Antigravity Agent
f8b21af4be
10_Wiki/Topics 대규모 정리: - 오류 캡처/미완성 stub 문서 227개 제거 - 교차폴더 중복 43클러스터 병합 (63파일 → redirect) - 링크명 정규화: 깨진 링크 수정·redirect 직결·개념 매핑 ~2,400건 - 카테고리 MOC 6개 신규 생성 - Graph 섹션 미해결 related-keyword 링크 10,058건 제거 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
140 lines
4.9 KiB
Markdown
140 lines
4.9 KiB
Markdown
---
|
|
id: wiki-2026-0508-global-neuronal-workspace
|
|
title: Global Neuronal Workspace
|
|
category: 10_Wiki/Topics
|
|
status: verified
|
|
canonical_id: self
|
|
aliases: [GNW, GNWT, global-workspace-theory]
|
|
duplicate_of: none
|
|
source_trust_level: A
|
|
confidence_score: 0.85
|
|
verification_status: applied
|
|
tags: [neuroscience, consciousness, cognitive-architecture]
|
|
raw_sources: []
|
|
last_reinforced: 2026-05-10
|
|
github_commit: pending
|
|
tech_stack:
|
|
language: python
|
|
framework: neuro-cognitive
|
|
---
|
|
|
|
# Global Neuronal Workspace
|
|
|
|
## 매 한 줄
|
|
> **"매 conscious access = workspace 로 의 broadcast"**. Dehaene & Changeux 의 GNWT — 매 prefrontal-parietal long-range neuron 의 ignition 시 정보 의 brain-wide 의 broadcast 가 발생, 이것이 conscious experience 의 neural correlate. 매 LLM/AGI architecture 의 inspiration source.
|
|
|
|
## 매 핵심
|
|
|
|
### 매 정의
|
|
- **Workspace neurons**: long-range pyramidal cells (layer 2/3 in PFC, parietal).
|
|
- **Ignition**: 매 sub-threshold processing 의 supra-threshold broadcast 로 의 nonlinear transition (~300ms post-stimulus, P3b component).
|
|
- **Module ↔ workspace**: 매 specialist module 의 결과 의 workspace 로 의 winner-take-all entry.
|
|
|
|
### 매 측거
|
|
- P3b ERP — ignition signature.
|
|
- Long-distance gamma synchrony.
|
|
- fMRI 의 prefrontal-parietal co-activation under conscious report task.
|
|
|
|
### 매 응용
|
|
1. Anesthesia monitoring — workspace breakdown 의 measure.
|
|
2. Vegetative state diagnosis — Owen et al. mental imagery paradigm.
|
|
3. AI architecture — Bengio's Consciousness Prior, Goyal's Coordination via Attention.
|
|
4. LLM analysis — 매 attention 의 workspace 로 의 mapping.
|
|
|
|
## 💻 패턴
|
|
|
|
### Workspace-style Coordination Layer
|
|
```python
|
|
import torch
|
|
import torch.nn as nn
|
|
|
|
class GlobalWorkspace(nn.Module):
|
|
"""매 specialist module 의 의 winner-take-all broadcast."""
|
|
def __init__(self, n_modules: int, dim: int, k_winners: int = 4):
|
|
super().__init__()
|
|
self.attn = nn.MultiheadAttention(dim, num_heads=8, batch_first=True)
|
|
self.k = k_winners
|
|
|
|
def forward(self, module_outputs: torch.Tensor) -> torch.Tensor:
|
|
# module_outputs: (B, n_modules, dim)
|
|
scores = module_outputs.norm(dim=-1) # (B, n_modules)
|
|
topk = scores.topk(self.k, dim=-1).indices
|
|
gather_idx = topk.unsqueeze(-1).expand(-1, -1, module_outputs.size(-1))
|
|
winners = module_outputs.gather(1, gather_idx) # (B, k, dim)
|
|
broadcast, _ = self.attn(winners, winners, winners)
|
|
return broadcast.mean(dim=1)
|
|
```
|
|
|
|
### Ignition Detector (P3b-like)
|
|
```python
|
|
import numpy as np
|
|
|
|
def detect_ignition(eeg: np.ndarray, fs: int = 1000, electrode_pz: int = 31):
|
|
"""Pz 의 250-450ms window 의 amplitude → ignition flag."""
|
|
window = eeg[electrode_pz, int(0.25 * fs):int(0.45 * fs)]
|
|
baseline = eeg[electrode_pz, :int(0.1 * fs)]
|
|
p3b = window.mean() - baseline.mean()
|
|
return p3b > 3 * baseline.std(), p3b
|
|
```
|
|
|
|
### Consciousness Prior (Bengio)
|
|
```python
|
|
class ConsciousnessPrior(nn.Module):
|
|
"""매 sparse high-level state z_t 의 의 dependency 의 sparse factor graph 의 학습."""
|
|
def __init__(self, dim, k_active=5):
|
|
super().__init__()
|
|
self.encoder = nn.Linear(dim, dim)
|
|
self.k = k_active
|
|
|
|
def forward(self, h):
|
|
z = self.encoder(h)
|
|
topk_vals, topk_idx = z.abs().topk(self.k, dim=-1)
|
|
mask = torch.zeros_like(z).scatter_(-1, topk_idx, 1.0)
|
|
return z * mask
|
|
```
|
|
|
|
### Long-range Gamma Coupling
|
|
```python
|
|
from scipy.signal import hilbert
|
|
|
|
def plv(x, y):
|
|
"""Phase-locking value — 매 long-distance gamma synchrony proxy."""
|
|
px = np.angle(hilbert(x))
|
|
py = np.angle(hilbert(y))
|
|
return np.abs(np.exp(1j * (px - py)).mean())
|
|
```
|
|
|
|
## 매 결정 기준
|
|
| 상황 | Approach |
|
|
|---|---|
|
|
| Conscious report task | P3b + gamma coupling |
|
|
| AI coordination | Workspace + sparse top-k |
|
|
| Anesthesia depth | Workspace breakdown index |
|
|
| Disorder of consciousness | Active paradigm (mental imagery) |
|
|
|
|
**기본값**: GNW + IIT 의 complementary — GNW 의 access consciousness, IIT 의 phenomenal.
|
|
|
|
## 🔗 Graph
|
|
- 부모: [[Cognitive-Architecture]]
|
|
- 변형: [[Global-Workspace-Theory]] (Baars) · [[GNWT]] (Dehaene)
|
|
- Adjacent: [[Predictive-Processing]]
|
|
|
|
## 🤖 LLM 활용
|
|
**언제**: cognitive architecture design / consciousness 관련 신경과학 정리 / multi-agent coordination.
|
|
**언제 X**: 매 phenomenal consciousness (qualia) 의 explanation — IIT 의 영역.
|
|
|
|
## ❌ 안티패턴
|
|
- **GNW = consciousness fully**: 매 access vs phenomenal 의 conflate.
|
|
- **Workspace = single bottleneck**: 매 실제로 distributed competition.
|
|
- **PFC = consciousness seat**: 매 posterior hot zone view 의 ignore.
|
|
|
|
## 🧪 검증 / 중복
|
|
- Verified (Dehaene 2014 Consciousness and the Brain, Mashour et al. 2020 Neuron).
|
|
- 신뢰도 A.
|
|
|
|
## 🕓 Changelog
|
|
| 날짜 | 변경 |
|
|
|---|---|
|
|
| 2026-05-08 | Phase 1 |
|
|
| 2026-05-10 | Manual cleanup — GNW + AI architecture inspiration 패턴 |
|