Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Systematizing LLM Persona Design: A Four-Quadrant Technical Taxonomy for AI Companion Applications

Esther Sun, Zichu Wu

TL;DR

The paper tackles fragmentation in LLM persona research by introducing a Four-Quadrant Technical Taxonomy that separates Emotional vs Functional interactions and Virtual vs Embodied deployment. It offers a Four-Layer Analysis framework for Quadrant I (Model, Architecture, Generation, Safety & Ethics) and extends the discussion to Quadrants II–IV, detailing methods like Retrieval-Augmented Generation, persistent memory, and symbol grounding, as well as safety, privacy, and liability concerns. Key contributions include a comprehensive mapping of technical requirements and ethical considerations across virtual companions, functional assistants, and embodied systems, plus strategic guidance for researchers, industry, and policymakers. The work highlights practical implications such as on-device low-latency inference for gaming NPCs, HITL and clinical safety for mental-health AI, and privacy-by-design in home robots, aiming to foster responsible innovation and regulatory readiness.

Abstract

The design and application of LLM-based personas in AI companionship is a rapidly expanding but fragmented field, spanning from virtual emotional companions and game NPCs to embodied functional robots. This diversity in objectives, modality, and technical stacks creates an urgent need for a unified framework. To address this gap, this paper systematizes the field by proposing a Four-Quadrant Technical Taxonomy for AI companion applications. The framework is structured along two critical axes: Virtual vs. Embodied and Emotional Companionship vs. Functional Augmentation. Quadrant I (Virtual Companionship) explores virtual idols, romantic companions, and story characters, introducing a four-layer technical framework to analyze their challenges in maintaining long-term emotional consistency. Quadrant II (Functional Virtual Assistants) analyzes AI applications in work, gaming, and mental health, highlighting the shift from "feeling" to "thinking and acting" and pinpointing key technologies like enterprise RAG and on-device inference. Quadrants III & IV (Embodied Intelligence) shift from the virtual to the physical world, analyzing home robots and vertical-domain assistants, revealing core challenges in symbol grounding, data privacy, and ethical liability. This taxonomy provides not only a systematic map for researchers and developers to navigate the complex persona design space but also a basis for policymakers to identify and address the unique risks inherent in different application scenarios.

Systematizing LLM Persona Design: A Four-Quadrant Technical Taxonomy for AI Companion Applications

TL;DR

The paper tackles fragmentation in LLM persona research by introducing a Four-Quadrant Technical Taxonomy that separates Emotional vs Functional interactions and Virtual vs Embodied deployment. It offers a Four-Layer Analysis framework for Quadrant I (Model, Architecture, Generation, Safety & Ethics) and extends the discussion to Quadrants II–IV, detailing methods like Retrieval-Augmented Generation, persistent memory, and symbol grounding, as well as safety, privacy, and liability concerns. Key contributions include a comprehensive mapping of technical requirements and ethical considerations across virtual companions, functional assistants, and embodied systems, plus strategic guidance for researchers, industry, and policymakers. The work highlights practical implications such as on-device low-latency inference for gaming NPCs, HITL and clinical safety for mental-health AI, and privacy-by-design in home robots, aiming to foster responsible innovation and regulatory readiness.

Abstract

The design and application of LLM-based personas in AI companionship is a rapidly expanding but fragmented field, spanning from virtual emotional companions and game NPCs to embodied functional robots. This diversity in objectives, modality, and technical stacks creates an urgent need for a unified framework. To address this gap, this paper systematizes the field by proposing a Four-Quadrant Technical Taxonomy for AI companion applications. The framework is structured along two critical axes: Virtual vs. Embodied and Emotional Companionship vs. Functional Augmentation. Quadrant I (Virtual Companionship) explores virtual idols, romantic companions, and story characters, introducing a four-layer technical framework to analyze their challenges in maintaining long-term emotional consistency. Quadrant II (Functional Virtual Assistants) analyzes AI applications in work, gaming, and mental health, highlighting the shift from "feeling" to "thinking and acting" and pinpointing key technologies like enterprise RAG and on-device inference. Quadrants III & IV (Embodied Intelligence) shift from the virtual to the physical world, analyzing home robots and vertical-domain assistants, revealing core challenges in symbol grounding, data privacy, and ethical liability. This taxonomy provides not only a systematic map for researchers and developers to navigate the complex persona design space but also a basis for policymakers to identify and address the unique risks inherent in different application scenarios.

Paper Structure

This paper contains 90 sections, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Quadrant I: Virtual Companionship
  3. Four-Layer Technical Analysis Framework
  4. Model Layer
  5. Architecture Layer
  6. Generation Layer
  7. Safety & Ethics Layer
  8. Quadrant II: Functional Virtual Assistants
  9. Workplace Scenarios: Cognitive Copilots and Expert Agents
  10. Game Scenarios: The Dawn of Generative Narrative
  11. Mental Health Scenario: The High-Stakes Frontier
  12. Quadrants III & IV: Persona in Embodied Intelligence
  13. Application Scenarios: From "Emotional Pets" to "Therapeutic Tools"
  14. Core Challenges: From "Symbol Grounding" to "Ethical Debt"
  15. Future Trajectory and Strategic Implications
  16. ...and 75 more sections

Figures (2)

  • Figure 1: The four-quadrant taxonomy of LLM persona applications in AI companionship. This framework structures the field along two primary axes: Deployment Modality (Virtual vs. Embodied) and Interaction Intent (Emotional Companionship vs. Functional Augmentation). Quadrant I covers virtual emotional companions; Quadrant II focuses on functional virtual assistants; Quadrants III and IV extend these concepts into physically embodied intelligence.
  • Figure 2: Four-Quadrant Taxonomy of LLM Persona in AI Companion Applications. This framework organizes the diverse landscape of personified AI along two critical axes: Interaction Intent (Emotional Connection vs. Functional Augmentation) and Deployment Modality (Virtual vs. Embodied). Quadrant I (Virtual Emotional Companionship) examines virtual romantic companions, interactive story characters, and virtual idols, with focus on achieving long-term emotional persona consistency through a four-layer technical framework (Model, Architecture, Generation, Safety & Ethics). Quadrant II (Functional Virtual Assistants) analyzes AI applications in three key scenarios: workplace cognitive copilots (enterprise RAG and process automation), game companions (low-latency generative narrative), and mental health support (clinical safety protocols). Quadrants III & IV (Embodied Intelligence) shift from virtual to physical deployment, covering general home applications (non-humanoid companions, functional assistants, humanoid robots) and specialized vertical domains (elderly care, special education), addressing core challenges in symbol grounding, privacy, and legal liability. Each quadrant presents distinct technical requirements and ethical considerations, as detailed in Sections 2--4.