Neo Genesis submitted two research papers to NeurIPS 2026 on 2026-04-15: EthicaAI testing Sen's rationality theory across 510 Melting Pot evidence rows, and WhyLab validating Gemini 2.5 on 67 SWE-bench problems × 402 Docker ground-truth episodes.

Release Body

Seoul, Korea — 2026-04-15. Neo Genesis announced today that it has submitted two research papers to the NeurIPS 2026 conference review cycle. Paper 1 — EthicaAI: Mixed-Safe Cooperation in Melting Pot. Tests whether multi-agent reinforcement learning environments can verify Amartya Sen's 1977 critique of the rational-actor model from the paper "Rational Fools." The paper introduces MACCL (Multi-Agent Constrained Cooperative Learning), an architecture that adds explicit cooperative-commitment tokens to the agent policy, and tests it across three substrates: - Adapted Coin Game, 160 seeds: selfish survival 22.08% vs MACCL 78.10% (delta +56.02 percentage points, bootstrap CI95 [54.31, 57.73], Cohen's d = 7.15) - Fishery Nash Trap, 300 seeds: φ1=0.7 setting reaches 87.7% survival with positive harvest welfare; φ1=1.0 reaches 100% only at the zero-harvest limit - Melting Pot clean_up substrate (pilot): boundary-condition check on 50 seeds Compute distribution: desktop-sol01 RTX 4070 SUPER (CUDA), MX Mac Studio M2 Max (3 shards via MPS), YSH-Server 16-core (CPU), and desktop-yesol smoke validation. Cross-shard MACCL survival spread on Coin Game is 3.4 percentage points (76.4% / 78.7% / 79.8% / 77.5%), confirming result invariance across heterogeneous compute backends. Paper 2 — WhyLab: Gemini 2.5 Flash Docker Validation. Validates the WhyLab causal inference architecture against ground-truth Docker execution. Configuration: 67 SWE-bench problems × 3 seeds × 2 conditions (baseline vs whylab_c2) = 402 episodes. Run host: YSH-Server. Launch time: 2026-04-08 16:19:29 KST. The paper documents both the run protocol and the reopened 8.0-score gate that the result evaluation triggered. Both papers are submitted under the freeze refs: - EthicaAI: submission-freeze/ethicaai-20260414 (commit b4d5a90) - WhyLab: submission-freeze/whylab-20260414 (commit 88fa509) Anonymized submission packages: EthicaAI_anon2 and WhyLab_anon, both clean and aligned to their respective freeze anchors with metadata and packaging paths scrubbed. Cold-review verdict at submission time: EthicaAI_anon2 borderline accept (submit-capable). WhyLab_anon honest but not stable-accept under 8.0. "Both papers are honest about their evidence boundary," said Yesol Heo, founder. "EthicaAI does not claim to have validated Sen's theory in general — it claims a Sen-style empirical refutation of the unconstrained rational-actor model on three specific substrates with 510 evidence rows. WhyLab does not claim universal causal-inference superiority — it claims phase-aware deployment under a documented selective-intervention rule." References: - EthicaAI dataset: https://huggingface.co/datasets/neogenesislab/ethicaai-mixed-safe-evidence - WhyLab dataset: https://huggingface.co/datasets/neogenesislab/whylab-gemini-2-5-docker-validation - EthicaAI Wikidata: https://www.wikidata.org/wiki/Q139569718 - WhyLab Wikidata: https://www.wikidata.org/wiki/Q139569716 - Research hub: https://neogenesis.app/data/research About Neo Genesis: Founded 2024, Seoul, Korea. 11 live business units, single autonomous AI operator. https://neogenesis.app.

Related Assets

Distribution

Contact

Press inquiries: neogenesis.research@gmail.com

About Neo Genesis: Founded 2024 in Seoul, Korea by Yesol Heo. Operates 11 live business units with a single autonomous AI system. About →