On the Use of Recognition in Inferential Decision Making


sanantonio450This post and the next post are based on Rudiger Pohl’s article, “On the Use of Recognition in Inferential Decision Making” that appeared in the Journal of Judgment and Decision Making in 2011.  The Journal had three issues devoted to recognition.  Pohl provides the best summary and is also the last.  I found that some of the articles with two or three authors trying to come up with a summary failed apparently because there was so much disagreement among the authors.

“Intuition is nothing more or less than recognition.”  Daniel Kahneman delivers this and credits Simon in Thinking Fast and Thinking Slow.  Pohl’s article does not address this statement, but it helps me address it. Maybe the statement is not making intuition simpler, but making recognition much more complicated.

Unfortunately, these posts are something of  prequels to Bidirectional Reasoning, Act-R Cognitive Architecture, and Cognitive Niches. Pohl explains the concepts (I did some of this in Bounded Rationality, the Adaptive Toolbox, Simple Heuristics at Work in the World), details the issues, and even points out the centrality of the Marewski v Glockner arguments discussed in the posts, Bidirectional Reasoning, Act-R Cognitive Architecture and Cognitive Niches.

This first post will give a brief overview of the Recognition Heuristic and Pohl’s list of controversial topics along with his conclusions. The next post examines some of the controversial issues in more detail. As one of the simplest heuristics in Gigerenzer’s “adaptive toolbox”, the recognition heuristic (RH) exploits recognition and may reach a high level of accuracy in inferential decisions. For example, if asked which of two cities is larger, A or B, and given one recognizes A, but not B, one may simply follow the recognition cue and infer that A is the larger city. In domains in which the probability of recognizing an object is substantially related to its criterion value (here the city’s size), such a simple strategy will lead to many correct answers, far above chance.  RH and TTB (Take the Best) and several other heuristics were assumed to form the cognitive tools in an “adaptive toolbox” that human decision makers possess. “Adaptive” means that, depending on the task and situation, different, ecologically valid tools could be applied. Because these tools exploit regularities of the given environment, they allow good and fast decisions with minimal effort. Hence, these strategies were accordingly also termed “fast and frugal heuristics” (FFH; as opposed to more effortful and presumably time-consuming, complex decision processes).

The RH was assumed to be domain-specific, that is, useful only in domains with a high correlation between probability of recognition and criterion value. Recognition was (and is) still used in a binary fashion, that is, objects are either recognized or not. The most important
feature, however, was that recognition should be used as the only cue: The authors, Gigerenzer and Goldstein, stated that, if recognition discriminates between the alternatives (i.e., when one object is recognized and the other not), then (1) no other information beyond recognition will be considered and therefore (2) nothing can overturn the inference based on the recognition cue. The first hypothesis is known as “one-reason decision making”, the second one as “non-compensatory strategy.” According to Pohl these “rather bold” proposals have fueled the strongest reactions of other researchers.

Pohl created this list of controversial topics surrounding the recognition heuristic (RH).
1. Recognition as a memory-based process
2. The RH as a cognitive process model
3. Proper conditions of testing the RH
4. Measures of using the RH
5. Reasons for not using the RH
6. The RH as a non-compensatory strategy
7. Evidence for a Less-is-more effect (LIME)
8. The RH as part of the toolbox

In their most recent presentation of the RH, Gigerenzer and Goldstein have clarified the conditions and predictions of the RH theory.  According to Pohl, they now posit that, before the RH is applied, an evaluation will be run that tests whether the recognition cue should be used or not.

Pohl summarizes the empirical findings by stating that the recognition heuristic is not used by all people all the time and under all circumstances, and that people appear to differ greatly in their reliance on recognition for inferences.  Thus, one viable and legitimate question, asked from the early days of the FHH approach, is which heuristics like the RH are suited for which domains and tasks. Another question, although also asked from the beginning, namely that of individual differences, might be more difficult to answer. That question as posed by Pohl is if the “adaptive” use of such decision strategies like the RH reflects environmental regularities, why should individuals differ so much in their perception or evaluation of these regularities? Why should, within the same domain, some people rely on recognition almost always, and others only occasionally (as has been reported in some studies)? This in Pohl’s view is still  puzzling.

In sum, the general question concerning the RH would then not be to ask whether it is used but rather when and by whom it is used . Pohl notes that when phrased in such a way, the current controversy surrounding the RH looses much of its impetus, and one may wonder why such a simple question has raised so many debates. Pohl suggests that the answer is that some have not stopped there and have instead questioned the RH as a valid tool and finally the whole FFH approach. One reason for such a fundamental critique may be grounded in the FFH’s central assumption that there are a number of different tools available from which the decision maker has to choose the appropriate one that best fits a given environment. Moreover, according to the theory, once a potentially useful strategy is identified, the decision maker has to check whether any reason would speak against using that otherwise optimal tool. All these reasons may lead to the selection, evaluation, and (finally) application of other tools. Such a series of strategy selection and evaluation steps, repeatedly for each single trial, seems quite cumbersome  and too complicated to work in practice. In addition to this more theoretical argument, some researchers came to a negative conclusion regarding the FFH approach when summarizing the available empirical evidence.

Pohl writes that alternative conceptions that posit fewer or only one mechanism instead of multiple tools have been proposed and may gain ground. As examples, Pohl mentions the evidence-accumulation models, reappearing in the “adjustable-spanner” metaphor, and the recently proposed “Parallel Constraint Satisfaction” network model. One of the major advantages of these approaches is that they can be applied to all comparison types (not just recognition cases) and also easily combine compensatory and non-compensatory use of probabilistic cues within the same architecture and thereby avoid the need to change tools from one trial to the next.  Pohl notes that is still too early to draw any further conclusions about how good these alternatives will fare in the end.

Marewski has provided a countering view in Cognitive Niches.

Pohl, Rudiger F, (2011). On the use of recognition in inferential decision making:  An overview of the debate. Judgment and Decision Making, Vol. 6, No. 5, July 2011, pp. 423–438.

Goldstein, D. G., & Gigerenzer, G. (2002). Models of ecological rationality: The recognition heuristic. Psychological Review, 109, 75–90.