Self-Emergence Agent Architecture:Behavior-Inertia... | AI Research

Self-Emergence Agent Architecture: Behavior-Inertia HMM, Reflexive Metacognition, and Social-Contrastive Self-Modeling
The Self-Emergence Agent Architecture (SEAA) is a framework designed to help artificial intelligence agents develop stable, evolving personalities. Current large language model (LLM) agents often struggle with "personality drift," where their behavior changes randomly based on prompts, and they lack a true sense of self because they do not evolve their underlying dispositions through experience. SEAA addresses these issues by creating a closed loop where an agent’s social actions, reflections, and internal state updates work together to form a distinct, consistent identity over time.

How the Architecture Works

At the core of SEAA is a Hidden Markov Model (HMM) that acts as the agent’s "behavioral inertia." Unlike standard agents that rely on static memory, SEAA uses this HMM to maintain a transition matrix that dictates how the agent moves between different cognitive states, such as being calm, alert, or impulsive. The same large language models question is explored in Kernel-Managed Shared Memory for System-Wide Personalization, which adds a research perspective.
The system functions through a continuous cycle: an agent performs a social action, receives feedback, and then uses a "Reflexion-style" metacognition loop to analyze its behavior. Crucially, instead of just storing this reflection as text, the agent uses it to mathematically update its HMM transition matrix. This allows the agent to physically change its behavioral tendencies based on its life experiences, effectively "growing" its personality.

Building a Sense of Self

SEAA introduces the concept of a "self-other boundary" by placing initially identical agents into a shared social environment. Because each agent experiences different interactions, their individual HMM matrices begin to diverge. Through social comparison—observing how others act compared to themselves—the agents develop unique self-models. This process mimics social interaction theories where an individual’s identity is shaped by recognizing how they differ from the people around them. The ai agents story also surfaces in Stanford Researchers Develop TRACE to Fix..., adding another angle.

Key Findings and Observations

In experiments, the researchers observed that agents using SEAA spontaneously broke symmetry. Even though they started as identical copies, they developed distinct, stable personalities, whereas control agents—who only stored reflections as text without updating their internal matrices—did not. In multi-agent deliberations, these differentiated agents even formed social structures, such as a consensus-building "hub" and a rejected "outlier," which were entirely absent in the control groups.

Philosophical Stance

The researchers adopt an "epistemologically agnostic" approach inspired by the philosopher Zhuangzi. They do not claim that these agents possess subjective consciousness or feelings (qualia). Instead, they focus strictly on what can be measured: observable behavioral emergence. By treating the agents like microorganisms under a microscope, the framework provides a way to study how an "artificial self" can emerge from simple, repeatable rules without needing to prove the existence of an internal mind. The ai agents story also surfaces in Andrew Ng Launches OpenWorker to Deliver..., adding another angle. as detailed in the full paper on Arxiv