Overview
Cognitive science historically viewed the mind and brain in terms of computation. There are three main research traditions that follow: computationalism, which emphasizes structured mental representations; connectionism, which focuses on learning and implementation in parallel hardware; and rational analysis, which focuses on why cognitive processes reliably lead to observable thought and behavior.
| Name | View of cognition | Key figures | Main limitations |
|---|---|---|---|
| Computationalism (a.k.a., the symbolic approach) | The mind operates as an information processing machine, especially performing computations over symbols following formal syntactical rules. (Heavily influenced by the development of digital computers in the 1950s). | Chomsky (e.g., computational arguments for Universal Grammar). Simon and Newell (e.g., @1976newellComputer). Fodor (e.g., The language-of-thought-hypothesis). Minsky. | Limited calculus for dealing with uncertainty (e.g., “local ambiguity” of natural language as well as sensory perceptions). ”Excessively rigid” compared to gradual biological learning and degradation. Unclear how to implement serialized process in in parallel brain structures. |
| Connectionism (a.k.a., parallel distributed processing) | Cognition is distributed signal activity, and learning involves the bottom-up process of locally adjusting network components (i.e., units). | McClelland (e.g., @1988mcclellandAppeal). Rosenblatt, developer of the perceptron model. Hinton, Sejnowski, Rumelhart. | Lack of representational power for rich symbolic systems (e.g., implicated in human language, reasoning, planning). |
| Rational approaches | Agents take optimal solutions to cognitive problems, with reasoning constrained by a well-defined notion of coherence. Focuses on the characteristics of the optimal solution, not the mechanisms for obtaining it. | Gibson’s ecological analyses of visual perception. Marr’s computational-level explanations. Shepard (e.g., 1987, 1994). Anderson (e.g., rational analysis). |
Related notes:
- Classical theories of mind
- Systematicity and productivity
- Nativist and empiricist positions in cognitive science
- (Resource-)rational analysis
Connectionism vs. computationalism
| Computationalism | Connectionism | |
|---|---|---|
| Examples | “Good old-fashioned AI”; expert systems | Parallel-distributed processing; § Deep Learning |
| Basic characteristics | - Symbol manipulation - (Domain-specific) rules of inference over explicit representations - Ability develops in distinct stages - Innateness - Systematicity and productivity | - Subsymbolic processes - Emergent representations - Gradual learning - Domain-general learning mechanisms - Flexible behavior/graceful degradation |
| Definition of learning | Developing explicit rules that capture powerful generalizations about the world. | Tune connections to capture interdependencies between activations, thus acting “as though it knows the rules.” |
Highlights
- Classical cognitive science studies how information processing—representation, transformation, and propagation—organizes the behavior of an individual: “Cognitive science thus concerns itself with the nature of knowledge structures and the processes that operate on them. The properties of these representations inside the system and the processes that operate on representations are assumed to cause or explain the observed performance of the cognitive system as a whole.”
References
General references
- Simon (1981), The Sciences of the Artificial
- Fodor & Pylyshyn (1988), “Connectionism and Cognitive Architecture: A Critical Analysis”
- @2011kimMind, “Mind as a Computing Machine”
- @2024griffithsBayesian, Bayesian Models of Cognition
Literature notes
| Cite key | One-line takeaway |
|---|---|
| @1995hutchinsHow, “How a cockpit remembers its speeds” | Proposes a theoretical framework for extending classical computationalist analysis—explaining behavioral properties in terms of internal representations—to larger “socio-technical systems”; demonstrates the framework by analyzing airplane cockpits. |