RFC-020: Governed Autonomy, Symbolic Intent, and D-DNA Evidence — 7. Human Language Is Not a Control Plane

AIGP Specification › RFC-020: Governed Autonomy, Symbolic Intent, and D-DNA Evidence › 7. Human Language Is Not a Control Plane

← 6. Policy Is Not the Primary Object · Section index · 8. Architectural Deterrence and Inhibition →

7. Human Language Is Not a Control Plane

Natural language is suitable for:

intent expression
human explanation
legal review
training
doctrine
post-action analysis

Natural language is not sufficient for:

runtime control
autonomous permission
force inhibition
geofence enforcement
sensor uncertainty handling
peer verification
machine-to-machine governance

An autonomous system should not be governed by prose such as:

Avoid danger.
Do not harm civilians.
Use reasonable caution.
Escalate if uncertain.

These statements must be compiled into symbolic, typed, machine-verifiable constraints.

Example:

constraints:
  - if civilian_presence == true then deny_force_action
  - if identity_confidence < roe.identity_threshold then escalate
  - if geofence_status == outside then return_to_base
  - if comms_loss_seconds > 10 then degrade_autonomy

AIGP doctrine:

Autonomous systems may explain themselves in human language, but they must be governed in machine-verifiable language.

← 6. Policy Is Not the Primary Object · Section index · 8. Architectural Deterrence and Inhibition →