Cognitive Science: Research Programmes & Stratified Disagreement

Cognitive science is better understood as a cluster of competing research programmes than a unified discipline. These programmes disagree not just on facts, but on deep commitments regarding what is real (Ontology), what counts as an answer (Explanation), and how to investigate (Methodology).

The "Hard Core" Stratification

Layer What it settles Example Disagreement
Ontological What cognition is Computation vs. coupling
Explanatory What counts as explanation Mechanism vs. optimality proof vs. phase portrait
Methodological How to investigate Behavioral experiments vs. dynamical modeling

These come apart. Two researchers can share ontology (representationalism) but diverge on explanatory norms (mechanistic vs. computational-level). Different components have different rigidity.

The Five Programmes

Classical/Symbolic

Cognition is rule-governed symbol manipulation.

Key Figures: Fodor, Newell, Pylyshyn

Connectionist

Cognition is pattern transformation across distributed representations.

Key Figures: Rumelhart, McClelland, Smolensky

Bayesian

Cognition is approximate optimal inference under uncertainty.

Key Figures: Tenenbaum, Griffiths, Friston

Embodied/Enactive

Cognition is inseparable from bodily action and environmental coupling.

Key Figures: Varela, Thompson, Di Paolo

Dynamical Systems

Cognition is continuous state-space evolution.

Key Figures: Kelso, Thelen, Van Gelder
Symbolic
Connectionist
Bayesian
Enactive
Dynamical
Ontological What cognition IS
Computation Rule-governed symbol manipulation & syntax.
Vector Transform Pattern transformation across distributed nodes.
Inference Approximate optimal inference under uncertainty.
Coupling Inseparable from bodily action & environment.
Evolution Continuous state-space evolution in time.
Explanatory What counts as explanation
Algorithm Program-level description of information processing.
Network Dynamics Learning rules, weight changes, energy landscapes.
Optimality Rational analysis: why is this behavior optimal?
History History of structural coupling & viability.
Attractor Phase portraits, manifolds, stability analysis.
Methodological How to investigate
Behavioral Reaction times, error rates, logic proofs.
Simulation Training artificial neural networks on tasks.
Ideal Models Comparing behavior to ideal observer models.
Phenomenology First-person experience + ethology.
Diff Eq Differential equations & coordination dynamics.

Programme Relations

Relations are a partial order, not pure rivalry:

  • Bayesian & Connectionist: Complementary across Marr levels.
  • Embodiment: Often about scope (boundaries) rather than mechanism.
  • Dynamical Systems: A mathematical vocabulary applicable within multiple ontologies.

This dissolves pseudo-debates (Bayesian vs. connectionist) while preserving genuine ones (classical vs. radical enactivism).

Case Study: Predictive Processing

PP attempts unification by straddling: Bayesian at computational level, Connectionist in implementation, Dynamical in mathematics.

However, there are limits to what it can absorb without becoming unfalsifiable:

  • Radical enactivism's rejection of representational content.
  • Strongly externalist claims where cognition crosses the skin.
  • Social constitution of norms (are practices just "priors"?).

Hierarchy of Presupposition

Level Example Status
Phenomenal Behavioral regularities needing explanation Minimal shared
Thin Theoretical Adaptive coordination with environment Already boundary-drawing
Thicker Framing "Cognition", "Representation" Contested
Programme-Specific "How is X represented?" Fully internal

Critical Realism as Regulative Commitment

The Pragmatist View

Bracketing metaphysics reveals that theoretical questions lack determinate answers. Fertility is truth.

My Stance

Bracketing is epistemically modest, not ontologically deflationary. Cognitive mechanisms and social processes are real and stratified.

What this buys:

  • Against instrumentalism: PP translating enactivism into priors might be wrong about the joints of reality, not just less useful.
  • Differentiated realism: Methodology is instrumental; Ontology tracks real mechanisms.
  • Unified stance: Experiments and ethnography are probing different strata of the same reality.