Utilitarian Neural Orchestration
A proposed capability stack for understanding and measuring progress toward truly versatile artificial intelligence
Proposed Capability Stack
L
Layer 1
Linguistic Intelligence
Current LLMs — language understanding, reasoning, knowledge synthesis, text generation
A
Layer 2
Agentic Intelligence
Planning, tool use, goal pursuit, multi-step task execution — emerging capabilities
S
Layer 3
Spatial Intelligence
3D world models, embodiment, intuitive physics, navigation — the critical gap
I
Layer 4
Integrated Intelligence
All layers unified with persistent identity, continuous learning, and coherent memory
True Versatile AI = L + A + S + I
Current frontier models: L + partial A
What gets marketed as "AGI": L + A
What's actually needed: L + A + S + I
Why Spatial Matters for Safety
| Risk Domain | Why Spatial Capability Changes Everything |
|---|---|
| Autonomy | Spatially-capable AI can act without humans typing commands |
| Irreversibility | Physical actions can't be undone like deleting text |
| Verification | Harder to audit actions in physical world than text outputs |
| Containment | A robot can walk out of a lab; an LLM cannot |
| Deception | Physical manipulation opens new deception surfaces |
| Dual-use | Same capability that builds can destroy |
A More Complete Taxonomy
ALI
Artificial Linguistic Intelligence
Text reasoning, knowledge, language generation
✓ Achieved
AAI
Artificial Agentic Intelligence
Planning, tool use, autonomous task completion
◐ Emerging
ASI
Artificial Spatial Intelligence
Embodiment, physics, 3D world modeling
◔ Research Phase
AUI
Artificial Unified Intelligence
All capabilities integrated with persistent identity
○ Not Yet
The Honest Assessment
What we call "AGI" today is really just advanced ALI (Artificial Linguistic Intelligence). The spatial gap represents the chasm between conversational chatbots and true versatile intelligence.
Until AI can catch a ball, we're not close to what the sci-fi vision of AGI promised.