Psychological Grounding - Foundational Models

Psychological Grounding: Foundational Models

The development of agentic AI systems with persistent, expressive personalities, systems that convincingly imitate human behavior, requires application of the psychological foundations that drive human behavior, decision-making, and social interaction.

To create AI agents that can navigate complex social environments and exhibit authentic, contextually appropriate responses, we draw upon established psychological theories that capture the multifaceted nature of human psychology. This section briefly examines four complementary psychological models that together provide a comprehensive framework for understanding the cognitive, emotional, and moral dimensions of human behavior. A more detailed treatment of each can be found in the referenced/linked notes.

The four categories of psychological model we are interested in are; cognitive appraisal theory for emotional responses, emotional state modeling for continuous affective representations, personality psychology for structuring individual differences, and evolutionary approaches to moral cognition.

In our work within NeoThymos the psychological models used from each of these categories are:

Research Area	Model
Cognitive Appraisal Theory & Emotions	OCC Model
Affective Emotional State	PAD Model
Personality Psychology	HEXACO Model
Moral Cognition	MAC Theory

While all of the psychological models discussed here serve as a comprehensive scientific grounding for the underlying NeoThymos personality projection architecture, the current design of NeoThymos leans most heavily on the HEXACO personality model which is the basis for the core personality editor in the NeoThymos Forge. This design aligns with the static model of personality that most agentic frameworks support today.
In the future we will need products and services that enable the real time cognitive appraisal of events and actions in order to support a dynamic homeostatic model of emotion, personality and moral reasoning.

OCC - The Ortony, Clore and Collins Model of Emotion

The OCC model of emotion, developed by Andrew Ortony of Northwestern University, Gerald Clore of University of Virginia, and Allan Collins of Northwestern University, is a highly influential cognitive appraisal theory that provides a structured framework for understanding how emotions are generated. The model posits that emotions arise from how individuals cognitively appraise situations in three distinct ways: as events relative to personal goals (producing joy or distress), as actions of agents relative to standards (producing pride or shame), or as objects relative to attitudes (producing love or hate).

The systematic, computational approach detailed in their seminal 1988 work The Cognitive Structure of Emotions made it particularly valuable for bridging psychology and computer science. In 2022, more than 30 years after its initial publication, the authors published a major second edition that refines and updates the OCC model, incorporating recent advances in emotion theory and affective computing, and includes a contributed chapter by Jonathan Gratch and Stacy Marsella describing the model’s profound influence on computational emotion modeling. Cambridge University PressResearchGate.

The PAD Emotional State Model

While the OCC model provides an approach for appraising events and interpreting their emotional content, there is a need to represent the resulting emotional state. To achieve this, we take inspiration from the PAD (Pleasure-Arousal-Dominance) emotional state model, developed by Albert Mehrabian and James A. Russell (1974). Unlike categorical models that treat emotions as distinct islands (e.g., “Joy” vs. “Sadness”), PAD treats three aspect of affect as orthogonal dimensions in a three dimensional vector space.

The PAD model defines emotional experience along these three orthogonal axes:

Pleasure (Valence): The qualitative evaluation of the state, ranging from negative (unpleasant) to positive (pleasant).

Arousal (Activation): The level of physical and mental energy associated with the state, ranging from low (sleepy/calm) to high (excited/alert).

Dominance (Control): The individual’s perceived level of control over the stimulus or situation, ranging from submissive (controlled by the environment) to dominant (in control of the environment).

Dominance is an important dimension. For example, emotions like Fear and Anger are both unpleasant and highly aroused; they are distinguished primarily by Dominance. Fear is characterized by low dominance (feeling threatened or overwhelmed), whereas Anger is characterized by high dominance (feeling an urge to attack or correct).

HEXACO - A Model of Personality

Developed by psychologists Kibeom Lee of the University of Calgary and Michael C. Ashton of Brock University, the HEXACO model of personality is an influential framework that refines and expands upon the traditional Big Five model. Based on extensive lexical studies across multiple languages, the model describes six major dimensions of personality: Honesty-Humility, Emotionality, eXtraversion, Agreeableness, Conscientiousness, and Openness to Experience.

The model’s most significant contribution over similar models (e.g. OCEAN, Big Five) is the refactoring of the five domains used in those models and the addition of the Honesty-Humility domain, which captures traits like sincerity, fairness, greed-avoidance, and modesty, and has proven crucial for predicting ethical behavior, cooperation, and various forms of counterproductive conduct. Each of the six domains contains four specific facets, plus a 25th interstitial facet called Altruism (versus Antagonism). This interstitial facet was added because altruistic characteristics related to compassion and kindness overlap with multiple domains—particularly Honesty-Humility, Agreeableness, and Emotionality—allowing for more precise measurement of altruistic behaviors that play key roles in social relationships.

HEXACO Domains and Facets:

• Honesty-Humility: Sincerity, Fairness, Greed Avoidance, Modesty
• Emotionality: Fearfulness, Anxiety, Dependence, Sentimentality
• eXtraversion: Social Self-Esteem, Social Boldness, Sociability, Liveliness
• Agreeableness: Forgivingness, Gentleness, Flexibility, Patience
• Conscientiousness: Organization, Diligence, Perfectionism, Prudence
• Openness to Experience: Aesthetic Appreciation, Inquisitiveness, Creativity, Unconventionality
• Interstitial Facet: Altruism (versus Antagonism)

The HEXACO model’s robust empirical foundation and superior predictive validity in moral and prosocial domains were formally established in Lee and Ashton’s foundational 2004 paper published in Multivariate Behavioral Research, making it the preferred choice over Big Five for applications requiring nuanced assessment of ethical and cooperative behaviors.

Note: In NeoThymos we use the label Integrity in place of Honesty-Humility and Curiosity in place of Openness to Experience.

MAC - Morality as Cooperation

Developed by anthropologist Dr. Oliver Scott Curry and his colleagues at the University of Oxford’s Institute of Cognitive & Evolutionary Anthropology, the Morality as Cooperation (MAC) theory provides a groundbreaking evolutionary framework for understanding human ethics. The theory posits that morality is not an arbitrary cultural construct, but rather a collection of biological and cultural solutions to the universal problems of cooperation that emerge from social living. Drawing extensively from evolutionary game theory and cooperative game theory, MAC identifies how different types of social dilemmas—from helping relatives (kin selection) to reciprocal exchanges (reciprocal altruism)—have shaped distinct moral intuitions across human societies.

The MAC model specifies seven universal moral domains that correspond to different cooperation problems:

• helping family (kin selection)
• helping your group (mutualism)
• returning favors (reciprocal altruism)
• being brave (competitive altruism)
• deferring to superiors (dominance hierarchy coordination)
• dividing resources fairly (coordination games)
• respecting property (conflict resolution)

Each domain represents an evolutionarily stable solution to specific cooperation challenges identified in game-theoretic models, from the Prisoner’s Dilemma to coordination games. This comprehensive framework was validated through cross-cultural research and formally presented in Curry et al.’s influential 2019 paper, “Is It Good to Cooperate? Testing the Theory of Morality-as-Cooperation in 60 Societies,” published in Current Anthropology.

A Unified Computational Framework

These psychological models—OCC, PAD, HEXACO, and MAC—form a synergistic framework that addresses distinct but interconnected aspects of human psychology.

Today

Behind the scenes in NeoThymos these models are already integrated into a unified software framework. The HEXACO model is used to define an agent’s personality through a detailed vector-based representation. This Core Personality then informs the derivation of a full Personality Profile composed of an emotional base state, the Baseline Disposition and other emotional subsystems (Sensitivity and Temperament) that eventually will be extended to enable homeostatic responsiveness to emotional stimuli. The HEXACO Personality vector also informs a persona’s moral priorities through the MAC model of morality.

Future Work

Today we use the HEXACO model to capture the stable differences in personality traits between individuals, the things that influence how people typically think, feel, and behave across situations. In the future, inspiration from the OCC and PAD models are guiding the design of an affective appraisal system that takes full advantage of the Personality Profile noted above. This will enable responsive, temporally realistic, homeostatic systems with specific, dynamic, emotional and behavioral responses. The MAC model completes this picture by enabling a persona’s moral priorities, what we today refer to as it’s Moral Center, to be coupled with a Catalogue of Moral Standards, enabling the formation of a personal Moral Compass for the agent that guides its ethical reasoning and cooperative behavior during the appraisal process.

This integration will lead to AI agents capable of exhibiting contextually appropriate emotional responses, consistent personality-driven behaviors, and principled moral reasoning - — essential capabilities for sophisticated agentic systems operating in complex social environments..