神枢
Project Shenshu — The Divine Pivot
A brain-inspired research route toward building AI agents with genuine memory, personality, and self-awareness — translating the vision of Artificial Fluctlights into real neuroscience-grounded architectures.
“A Fluctlight is not simply data. It is a quantum field that encompasses memories, emotions, and will — in other words, a soul.”
Research Vision
From the fiction of Project Alicization to real brain-inspired AI architectures
Project Alicization
Creating true Bottom-up AI by simulating human souls (Fluctlights) in a virtual world, allowing artificial consciousness to grow organically.
BMAM Framework
Building AI agents from brain-region primitives — hippocampus, prefrontal cortex, temporal lobe — with real memory consolidation, forgetting, and personality.
In Sword Art Online: Alicization, Project Alicization pursues the creation of Artificial Fluctlights — bottom-up AI born not from hand-crafted rules, but by replicating the quantum structures of human consciousness and letting them grow in a simulated world called the Underworld.
Project Shenshu (神枢) takes this fiction as its north star, grounding it in real neuroscience and LLM research. The BMAM framework is the first step: modeling actual brain regions to give AI agents the capacity for memory consolidation, emotional stress response, personality-driven reasoning, and even forgetting — the cognitive primitives that make a Fluctlight more than data.
From Alicization to Shenshu
The conceptual bridge between SAO's fictional technology and our real research architecture
Fluctlight
摇光 / Quantum Soul
The quantum field comprising human consciousness — memories, emotions, and will encoded in photon spin states within neural microtubules.
BMAM Brain Agents
神枢 Core
Brain-region agents (hippocampus, prefrontal cortex, temporal lobe) that model memory, personality, reasoning, and emotional response as cognitive primitives.
Soul Translator (STL)
灵魂翻译机
The device that reads Fluctlight quantum states and translates them into a form readable by computers, bridging biological consciousness and digital systems.
Neuroscience-to-LLM Translation
神枢 Protocol
The BMAM methodology: translating neuroscience principles (memory consolidation, synaptic pruning, stress response) into computational architectures that run on LLM agents.
Underworld
异界 / Virtual Civilization
A simulated world where artificial Fluctlights are raised from birth, developing genuine consciousness through organic experience rather than programmed rules.
Agent Simulation Environment
神枢 World
Environments where BMAM-powered agents develop memories, form personalities, and learn through interaction — evaluated on benchmarks like GAIA and Sotopia scenarios.
Bottom-up AI (A.L.I.C.E.)
人工超级智能
Alicization's goal: creating Artificial Labile Intelligent Cybernated Existence — true AI that emerges from cognitive growth, not top-down engineering.
Brain-Region Primitive AI
神枢 Goal
Building AI from the bottom up using brain-region agents as primitives. Complex cognition emerges from the interplay of memory, reasoning, and personality modules — not from a monolithic model.
BMAM: The Core of Shenshu
The published framework (Findings of ACL 2026, arXiv:2601.20465) that implements the first layer of the Shenshu vision
Cognitive Operations — What Makes a “Fluctlight”
Memory Consolidation
Like Fluctlights forming lasting memories from experience, BMAM transfers short-term observations into stable long-term representations through a hippocampal replay mechanism.
Forgetting
Underworld inhabitants forget — and so do BMAM agents. Controlled forgetting via synaptic decay prevents information overload and prioritizes salient memories, mirroring biological pruning.
Memory Retrieval
Context-driven recall that mirrors how Fluctlights access relevant memories: not by exhaustive search, but by associative cues that activate related memory clusters.
Reflection
Alice's self-awareness emerged through reflection on her experiences. BMAM agents perform meta-cognitive reflection, re-evaluating past actions and updating self-models.
Stress Response
Emotional pressure changes how Fluctlights process information. BMAM models stress-induced cognitive changes: narrowed attention, enhanced threat memory, and altered decision-making.
Personality-Driven Reasoning
Every Fluctlight has a unique personality. BMAM's prefrontal agent validates reasoning through personality constraints, ensuring agents respond consistently with their character traits.
Published Paper
Y. Li, J. Liu, Y. Wang, Y. Wu, and M. Xu. “BMAM: Brain-inspired Multi-Agent Memory Framework.” arXiv:2601.20465.
Shenshu Roadmap
The long-term research trajectory toward brain-inspired AGI agents
Fluctlight Core — BMAM
Build the foundational cognitive engine: brain-region agents with memory consolidation, forgetting, retrieval, reflection, stress response, and personality-driven reasoning.
Soul Translator — Domain Grounding
Ground the cognitive engine in real-world domains. Pain LLM (20+ H100 GPUs) applies brain-inspired architecture to medical pain analysis under the China Brain Project. CondAmbigQA (EMNLP 2025) validates reasoning under ambiguity.
Underworld — Agent Civilization
Multi-agent societies where BMAM-powered agents interact, form relationships, develop shared memories, and evolve collective intelligence in persistent simulated environments.
A.L.I.C.E. — Emergent Cognition
The long-term goal: AI agents whose complex cognition, self-awareness, and adaptability emerge from the interplay of brain-inspired primitives — not from monolithic training. A step toward the Artificial Fluctlight.
Inspiration Gallery
The fictional world that inspired a real research direction
Images from Sword Art Online: Alicization © Reki Kawahara / KADOKAWA / SAO-A Project
Join the Shenshu Journey
Interested in brain-inspired AI, multi-agent memory, or bringing fictional AI visions to reality? Let's collaborate.