# Faster smoke run
PROFILE=quick ./run_experiments.sh

Note: the script deletes and recreates results/ and logs/ at startup.

For the ALIFE paper specifically, use the dedicated commands in Paper-Specific Reproduction, because the social module __main__ entry points default to quick unless you call their functions explicitly with profile="paper".

./run_pytest.sh

• core/: ReCoN primitives and Ipsundrum(+affect) dynamics.
• experiments/: Assays and figure generation (used by run_experiments.sh).
• analysis/: “Claims-as-code” exports used by the paper and tables.
• docs/recips_aaai2026/: AAAI 2026 manuscript source and compiled paper.
• docs/recips_social_alife2026/: ALIFE 2026 social-extension manuscript, figures, and supplementary GIFs.

Requires a LaTeX toolchain (e.g. latexmk).

For the AAAI 2026 paper:

cd docs/recips_aaai2026
latexmk -pdf paper-3-v9.tex

For the ALIFE 2026 social paper:

cd docs/recips_social_alife2026
latexmk -pdf main.tex

The static paper website lives in paper-site/ and is generated from the current results/ artifacts plus a few exported GIFs.

python3 -m analysis.build_paper_site

That command writes:

• paper-site/static/data/site-data.json
• paper-site/static/media/*

To preview locally:

python3 -m http.server 8000 --directory paper-site

A GitHub Pages workflow is included at .github/workflows/deploy-paper-site.yml and rebuilds/deploys the page from main.

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.

@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}
}