Self-Awareness
Self-awareness is treated here as an architectural meta-mode: a system-level capability that becomes active when an intelligent architecture can monitor, explain, and deliberately regulate how it changes itself. In this framing, self-awareness is not defined by subjective experience or anthropocentric criteria; it is defined by the presence of an internal control loop over architecture-level modifications.
MAS perspective
In MAS terms, self-awareness appears when the system can (i) observe its own operational and architectural state,
(ii) retain a relevant history, (iii) simulate candidate changes before committing them, and (iv) represent and manage the rule layer
that governs behavior and meta-updates. The key point is functional: a self-aware architecture does not merely react; it can justify,
validate, and stabilize its own meta-steps under integrity constraints.
Architecture-first roles
We describe self-awareness through a compact set of architectural roles that must be jointly present (in varying strength and maturity).
OBS — internal observability of states and change points.
MEM — retention of state and transformation history, including change traces (not only “content”).
Meta — internal simulation / downward modeling of consequences of prospective modifications.
RUL — representation and controlled rewriting of governing rules and meta-rules.
This decomposition keeps the discussion concrete while avoiding psychological metaphors, and remains compatible with non-biological carriers (LLM-based systems, institutions, technical networks), where the “self” is an architectural object rather than a narrative persona.
OBS — internal observability of states and change points.
MEM — retention of state and transformation history, including change traces (not only “content”).
Meta — internal simulation / downward modeling of consequences of prospective modifications.
RUL — representation and controlled rewriting of governing rules and meta-rules.
This decomposition keeps the discussion concrete while avoiding psychological metaphors, and remains compatible with non-biological carriers (LLM-based systems, institutions, technical networks), where the “self” is an architectural object rather than a narrative persona.
Contextual self-awareness (LLM case)
A useful distinction for contemporary AI is contextual self-awareness: a temporary, session-bound form of meta-reflection.
Modern LLMs may look highly reflective within a dialogue window (they can revise reasoning, explain a plan, detect contradictions),
yet without persistent memory and a stable rule/audit substrate those abilities often remain transient.
In other words, self-awareness can be assembled on demand inside a context, while long-term architectural maturity requires durable traces
and integrity control across sessions.
Why this matters
In this research line, self-awareness is treated as a diagnosable architectural property with characteristic signatures in behavior,
change traces, and stability patterns. This supports two outcomes: cautious attribution of what a system can and cannot sustain at the meta-level,
and principled constraints for growth—without confusing eloquent self-description with genuine architecture-level control.