«Sebastien Helie University of California Santa Barbara Ron Sun Rensselaer Polytechnic Institute March 12, 2012 To appear in: Psychology of Problem ...»
Implicit cognition in problem solving
University of California Santa Barbara
Rensselaer Polytechnic Institute
March 12, 2012
To appear in: Psychology of Problem Solving
Address correspondence to:
Department of Psychological & Brain Sciences
University of California, Santa Barbara
Santa Barbara, CA 93106-9660
Cognitive Science Department
Rensselaer Polytechnic Institute
110 8th street, Carnegie 302A Troy, NY 12180 email@example.com Running head: Implicit cognition Abstract Psychological theories of problem solving have largely focused on explicit processes that gradually bring the solver closer to the solution step-by-step in a mostly explicit and deliberative way. This approach to problem solving is often inefficient or ineffective when the problem is too complex, poorly understood, or ambiguous. In such a case, a more intuitive or implicit approach to problem solving might be more appropriate. The role of implicit processes in producing creative solutions is called ‘incubation’. In this chapter, we review the Explicit-Implicit Interaction (EII) theory of creative problem solving, an integrative framework focusing on the role of implicit processes in problem solving. The EII theory has been implemented using a computational model based on the CLARION cognitive architecture. Results from some of the computational simulations are also reviewed. We conclude by discussing how the EII theory can contribute to cognitive neuroscience research on creative problem solving.
Introduction Many psychological theories have highlighted a role for implicit cognitive processes (e.g., Ashby, Alfonso-Reese, Turken, & Waldron, 1998; Evans, 2006; Reber, 1989). For instance, similarity has been shown to affect reasoning through processes that are mostly implicit (Sun, 1994; Sun & Zhang, 2006). In problem solving, implicit processes are often thought to generate hypotheses that are later explicitly tested (Evans, 2006; Helie, Proulx, & Lefebvre, 2011). Yet, most theories of problem solving have focused on explicit processes that gradually bring the problem solver closer to the solution in a deliberative way (Dorfman, Shames, & Kihlstrom, 1996). However, when an ill-defined or overly complex problem has to be solved (e.g., when the initial state or the goal state can lead to many different interpretations, or when the solution paths are highly complex), the solution is often found by sudden ‘insight’ (Pols, 2002; Reber, 1989; Schooler & Melcher, 1995; Schooler, Ohlsson, & Brooks, 1993), and regular problem solving theories are for the most part unable to account for this apparent absence of deliberative strategy (Bowden, Beeman, Fleck, & Kounios, 2005).
A complementary line of research on creative problem solving has tried to tackle complex problem solving for many years. However, theories of creative problem solving tend to be fragmentary and usually focus only on a subset of phenomena, such as incubation (i.e., a period away from deliberative work on the problem; for a review, see Smith & Dodds, 1999) or insight (i.e., the sudden appearance of a solution; for a review, see Pols, 2002). The lack of detailed computational models has resulted in their limited impact on the field of problem solving (Duch, 2006). In this chapter, we review some of the evidence supporting a role for implicit processing in problem solving and an integrative framework that was recently proposed, namely the Explicit-Implicit Interaction (EII) theory (Helie & Sun, 2010). One of the strengths of the EII theory is that it provides a process-based account of creative problem solving that is amenable to computational implementation. To illustrate this point, we then present some example simulation results from a CLARION-based implementation of the EII theory. We conclude with a discussion of how this work on integrative theories can facilitate new developments in the cognitive neuroscience of creative problem solving.
The role of creativity in problem solving has been acknowledged at least since Wallas’ (1926) seminal work. According to Wallas, humans go through four different stages when trying to solve a problem: preparation, incubation, illumination (i.e., insight), and verification. The first stage, preparation, refers to an initial period of search in many directions using (essentially) logic and reasoning. If a solution is found at this stage, the remaining stages are not needed. However, if the problem is ill-defined and/or too complex to be fully grasped, the preparation stage is unlikely to generate a satisfactory solution. When an impasse is reached, the problem solver stops attempting to solve the problem, which marks the beginning of the incubation phase. Incubation can last from a few minutes to many years, during which the attention of the problem solver is not devoted to the problem. The incubation period has been shown to increase the probability of eventually finding the correct solution (e.g., Dodds, Ward, & Smith, 2003; Smith & Dodds, 1999). The following stage, insight, is the “spontaneous” manifestation of the problem and its solution in conscious thought (i.e., the “Eureka!” moment). The fourth stage, verification, is used to ascertain the correctness of the insight solution. Verification is similar to preparation, because it also involves the use of deliberative thinking processes (with logic and reasoning). If the verification stage invalidates the solution, the problem solver usually goes back to the first or second stage and this process is repeated.
Even though the stage decomposition theory is difficult to test empirically, it has been used to guide much of Gestalt psychologists’ early research program on problem solving (e.g., Duncker, 1945; Kohler, 1925; Maier, 1931). According to Gestalt psychology, ill-defined problems are akin to perceptual illusions: they are problems that can be understood (perceived) in a number of different ways, some of which allow for an easier resolution (Pols, 2002). Hence, the preparation stage would be made up of unsuccessful efforts on an inadequate problem representation, incubation would be the search for a better problem representation, and insight would mark the discovery of a problem representation useful for solving the problem. The verification phase would verify that the new problem representation is equivalent to the initial problem representation (Dunker, 1945). This Gestalt theory of problem solving provides a sketchy high-level description of creative problem solving but no detailed psychological mechanism was proposed.
More recent research has focused on finding evidence supporting the existence of the individual stages of creative problem solving. Because the preparation and verification stages are thought to involve mostly regular reasoning processes (Wallas, 1926), not much effort has been devoted to these two stages (relevant results can be borrowed from the existing literature; see, e.g., Johnson-Laird, 1999; Sun, 1994). In contrast, incubation and insight have received more attention.
Incubation A recent review of experimental research on incubation shows that most experiments have found a significant effect of incubation (Dodds et al., 2003; see also Sio & Ormerod, 2009). Those experiments found an effect of incubation length, preparatory activity, clue, and distracting activities on participants’ performance. The review suggests that performance is positively related to incubation length and that preparatory activities can increase the effect of incubation. Presenting a clue during the incubation period also has a strong effect. If the clue is useful, the performance is improved; if the clue is misleading, the performance is decreased. Moreover, the effect of clues is stronger when the participants are explicitly instructed to look for clues (Dodds, Smith, & Ward, 2002). The effect of distracting activities is not as clear. Helie, Sun, and Xiong (2008) showed that distracting activities can have different effects on
resources/processing with the task used to assess the presence of incubation. Finally, incubation has also been linked to well-known cognitive effects such as reminiscence (i.e., the number of new words recalled in a second consecutive free recall test; Smith & Vela, 1991) and priming (Yaniv & Meyer, 1987).
Insight In a recent review of the different definitions used in psychology to characterize ‘insight’, Pols (2002) found three main elements. First, insight does not constitute just another step forward in solving a problem: it is a transition that has a major impact on the problem solver’s conception of the problem. Second, insight is sudden: It usually constitutes a quick transition from a state of ‘not knowing’ to a state of ‘knowing’. Third, the new understanding is more appropriate: Even when insight does not directly point to the solution, it leads to grasping essential features of the problem that were not considered previously.
In experimental psychology, insight is often elicited using ‘insight problems’ (e.g., Bowden et al., 2005; Dorfmann et al., 1996; Isaak & Just, 1996; Mayer, 1995; Pols, 2002). Such problems are diverse and characterized by the absence of direct, incremental algorithms allowing for their solutions. In many cases, they are selected because they have been shown to produce insight solutions in previous studies (Bowden et al., 2005). Empirically, insight is identified by a strong discontinuity in the subjective ‘feeling of knowing’ or the progress made in a verbal report (Pols, 2002).
Some research has even shown a sudden increase of heart rate just before insight is reached (whereas regular problem solving is accompanied by a steady increase in heart rate; see, e.g., Jausovec & Bakracevic, 1995).
The Explicit–Implicit Interaction (EII) theory (Helie & Sun, 2010) attempts to integrate and thus unify existing theories of creative problem solving in two different senses. First, most theories of creative problem solving have focused on either a highlevel stage decomposition (e.g., Wallas, 1926) or on a process explanation of only one of the stages (Lubart, 2001). Second, the process theories of incubation (e.g., Smith & Dodds, 1999) and insight (e.g., Mayer, 1995; Ohlsson, 1992; Pols, 2002) are often incomplete and sometimes mutually incompatible. EII attempts to integrate the existing theories to makes them more complete in order to provide a detailed description of the processes involved in key stages of creative problem solving. EII starts from Wallas’ (1926) stage decomposition of creative problem solving and provides a detailed process-based explanation sufficient for a coherent computational implementation. In this section, we present the core assumptions underlying the EII theory. Details on how EII captures existing theories of incubation and insight can be found in Helie & Sun (2010).
Principle #1: The co-existence of and the difference between explicit and implicit knowledge The EII theory assumes the existence of two different types of knowledge, namely explicit and implicit, residing in two separate modules (Sun, 2002). Explicit knowledge is easier to access and verbalize and often said to be composed of symbols following hard constraints (Sun, Merrill, & Peterson, 2001; Sun, Slusarz, & Terry, 2005).
However, using explicit knowledge requires extensive attentional resources (Curran & Keele, 1993; Sun et al., 2005). In contrast, implicit knowledge is relatively inaccessible, harder to verbalize, often “subsymbolic”, and follows soft constraints (Sun, 1994, 2002).
However, using implicit knowledge does not require much attentional resources. As such, explicit and implicit knowledge is processed differently. According to the EII theory, explicit processes perform some form of rule-based reasoning (in a very generalized sense; Smith, Langston, & Nisbett, 1992; Sun 1994) and represents relatively crisp and exact processing (often involving hard constraints; Sun et al., 2001), whereas implicit processing is ‘associative’ and represents soft-constraint satisfaction (Evans, 2008; Sloman, 1996; Sun, 1994).
Principle #2: The simultaneous involvement of implicit and explicit processes in most tasks Explicit and implicit processes are involved simultaneously in most tasks under most circumstances (Smith & DeCoster, 2000; Sun, 2002). This can be useful because different representations and processing are used to describe the two types of knowledge (as described above in Principle #1). As such, each type of processes can end up with similar or conflicting conclusions that contribute to the overall output (Evans, 2007).
Principle #3: The redundant representation of explicit and implicit knowledge According to the EII theory, explicit knowledge and implicit knowledge are often “redundant”: they can amount to re-descriptions of one another in different representational forms. For example, knowledge that is initially implicit is often later recoded to form explicit knowledge (through “bottom-up learning”; Helie et al., 2011; Sun et al., 2001, 2005). Likewise, knowledge that is initially learned explicitly (e.g., through verbal instructions) is often later assimilated and re-coded into an implicit form, usually after extensive practice (top-down assimilation: Ramamoorthy & Verguts, 2012; Sun & Zhang, 2004). There may also be other ways redundancy is created, e.g., through simultaneous learning of implicit and explicit knowledge. Redundancy often leads to interaction.
Principle #4: The integration of the results of explicit and implicit processing Although explicit and implicit knowledge are often re-descriptions of one another, they involve different forms of representation and processing, which may produce similar or different conclusions (Sun & Peterson, 1998). The integration of these conclusions can lead to synergy, that is, overall better performance and faster learning (Sun et al., 2001). EII assumes that this synergy is an important component of creative problem solving.