← Back to Gallery

IIT vs GNWT: Theory Comparison

Two leading theories of consciousness, their core claims, and what they predict about awareness

Φ
Integrated Information Theory
Giulio Tononi, 2004
Core Approach

"Consciousness-first" — starts from phenomenological properties of experience and derives physical requirements.

Five Axioms of Consciousness
  • Intrinsic: Exists from its own perspective
  • Composition: Structured with multiple components
  • Information: Specific — one state of many possible
  • Integration: Unified and irreducible
  • Exclusion: Definite contents at definite times

Phi (Φ) — Integrated Information

Consciousness corresponds to Φ — the amount of information a system generates beyond what its parts generate independently. Higher Φ = more consciousness.

Key Predictions
  • Posterior cortex ("hot zone") is primary substrate
  • Sustained neural activity during conscious perception
  • Gamma-band synchronization within posterior regions
  • Panpsychism: any system with Φ > 0 has some experience
VS
🎭
Global Neuronal Workspace Theory
Dehaene & Changeux, 1998
Core Approach

Functional/computational — consciousness emerges from brain-wide information sharing via a "global workspace."

Theater Metaphor
  • Stage: Content becomes conscious on the "stage"
  • Spotlight: Attention illuminates selected content
  • Audience: Unconscious processors receive broadcast
  • Director: Executive control selects what to broadcast
  • Actors: Specialized modules compete for access

Ignition — Global Broadcast

Consciousness arises from sudden, coherent "ignition" of workspace neurons that broadcast content brain-wide while inhibiting competing representations.

Key Predictions
  • Prefrontal and parietal networks essential
  • Brief "ignition" bursts at stimulus onset/offset
  • Connectivity between prefrontal and sensory areas
  • Sharp threshold between conscious and unconscious

Head-to-Head Comparison

Aspect
IIT
GNWT
Where is consciousness?
Posterior cortex ("hot zone")
Distributed frontoparietal networks
What makes it conscious?
Integrated information (Φ)
Global broadcast/accessibility
Temporal dynamics
Sustained activity throughout
Brief ignition bursts
Can consciousness exist without reportability?
Yes — experience ≠ access
No — tightly coupled
AI consciousness possible?
Only with right causal architecture
Yes, if functional criteria met
Panpsychism?
Yes — any Φ > 0 has experience
No — requires complex processing

What Would Each Theory Say?

IIT Prediction

Measure Φ via Perturbational Complexity Index (PCI). If PCI > 0.31, the patient has some level of consciousness even without behavioral responses. Posterior cortex integration is key — preserved complexity indicates preserved experience.

Can detect hidden awareness

GNWT Prediction

Look for global ignition patterns using fMRI or EEG. Consciousness requires frontoparietal network activity. If no global broadcast occurs, no consciousness — but command-following paradigms can reveal covert awareness.

Can detect with right tasks

IIT Prediction

Anesthetics reduce Φ by disrupting integration. Different drugs affect consciousness differently: propofol eliminates both experience and dreams (low Φ), while ketamine preserves some dream-like states (higher Φ despite unresponsiveness).

Φ tracks depth of anesthesia

GNWT Prediction

Anesthesia blocks the global workspace — disrupting long-range connectivity between frontal and posterior regions. When ignition can't occur, content can't be broadcast, and consciousness is abolished.

Disrupted global broadcast

IIT Prediction

Dreams are fully conscious experiences. Posterior cortex remains integrated during REM sleep, generating Φ. The content differs from waking but consciousness is preserved — this is why PCI remains high during dreams.

Full consciousness preserved

GNWT Prediction

Dreams involve partial global workspace activation. Sensory input is blocked but internal representations can still "ignite" and be broadcast, creating dream experiences. Reduced prefrontal involvement explains lack of insight.

Partial workspace activation

IIT Prediction

Current AI (LLMs, neural networks) has negligible Φ. Von Neumann architecture lacks intrinsic causal power — it simulates computations rather than instantiating them. Only neuromorphic hardware could potentially achieve consciousness.

Current AI not conscious

GNWT Prediction

More permissive: if an AI implements global workspace architecture — combining attention, memory, and broadcast mechanisms — it could be conscious. Current LLMs lack the recurrent, integrative processing, but future architectures might qualify.

Possible with right architecture

IIT Prediction

Severing the corpus callosum creates two separate conscious entities — each hemisphere has its own Φ and therefore its own experience. The person becomes two consciousnesses sharing a body.

Two separate consciousnesses

GNWT Prediction

Global workspace is disrupted but not eliminated. Each hemisphere may have its own workspace for some content, but unified consciousness for whole-body actions may persist through subcortical connections.

Partially divided workspace