This post is based on the 2011 paper by Julian Marewski and Lael Schooler published in the Psychological Review, “Cognitive Niches: An Ecological Model of Strategy Selection.” How do people select among different strategies to accomplish a given task? By using ACT-R along with heuristic decision strategies, the authors can create a more general bidirectional model that seems to be competitive with such models as parallel constraint satisfaction. In 14 simulations and 10 experiments, they consider the choice between strategies that operate on the accessibility of memories and those that depend on elaborate knowledge about the world. Based on Internet statistics, their model quantitatively predicts people’s familiarity with and knowledge of real-world objects, the distributional characteristics of the associated speed of memory retrieval, and the cognitive niches of classic decision strategies, including those of the fluency, recognition, integration, lexicographic, and sequential-sampling heuristics.
Marewski and Schooler find that there are a couple of things that ease strategy selection. First, when faced with a decision task, people will not actually choose among all the strategies in their repertoire. Rather, the workings of the cognitive system will limit the number of strategies that can be executed, simplifying strategy selection by reducing the consideration set of applicable strategies. This limitation arises from how the strategies and cognitive capacities, such as memory or time perception, interact and represent regularities in the environment. They use the term cognitive niche to refer to the situations in which a strategy is applicable, or afforded. They model these niches in terms of landscapes that quantify the probability of an individual being able to apply a strategy as a function of the interplay between the cognitive system and the environment. They show that this interplay results in partially disjunctive probability landscapes for different strategies, that is, in nonoverlapping niches, which limits strategy selection to situations where the niches of two or more strategies overlap, and thus for this aspect of the selection process does not require the assumption of cost– benefit calculations, learning, or any of the other mechanisms of strategy choice.
Second, where different strategies’ niches overlap, strategy selection depends on the cost– benefit, learning, and other selection mechanisms that assume accuracy, effort, and time as currencies of strategy choice. The model maps out how these currencies are shaped by the interplay between the cognitive system and the environment. In doing so, the model finds that there is less need for trade-offs between making accurate, effortless, and quick decisions than has been found by others.
To illustrate the cognitive niche framework, they consider how people choose among classic strategies that can be used, for example, to infer which of two cars is likely to be of better quality. These strategies can be roughly divided into two types. The first leads to decisions based on knowledge about the world, say, about a manufacturer’s country of origin. These decisions depend on the content of what is retrieved from memory. The second type depends on the characteristics of the retrieval. Such strategies are guided by the accessibility of memories, that is, the ease with which mental content comes to mind. This can take the form, for instance, of a sense of recognition of brand names.
By modeling heuristics in accord with the ACT–R cognitive architecture, quantitative precision can be lent to memory and other capacities the heuristics exploit. At the same time, ACT–R provides a theory of how these capacities interact with the environment. Such an integrative theoretical framework is essential to modeling how a strategy’s niche emerges. The authors have created an ACT–R model of how the environment shapes memory. As the perception of the timing of memory retrieval is critical to the accessibility-based strategies, they also model time perception. Using environmental data, such as how often an object (e.g., Volkswagen) is mentioned in the media, the resulting integrated model of memory and time perception predicts different types of behavioral data. These data include (a) whether people recognize an object and (b) whether they additionally know something about it (e.g., that Volkswagen is a German company). These behavioral data also include (c) the speed of memory retrieval (e.g., recognition time of Volkswagen) as well as (d) people’s perception thereof. The models’ predictions about memory retrieval and time perception, in turn, enable detailed quantitative predictions (e) about the niches where different strategies are applicable and (f) in what regions of their niches the strategies help a person make accurate inferences, as well as (g) in what regions using the strategies requires little effort and time.
For instance, when inferring which of two car brands, Volkswagen or Daewoo, is of higher quality, a person may recognize the name Volkswagen but not the name Daewoo. Following the recognition heuristic, the person would infer that the Volkswagen is of higher quality. When both car brands are recognized, a person can use the fluency heuristic. This heuristic consists of production rules that fire when the memories of two objects are both available but are perceived to have been recalled with different speeds. A person using this heuristic would infer car brands that are more quickly recognized to be of higher quality: If two chunks with the names A and B are retrieved and recognized, and A is perceived to have been recognized faster than B, Then set the goal to respond that A has the larger value on the criterion. Finally, the production rules implementing knowledge-based strategies can fire when a person remembers attributes of objects beyond just recognizing their names. For instance, tally2’s production rules can be summarized as follows: If two chunks with the names A and B are retrieved and recognized, and chunks with knowledge representing more positive cue values for A can be retrieved than can be retrieved for B, Then set the goal to respond that A has the larger value on the criterion. The sense of recognition, ease of retrieval, and knowledge on which the recognition, fluency, tally2, and other heuristics depend can be modeled with ACT–R’s declarative memory.
By mapping out the niches where different strategies are applicable, the cognitive niche framework can be seen as a complement to earlier approaches to strategy selection, rather than as an alternative. For instance, the learning models, including ACT–R’s mechanism for production rule selection, cannot easily explain how an adaptive, systematic selection occurs in the absence of feedback and learning processes. The cognitive niche framework provides a theoretical basis for grouping of strategies, because it allows modeling how the interplay of the environment and the cognitive system reduces the consideration set of strategies to those whose cognitive niches overlap. Grouping allows the number of parameters to be kept reasonably low. At the same time, for those situations where two or more strategies’ niches do not overlap, the cognitive niche framework facilitates understanding how strategy selection emerges as a bottom-up process—in the absence of feedback and learning—solely through the interplay between the cognitive system and the environment.
Moreover, by mapping out how the interplay between the cognitive system and the environment shapes the regions in a strategy’s niche where that strategy will help a person make accurate, fast, and effortless decisions, the cognitive niche framework provides the currencies needed by the cost– benefit, learning, and other earlier approaches to explain the selection among simultaneously applicable strategies. For instance, the cognitive niche framework enables us to model when positive feedback on the success of a strategy will be available for reinforcement learning and when not. To illustrate this, the fluency heuristic (see map above) will most likely be reinforced when recognition time differences are easy to detect, because it is in this situation that a person using the heuristic is most likely to make accurate inferences.
In this investigation of the cognitive niches of decision strategies Marewski and Schooler were careful not to disrupt the natural covariation of memory variables. But as the strategies depend on the knowledge people bring with them to the experiment about, say, cities or companies, they needed to model in some detail how this knowledge is represented in memory. Using little more than web frequency data, they developed an ecological ACT–R memory model for predicting (a) how likely people are to recognize objects in the world, (b) how likely they are to know something more about these recognized objects, and (c) the associated recognition time distributions. The accessibility of these simulated memories reflects not only the natural environment, outside the laboratory, but also how easily a person in the laboratory can retrieve like memories.
The post Option or Strategy Routines is related to this post if only to emphasize how complicated it is.
Marewski, J.N., and Schooler, L.J. (2011). “Cognitive Niches: An Ecological Model of Strategy Selection,” Psychological Review, Vol. 118, No. 3, 393–437.