TL;DR
This paper presents a formal model for integrating the qualitative (triggering conditions) and quantitative (intensity) aspects of human emotions into intelligent agent architectures. By formalizing the Ortony, Clore, and Collins (OCC) model using a dynamic logic-based agent specification language, the authors provide a framework for agents to not only recognize emotional triggers but also to simulate the experience and decay of emotions over time.
What Problem Does This Paper Address?
Building believable and sophisticated artificial agents requires them to exhibit emotional responses. However, existing psychological models of emotion, like the influential OCC model, are often descriptive rather than formal, leaving significant gaps in how to implement them computationally. Specifically, while the OCC model outlines what triggers an emotion (qualitative aspects) and what factors influence its intensity (quantitative aspects), it lacks concrete mechanisms for how these intensities are calculated, experienced, and evolve. This paper aims to bridge this gap by providing a formal, computational framework for integrating both qualitative and quantitative emotional aspects into agent models, making them more robust and realistic.
Methods
The authors leverage a modified version of the KARO (Knowledge, Action, Reasoning, and Other) agent specification language, which is based on dynamic and epistemic logic. This language is extended with modal operators to handle motivational aspects of agents.
Key methodological contributions include:
- Qualitative Formalization: The paper translates the OCC model's emotion-triggering conditions into formal logical propositions within the KARO framework. For instance, emotions related to consequences of events are mapped to goal accomplishments or failures, and emotions related to actions are mapped to plans and their sequential compositions.
- Quantitative Integration: The model incorporates quantitative aspects by defining "emotion triggering fluents" that, when satisfied, associate a potential, threshold, and intensity with an emotion. A novel inverse sigmoid function is proposed to model the decay of emotion intensity over time, allowing for parameters like initial intensity, half-life, and fall-off speed.
- Agent Specification Language (LPAG): A language is defined with operators for belief (
Bi), goals (Gi), accomplishments (Acci), differences (Diff), and action outcomes ([doi(π)]). This language allows for the formal representation of agent states, beliefs, goals, and the effects of actions on these. - Semantics: The paper defines a formal semantics for the LPAG language using Kripke structures, extending them to include functions for goals, emotion triggers, and history tracking.
Datasets
Not discussed in the paper. The paper focuses on formalizing an existing psychological model rather than empirical evaluation on datasets.
Key Insights and Ideas
- Bridging the Gap: The core insight is that a formal logic-based framework can effectively integrate the descriptive nature of psychological emotion models with the computational needs of intelligent agents.
- Formalizing "Experience": The introduction of an "emotional experience fluent" (e.g., "joyous") that is satisfied only when an emotion's intensity is above zero provides a formal way to model an agent experiencing an emotion, distinct from merely having the conditions for its trigger met.
- Dynamic Intensity Decay: The use of an inverse sigmoid function for emotion intensity decay is a pragmatic and intuitive approach to model how emotions naturally fade over time, offering parameters to control this decay rate.
- Explicating Implicit Assumptions: The formalization process reveals and clarifies implicit assumptions within the OCC model, particularly concerning the relationships between opposing emotions (like hope and fear) and their intensities.
Key Results
The paper presents a formalization of the qualitative and quantitative aspects of the OCC model within the LPAG language. While no empirical results are presented, the key outcome is the development of a formal framework that:
- Defines how specific emotions (e.g., gratification) are triggered based on agent actions, goals, and events.
- Proposes a mechanism for calculating and modeling the decay of emotion intensity over time using an inverse sigmoid function.
- Establishes a formal link between emotion triggering conditions and the agent's subjective experience of that emotion.
- Demonstrates how formalization can highlight and address implicit assumptions in psychological models.
Limitations and Future Work
Limitations:
- No Empirical Validation: The paper focuses on formalization and does not present empirical validation of the model with real-world data or simulations.
- Simplified Quantitative Aspects: While a decay function is proposed, the paper acknowledges that the calculation of emotion potentials and thresholds is not the focus and is left as a subject for future work.
- Specific Emotion Focus: The detailed formalization is presented for "gratification," with the understanding that a full formalization of all 22 OCC emotions is available but not fully detailed due to space constraints.
Future Work:
- Full Formalization: Completing the formalization of all 22 OCC emotions.
- Potential and Thresholds: Developing concrete methods for calculating emotion potentials and thresholds, potentially linking them to an agent's global "mood."
- Parameter Dynamics: Investigating how the parameters of the intensity decay function (half-life, fall-off speed) might change over time or in response to events.
- Influence on Deliberation: Exploring how the modeled emotions influence an agent's deliberation and decision-making processes.
- Opposing Emotions: Further investigating the relationships between intensities of opposing emotions beyond hope and fear.
Summary updated