Higher-Order Potentialities and their Reducers: A Philosophical Foundation Unifying Dynamic Modeling Methods

Tibor Bosse, Jan TreurIJCAI 2007

Abstract

In cognitive science and artificial intelligence, a persistent divide exists between quantitative dynamical systems theory and qualitative symbolic modeling. This paper proposes a unified philosophical foundation to bridge this gap by introducing the concepts of higher-order potentialities and reducers. By viewing state changes as anticipatory properties that determine future evolution, the authors show that differential equations and cognitive rules perform the same underlying function: reducing complex future possibilities to current state properties. This framework provides a consistent language for modeling systems ranging from single neurons to complex human reasoning.

Introduction

The modeling of intelligent systems has historically been split into two paradigms. Dynamical Systems Theory (DST) utilizes differential equations to model low-level, continuous processes like motor control. Conversely, symbolic modeling uses qualitative rules, such as production systems or Belief-Desire-Intention (BDI) models, to describe high-level cognition. These paradigms are often viewed as mutually exclusive. This work seeks a common philosophical foundation that can describe both, enabling the creation of hybrid models that are philosophically consistent across all levels of abstraction.

The Framework of Higher-Order Potentialities

The core of the proposed framework rests on three interrelated concepts:

  1. Potentialities: Drawing from Aristotelian philosophy, a potentiality ($p$) is an anticipatory state property. If a system state possesses potentiality $p$ for property $a$, then $a$ will manifest in the subsequent state.
  2. Higher-Order Potentialities: Change itself is subject to change. A first-order potentiality ($p^{(1)}$) explains a change in a basic property. A second-order potentiality ($p^{(2)}$) explains the change in the first-order potentiality. This creates a hierarchical chain analogous to derivatives in calculus (e.g., position, velocity, acceleration).
  3. Reducers: To prevent an infinite regress of "why" a state changes, the authors introduce "reducers." A reducer is a relationship that equates a higher-order potentiality to a specific combination of lower-level or basic state properties.
The hierarchy of higher-order potentialities $p^{(n)}$ determining state transitions over time.

In this view, a law of nature or a cognitive rule acts as a reducer that grounds the dynamics in the current state, effectively stopping the regress of higher-order potentialities.

Applications and Unification

The framework demonstrates its versatility by mapping onto diverse fields:

  • Physics: In Newtonian mechanics, momentum is viewed as a first-order potentiality for position, while force ($F=ma$) acts as a second-order potentiality. Newton’s Second Law serves as the reducer, relating force back to mass and acceleration.
  • AI and BDI Models: Mental states can be framed dynamically. Intentions are first-order potentialities for action, while desires act as second-order potentialities for intentions. This grounds "mental" states in the same formal structure as physical momentum.
  • Production Systems: A rule ($\phi \to \psi$) is interpreted as a logical implication where the antecedent $\phi$ reduces the potentiality of the consequent $\psi$.

Biological Learning in Aplysia

The framework is further validated through the classical conditioning of the sea hare, Aplysia. The learning process involves higher-order potentialities—the potential to learn to react—which eventually reduce to physical neurological states, such as synaptic strengths, as the animal becomes conditioned.

Neural circuit of the sea hare Aplysia, illustrating the reduction of behavioral potential to synaptic states.

Conclusion

By framing both differential equations and symbolic rules as reducers of higher-order potentialities, this work provides a qualitative unification of disparate fields. While the framework is currently theoretical, it offers a robust foundation for developing hybrid modeling languages (such as LEADSTO) that combine symbolic logic and numerical dynamics. Future work will focus on identifying reducers in complex cognitive systems where the underlying "laws" of change are less well-defined than in physics.

Summary updated

BibTeX
@inproceedings{bosse2007higher,
  author = {Tibor Bosse and Jan Treur},
  title = {{Higher-Order Potentialities and their Reducers: A Philosophical Foundation Unifying Dynamic Modeling Methods}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2007},
}
Figure