Program

A Methodological Framework for Theory Formalization in Psychology: An Application to Resource Competition in Cognitive Development

Psychological theories often describe complex dynamic systems, yet they typically remain verbal and underspecified, making it difficult to derive precise predictions or explain how psychological phenomena emerge over time. Mechanistic computational models offer a promising way to formalize such theories.

In this talk, I present a methodological framework for theory formalization in psychology based on the theory construction methodology (TCM; Borsboom et al., 2021) and the productive explanation framework (van Dongen et al., 2025). These methodologies provide a structured sequence of steps to guide researchers in transforming verbal theories into computational models.

To illustrate these methodology, I present a formal model of resource competition in cognitive development (de Ron et al., 2023). The model integrates mutualistic interactions between developing cognitive abilities with competition for limited resources, such as time. Through simulation, the model reproduces a range of established developmental phenomena, including the positive manifold, developmental stages, and characteristic profiles of atypical development.

Using this example, I demonstrate how each step of the methodologies can be implemented in practice and discuss the broader applicability of this approach for theory development in psychology.

Computational Frameworks for Theory Building in Psychology

Psychological theories are often incompletely specified—that is, certain cases are overlooked, and some statements are ambiguous or imprecise. This is largely due to the fact that such theories are rarely expressed using formal models. Yet in many scientific disciplines, formal modeling plays a crucial role in theory-building: it enhances the precision of predictions, strengthens scientific reasoning, and facilitates collaboration. We propose two automata-based frameworks and a methodology for modeling psychological theories—that is, for transforming informal, verbal descriptions into formal, analyzable models.

These frameworks use communicating finite automata, along with a refinement procedure for theory-building, as well as model-checking to verify that a model is faithful to the original theory. We prove the pseudo-inclusion of traces of the refinement procedure, as well as a completeness property of our framework.

This method highlights theoretical ambiguities and inconsistencies, and prompts research questions. We illustrate our method using Lazarus and Folkman's theory of stress as well as the consciousness theory called the Global Neuronal Workspace hypothesis. We provide a step-by-step formalization of these theories.

No Psychology Without Viewpoint: The Projective Consciousness Model as a Formal Architecture for Computational Psychology

Computational psychology cannot be satisfied with translating verbal theories into code, nor with adding statistical machinery to behavioral data. If it is to become a theoretical framework, it must formalize the hidden structures by which psychological processes generate observable trajectories. I will argue that one such structure is indispensable: the viewpoint. Perception, imagination, affect, learning, social inference, and action are not organized from nowhere. They are organized from an embodied first-person perspective.

The Projective Consciousness Model turns this fact into an explicit computational architecture. Its central claim is that conscious access is structured as a three-dimensional projective field, acted on by the projective group and governed by active-inferential and optimal-control principles. This is not a phenomenological ornament. From minimal constraints on sensorimotor projection, coherent updating under changes of viewpoint, invariance, and control, the projective format emerges as the uniquely economical representational structure for coherent egocentric experience. Euclidean and purely topological alternatives either reintroduce the same structure implicitly or fail to preserve the phenomenon.

I will show how this formal core yields a unified model of psychological computation. First, it explains perceptual regularities and anomalies, including the Moon illusion, as consequences of projective frame calibration. Second, it links affective and epistemic value to policy selection, showing why projective geometry changes exploration whereas Euclidean control does not. Third, it extends to social cognition: virtual agents equipped with PCM can simulate theory-of-mind orders, infer preferences, and generate approach–avoidance strategies in immersive environments. Fourth, when coupled to large language models, PCM supplies the missing inspectable affective-control loop: felt appraisal, voluntary expression, involuntary leakage, belief management, and linguistic report no longer collapse into unconstrained text generation.

The contribution is a disciplined route from phenomenology to executable models: a projective, value-laden, socially situated workspace in which perception, imagination, emotion, and action are computationally coupled. PCM therefore advances computational psychology not as another metaphor of mind, but as a testable architecture for model-based behavioral science, psychopathology, and explainable human–machine interaction.

The Global Latent Workspace: a model of cognition with AI applications

Global Workspace Theory (GWT) is a leading account of human cognition and consciousness. In this view, a number of independent specialized modules connect to a shared central representation space; when a module is selected by attention, its contents are mobilized into the Global Workspace, and broadcast across the entire brain, resulting in a unified and integrated experience.

Inspired by this framework, we have developed a deep learning architecture that captures key features of GWT: the Global Latent Workspace (GLW). I will present our GLW and its initial implementations, with promising applications in various AI domains. The model shows improvements in sample efficiency for multimodal representation learning. It can be leveraged for downstream classification and retrieval tasks.

When an action module is connected to the GLW, the whole system exhibits affordance-like properties. The GLW is also beneficial as an input space for RL policy training: the policy is learned with fewer environment steps, and displays zero-shot cross-modal transfer abilities. Finally, augmenting the GLW with "operation" modules and an attention-controlled routing mechanism could open the way toward System-2 reasoning and sequential problem-solving.

Psychology and HCI: emotions, personality, motivation

Research in Human-Computer Interaction requires considering psychological theories in order to design user centred intuitive interfaces. In this presentation we will illustrate how psychological theories of major cognitive dimensions (such as emotion, personality, motivation) can be considered and implemented via computational models or HCI design approaches.