Towards a Brain Inspired Model of Self-Awareness for Sociable Agents

Budhitama Subagdja, Ah-Hwee Tan

Abstract

Current artificial intelligence agents often lack a cohesive sense of "self," limiting their ability to engage in truly subjective or socially aware interactions. This paper introduces ARTSELF, a hierarchical neural architecture based on Fusion Adaptive Resonance Theory (Fusion ART). ARTSELF integrates five core dimensions of identity—physical, mental, experiential, and social—to create a dynamic, constructive model of self-awareness. Implemented on a NAO humanoid robot, the model demonstrates successful self-recognition and the ability to resolve ambiguous social queries using autobiographical memory.

Introduction

While modern AI assistants like Siri or Alexa can process complex commands, they remain "applications" rather than "individuals." They lack a persistent identity, an understanding of their physical embodiment, and the capacity to reflect on how they are perceived by others. Existing robotic self-awareness research often focuses on narrow domains, such as motor control or logical reasoning. ARTSELF addresses this gap by providing a holistic framework for both subjective (first-person) and objective (third-person) introspection, moving robots toward becoming sociable agents with a genuine sense of self.

The ARTSELF Architecture

The ARTSELF model is built upon a multi-layered neural architecture designed for multi-channel data fusion. It defines an individual through five distinct dimensions:

  1. Identity: Unique identifiers and personality traits.
  2. Physical: Embodiment and sensory-motor data.
  3. Mental: Internal states, including goals and intentions.
  4. Experiential: Memories of past events used for future projection.
  5. Social: Relationships and adherence to social norms.
The multi-layered ARTSELF architecture integrating five dimensions of self into a hierarchical structure.

The architecture operates through a "bottom-up" categorization process and a "top-down" predictive readout. Raw data from the five dimensions enter at Layer 1 (Input Fields). Layer 2 (Introspection Units) groups these attributes into specific reflections, such as "I am looking at a mirror." Layer 3 organizes these reflections into narratives, while Layer 4 maintains the current "state of self," managing multiple hierarchical "possible selves."

The Working Self Memory system integrated with external modules like the Identity Catalog and Autobiographical Memory.

As shown in the system overview, the Working Self Memory system acts as a central hub, organized and controlled by executive processes, and interacting with specialized repositories like the Social Repository and Physical Model.

Experimental Evaluation

The model was evaluated using a NAO humanoid robot in two primary scenarios:

The Mirror Test

A classic benchmark for self-awareness, the mirror test required the robot to recognize its own reflection. Using the ARTSELF hierarchy, the robot successfully created a representation of itself. By halting the recursive loop of awareness at the appropriate level, the robot could distinguish between "another agent" and "me."

Social Dialog and Memory

In scripted interactions, the robot demonstrated the ability to use its autobiographical memory to resolve ambiguity. When a human provided a vague cue (e.g., "the person with the green shirt"), the robot queried its past experiences to identify the individual. Furthermore, the robot could reflect on its social history, such as recalling if a specific person had been "rude" in previous encounters.

Key Insights

  • Constructive Self: Self-awareness in ARTSELF is not a static database but a dynamic, mentally constructed representation that adapts to context.
  • Pattern Completion: The top-down readout allows the robot to "fill in the blanks" when faced with incomplete information, a critical feature for natural human-robot interaction.
  • Nested Awareness: The model supports "nested" awareness, enabling the robot to represent what a human might think about the robot itself.

Conclusion and Future Work

ARTSELF represents a significant step toward creating robots with a holistic sense of self. While successful in controlled scenarios, future research will focus on incorporating complex social emotions—such as pride or embarrassment—and testing the model in more unpredictable, long-term human environments. Integrating deeper emotional intelligence will be essential for the next generation of truly sociable agents.

Summary updated

Abstract

Self-awareness is a crucial feature for a sociable agent or robot to better interact with humans. In a futuristic scenario, a conversational agent may occasionally be asked for its own opinion or suggestion based on its own thought, feelings, or experiences as if it is an individual with identity, personality, and social life. In moving towards that direction, in this paper, a brain inspired model of self-awareness is presented that allows an agent to learn to attend to different aspects of self as an individual with identity, physical embodiment, mental states, experiences, and reflections on how others may think about oneself. The model is built and realized on a NAO humanoid robotic platform to investigate the role of this capacity of self-awareness on the robot's learning and interactivity.
BibTeX
@inproceedings{subagdja2017towards,
  author = {Budhitama Subagdja and Ah-Hwee Tan},
  title = {{Towards a Brain Inspired Model of Self-Awareness for Sociable Agents}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2017},
  doi = {10.1609/aaai.v31i1.11154},
}
Figure