Introduction
In the early 1990s, the field of Artificial Intelligence and educational psychology faced a significant challenge from the "situated cognition" movement. Proponents of this movement, such as William Clancey and Brown et al., argued that knowledge is not a discrete substance stored within the mind, but rather a dynamic construction emerging from interactions with the environment. In their paper, How Situated is Cognition?, Jacobijn Sandberg and Bob Wielinga provide a critical theoretical analysis of this paradigm shift, arguing that the radical rejection of internal representations is a fundamental error—essentially "throwing the baby out with the bathwater."
The Identity Hypothesis vs. Functional Equivalence
A core tenet of the situationist critique was the "Identity Hypothesis"—the idea that traditional AI claims its computer code is a literal map of human neural architecture. Sandberg and Wielinga argue that this is a straw man. Most cognitive scientists and AI researchers actually operate under a "functional-equivalence hypothesis." This perspective suggests that while the "low-level" implementation (the "cognitive wheels") of a computer differs from biological "wetware," the high-level models can still accurately represent human-like behavior and knowledge processing. We build models that behave like humans at a high level without assuming the underlying architecture matches biological neurons.
The Necessity of Decontextualization
Situationists championed "authentic" and "situated" learning, suggesting that knowledge should always be acquired in the context of its use (e.g., learning mathematics through carpentry). However, Sandberg and Wielinga contend that the primary goal of formal education is the provision of abstract, decontextualized tools. These abstractions are what allow individuals to apply knowledge across diverse and unforeseen contexts. They argue that the entire point of formal education is to provide these general-purpose tools that can be used in any context.
The Transfer Problem
The authors highlight the "Transfer Problem" as a significant weakness in purely situated models. If knowledge is inextricably tied to a specific context—such as solving a math problem using a specific physical medium like cottage cheese (a famous example in situated literature)—the learner's ability to solve the same problem in a different context (e.g., with a liquid) is severely compromised. Formal, abstract rules are the mechanism that enables the transfer of skills between different domains. Without these internal structures, knowledge remains fragmented and brittle.
Memory as a Functional Substance
While situationists viewed memory merely as a "capacity for behavior," Sandberg and Wielinga defend the utility of treating knowledge as a "mental substance." They argue that this abstraction is scientifically productive, explaining how humans communicate, store, and reuse information over time. Reducing the mind to a simple organism reacting to its immediate environment ignores the complex, long-term training and internal consolidation required to achieve expertise.
Conclusion and Educational Implications
The paper concludes that while situationism correctly identifies the importance of context in everyday cognition, it lacks the explanatory power of traditional cognitive theories. In the realm of education, the authors warn that "cognitive apprenticeship"—learning solely through practice within a community—risks leaving students with "partial, half-remembered" knowledge that fails to generalize to new, complex problems.
Instead of a radical paradigm shift, Sandberg and Wielinga advocate for advancing traditional AI and cognitive frameworks to better account for situated behavior without abandoning the essential role of internal representation. They suggest that the functional level of knowledge and representation must remain central to the field, even as new architectures like neural networks emerge.
Summary updated