A new framework for the research on Introspection

Get Complete Project Material File(s) Now! »

The Nisbett and Wilson Canon (NWC)

The renewed interest for introspection in the early days of cognitivism was accompanied by the question about its reliability. Nisbett and Wilson (1977) famously suggested that it is only possible to trust those mental contents available to the participants’ consciousness. The authors proposed the distinction between accessible mental contents (cognitive states) and inaccessible ones (cognitive processes). The research comes from a systematic review of experiments in social psychology (research areas included: attribution, cognitive dissonance, subliminal perception, problem solving, etc.), from which Nisbett and Wilson concluded that individuals might be: (a) unaware of the existence of a stimulus that importantly influenced a response, (b) unaware of the existence of the response, and (c) unaware that the stimulus has affected the response. These points form what we shall call Nisbett and Wilson’s canon (henceforth, NWC): introspective access is restricted to mental products or perceptual contents; the underlying sensory transformations are inaccessible to the 9 The pessimist version to this thesis (Schwitzgebel, 2011, 2012) suggests that such multiplicity of sources of information would be the cause of inter and intra individual variability, traditionally evidenced in a report of this type. Optimistic versions of this thesis (Bayne, 2015), converge on the idea that not all sources of information are equally weighted in the processing of introspective data.
individuals’ consciousness10. In summary, the NWC maintains that cognitive processes would not be accessible to introspection; when participants try to introspect on processes, they make use of a priori theories about the causal relationship between stimulus and response.
In order to illustrate the above, Nisbett and Wilson present an experiment where participants are shown four pairs of socks and are asked to choose the best pair. Participants were not informed that all four pairs were identical. After the participants’ decision, they were requested to justify their choice. The experiment presented a robust linear effect of position: as the position of the pair of socks advanced from left to right, the preference increased. The experiment does not investigate or speculate about the origin of this effect. The point of interest is that, when asking them for the cause that motivated their preference, none of the participants mentioned position as a determining factor. Even after asking them directly if position could have affected their preference, all participants denied this possibility. On the contrary, all participants confabulated regarding the motives of their choice.
It is important here to distinguish the cognitive process from the cognitive state. Individuals successfully access their sensory experience regarding the apparently “better quality” of the selected pair of socks; however, they do not access the cognitive process that motivated this decision. According to Schwitzgebel (2011): « They (Nisbett and Wilson) are skeptical about our knowledge of why we selected a particular brand of socks, not about the fact that we do judge them to be superior or about our sensory experience as we select them”. According to this, Nisbett & Wilson do not reject all kinds of introspective access; the authors accept that it is possible to introspectively describe mental contents such as perceptual experiences; however, there would be no introspective knowledge of the causal process underlying and driving our judgments, decisions, emotions and sensations. The difficulty in this (and other) experiment is that it is nearly impossible to tell apart the veridical introspection of hallucinated properties of the stimulus caused by a mechanism the participants are not aware of from a pure confabulation, that is: the invention of a justification for the choice that is rationally and socially acceptable in the absence of any underlying cognitive difference. Our studies will try to offer solutions to this issue.
10 The common definition of a cognitive state refers to a mental content that accompanies or precedes the decision (e.g., the level of confidence in a decision or the perception of the response time on a cognitive task). On the other hand, the cognitive processes refer to the cognitive treatments or sensorial transformation that both precedes and determines the decision (Rich, 1979). Another way to introduce the concepts of cognitive “states” and “processes” is proposed by White (1980) through the distinction between “knowing that” and “knowing how”, respectively.
The contemporary formulation of the NWC (Wilson, 2002, 2003; Wilson & Dunn, 2004) maintains that introspection would have retrospective access only to consciously processed mental contents (cognitive states) and not to those contents with unconscious processing (cognitive processes). This would constitute a central characteristic of our cognitive architecture. Relevant information about mental states is maintained in consciousness and superfluous information is kept underneath: this is precisely the notion of the adaptive unconscious. Moreover, recent studies in social cognition (Johansson, Hall, Silkström & Olsson, 2005; Johansson, Hall, Silkström, Tärning & Lind, 2006) have supported the NWC through a phenomenon called choice blindness (see also Froese, 2013; Jack, 2013; Petitmengin, Remillieux, Cahour & Carter-Thomas, 2013). A famous experiment on this phenomenon consists in presenting to participants two pairs of cards with two faces. Immediately afterwards, participants were asked to determine which one they preferred. In certain trials, researchers asked participants to report the cause that motivated their preference. The critical point of the experiment is that in some trials and immediately after the individuals stated their preference, researchers, by means of legerdemain, presented to participants the card they did not select, without their noticing it (Figure 2). Surprisingly, participants did not detect that the researcher had handed them a different card and furthermore they continued to articulate justifications for their decisions that are related to the new card handed to them by the researcher. Participants are blind to the causes that guided their original choice. The results of this study support the NWC model, as it provides evidence for a lack of conscious access to the causes that motivate behaviour. Importantly, recent studies (Hall et al., 2013; Johansson, Hall, Tärning, Sikström & Chater, 2013; McLaughlin & Somerville, 2013) confirm that this phenomenon is independent from the experimental context. An important aspect to note is that all studies commented upon by Nisbett and Wilson, in addition to the more recent evidence in favor of the NWC, comes from social psychology experiments. There is no evidence from basic cognitive experimental psychology that shows blindness to cognitive processes thus far. All evidences seem to comprise of crucial psychosocial contexts or researcher-participant relationship as driving factors (Petitmengin et al., 2013).
Figure 2. Choice blindness phenomenon (from: Johansson, Hall, Silkström & Olsson, 2005): individuals tend to justify their choice of Y, even when their original choice was, in fact, X. In first place (A) the participant is shown two faces, then (B) participants are asked to select the card with their highest preference. Immediately afterwards, (C), only on certain trials the experimenter switches the card previously selected by the participant. Finally (D), the experimenter hands the participant a card (card with the face not originally selected by the participant) and asks participants to justify their preference.
In opposition to the NWC, several conceptual and methodological objections have been proposed (Ericsson & Simon, 1980; Smith & Miller, 1978; White, 1988). Among the most important objections are the following: Firstly, (i) the NWC does not propose a clear distinction between a cognitive process and a cognitive state. Even though Nisbett and Wilson (1977) specify a list of contents or introspectively accessible mental states, the authors only offer a vague definition of a cognitive process: “causes that guided, lead to or motivate a decision” (p. 232). In a subsequent article, Nisbett and Ross (1980) define a cognitive process as “the causal relation between events or mental contents”. Notice in both cases the pivotal notion of a cause, which is, in itself a very elusive concept. Since Hume (1711-1776), many philosophers have criticized causality as a relation or theoretical construct that is inaccessible to individuals’ consciousness. Perhaps introspection of processes defined as causal is impossible because mental causation is difficult to define. Defenders of introspection (Engelbert & Carruthers, 2010), in line with the posture of the present thesis, in no case argue in favor of an introspective access of that type. My experiments will investigate access to the ongoing mental states occurring before, during and after the decision, independently of their causal status with respect to our behavior. In fact, Nisbett and Wilson (1977) only distinguish states and processes as a function of their introspective accessibility, generating an arbitrary (Smith & Miller, 1978) and circular argument (White, 1980)11. A conceptual strategy to avoid this circularity, and that additionally facilitates the design of experiments, is to define a cognitive process simply as “the sensory and representational transformations that occur between a stimulus and a response”. This operational redefinition has a double objective: eliminating accessibility as a criterion distinguishing cognitive states and processes, but also to simplify participants’ introspective tasks. Indeed, when requesting participants to report “the causes that motivate behavior” it is highly probable that they would resort to a priori theories (because folk epistemology associates “causes” with “theory”). This effect should be decreased when it is demanded to simply report “the sensory transformation” that precedes a decision. This formulation can also be seen in many other authors (Engelbert & Carruthers, 2011; Schwitzgebel, 2008, 2011). The experimentation in this research will propose operational definitions for “sensory transformation”, appropriate for the first order task.
Secondly, (ii) the NWC presumes that even in the case of evidence of a correct description of a cognitive process, this must not be considered as a real access but rather an ad-hoc theory, one that is only accidentally correct12. Indeed, the NWC is presented as a hypothesis that cannot be rejected in principle (Smith & Miller, 1978). In subsequent developments, (Nisbett & Bellows, 1977; Nisbett & Ross, 1980) the authors argue that introspection of complex processes must be studied through an actor-observer paradigm. That is, introspection should be investigated in an actor-group (actors: individuals that evaluate the causes of their own behavior), but also in an observer-group (observers: individuals that evaluate the causes that motivate the actors behavior). Under the premise that only actors have a privileged access to their cognitive processes by introspection, this group’s report not only should be adequate (that is, agree with the real causes of behavior, known to the researcher), but also present a higher accuracy than the observer-group reports, which does not possess 11 In terms of Smith and Miller: “[…] the only difference between these two kinds of information (states and processes) is that people may in general be able to report on the first but not on the second. There is no other reason to consider one content (state) and the other process […]” (Smith & Miller, 1978, p. 360). This circularity is also present in the new version of NWC, the adaptive unconscious (Wilson, 2002).
12 In terms of the authors: “subjective reports about higher mental processes are sometimes correct, but even the instances of correct report are not due to direct introspective awareness. Instead, they are due to the incidentally correct employment of a priori causal theories” (Nisbett & Wilson, 1977, p. 233).
privileged access to such information13: Therefore, this was the reliability test for introspective data. Many authors criticized the use of between-subject designs to study introspection of cognitive processes (Kraut & Lewis, 1982; White, 1980; Wright & Rip, 1981). Smith and Miller (1978) argue that consciousness of a mental process is something that happens within the individual, therefore the evaluation of accuracy should be done on a case-by-case basis. Consequently, a hypothesis regarding consciousness of any mental content is always best evaluated with within-subject designs. In line with this suggestion, the vast majority, if not all contemporary research programs interested in introspective access consider within-subject designs (e.g., Corallo et al., 2008; Fleming et al., 2010; Marti et al., 2010; Palmer et al., 2014). Surprisingly, as far as we know, there are still no experimental studies that evaluate the NWC with such a design. This thesis is partly motivated by a desire to fill this gap in the literature.
Thirdly, (i) not only does the conceptual distinction between cognitive states and processes remain unclear, but (ii) also it is largely unknown whether the experimental context has a preponderant role in introspective inaccessibility. In effect, (iii), Nisbett and Wilson do not describe thoroughly the experimental conditions of the studies reviewed. Along these lines, the most problematic aspect14 is the response mode15, which is often based on verbal reports. The problems associated with introspective verbal reports have been extensively discussed in the literature surrounding it (Ericsson & Fox, 2011; Ericsson & Simon, 1980; Fox, Ericsson & Best, 2011; Schooler, 2002b, 2011). Many authors converge on the idea that verbal reports must involve a translation of the information of interest (Schooler, 2002a). This opens the possibility that cultural factors, a priori beliefs, personality factors or aspects of the researcher-participant relationship (demand characteristics) could alter the formation of introspective judgments16. Indeed, the problem of translation is particularly relevant in the NWC. Most, if not all, cognitive processes investigated by Nisbett and Wilson have high cognitive complexity; as a 13 In summary, “[…] if actor consistently gave more accurate reports about the reasons for their behavior than observers did, then this would indicate privileged sources of information underlying these reports. If not, then the position of NWC would be further supported […]” (Johansson et al., 2006, p. 674).
14 For instance, White (1980, 1988) suggests that most of the studies reviewed by the authors favor the automation of processing, which leads to a systematic loss of attention about the task. The introspective deficit could be due to this effect. Ericsson and Simon (1980) note that Nisbett and Wilson provided considerable background information to their participants in their experiments, so that they could have chosen to base their reports on the background information rather than on internal information.
15 Particularly important is the fact that introspective task in a most of Nisbett and Wilson’s experiments was applied not immediately after the cognitive process took place. In consequence, and under the conceptualization of introspection as retrospection, individuals when asked about the nature of their cognitive processing might simply not have such information in working memory (Bona & Silvanto, 2014). Therefore, they probably resorted to other sources of information in the elaboration of the introspective judgment (Ericsson & Simon, 1980).
16 According to Schooler: “If meta-consciousness (introspection) requires re-representing the contents of consciousness, then, as with any recording process, some information could get lost or become distorted in the translation. The likelihood of noise entering the translation process is particularly great when individuals (1) verbally reflect on inherently non-verbal experiences, and/or (2) assess ambiguous or subtle visceral signals” (Schooler, 2002a, p. 342).
consequence, they demand that participants execute high-level forms of reasoning or at least integrate many information sources (e.g., selection of the best quality product). The high complexity of the task, added to the low experimental control of introspective verbal-report, are factors that favor confabulation (Ericsson & Simon, 1980). In fact, Ericsson and Simon observed in many studies that verbalization tends to be incomplete when participants are under high cognitive load. In response, recent advances in the field of introspection of mental states (Corallo et al., 2008; Marti et al., 2010) have all been achieved by focusing on elementary cognitive tasks, and replaced verbal reports by quantified reports.
The general objection can be summarized as the absence of a model which explains under which conditions an introspective report is reliable information and under which it is the end-product of inferential reasoning. Synthesizing the three criticisms, as well as considering new paradigms in experimental introspection, our proposition consists in re-evaluating the NWC from an (i) operational redefinition of cognitive processes and states, (ii) the implementation of a within-subject design to evaluate the introspective access to complex processes and (iii) a simplification of the experimental context with the purpose of avoiding translation problems.

A new framework for the research on Introspection

Our conceptual and experimental work is guided by the idea that the perceived inaccessibility of cognitive processes to introspective reports advocated by the NWC, derives from a poor representation of the mental content of interest, and this in turn is the product of a poor control over the multiple sources of information that contribute to the formation of an introspective judgment. In summary, a poor representation of a mental state is not necessarily evidence of a lack of access (White, 1988). Therefore, our objective is two-fold. On the one hand, we will try to experimentally re-evaluate the NWC (taking into account the previous critique). On the other hand, we will try to conceptually design a general model of introspective access from recent evidence in experimental metacognition. In order to do so, we need to determine the structural and functional factors that contribute to the limits of introspective access. The first process will be based on two functional and structural properties of introspection that can be inferred from the recent advances in studies on metacognition: (i) Evidence suggests that the formation of introspective judgments results from the accumulation of information from multiple (inner) sources. In this sense, introspections are decisions, much like perceptions. (ii) There is also evidence of a high inter-individual variability in the introspective capacity.
Firstly, recent experimental evidence indicates that introspection would act as an information accumulator (Yeung & Summerfield, 2012), based on many sources of information, that may not strictly depend on the cognitive processes leading to the first order decision. For instance, Schwartz and Diaz (2014) showed that there is not a 1:1 correspondence between first order processes and the processes that drive introspection. Multiple sources of information contribute to the elaboration of a metacognitive judgment. A tip-of-the-tongue (TOT) experience (an indication of introspective judgment, as there is absolutely no external fact that can validate the feeling), would be determined by the familiarity of the stimuli, the combination and intensity of the surrounding information, the force of target activation (i.e., items of interest) by itself, among other factors. Schwartz and Diaz hypothesize that there would be multiple cognitive processes (i.e., sources of information) that underlie each introspective judgment (TOT, JOL or confidence judgments). Similarly, Lempert et al. (2015) find evidence for the dissociation between the mechanism through which individuals build an introspective judgment and first order decision, in that an increase in pupil dilatation correlates negatively with the introspection precision (confidence in decision making), independently of the difficulty during an auditory decision task. Thus, introspective judgments seem to integrate multiple sources of information, so as to create a new decision variable. Admittedly, it is unclear whether there is a single introspective mechanism or if such property requires multiple introspective focuses (Prinz, 2004) accumulating information in parallel (Merkle & van Zandt, 2006).
Moreover, the locus of the introspective decision, relative to the first order processes, is still highly controversial: is it the case that introspective evidence accumulates up until the moment of the first order decision (Decisional Locus Models (DLM): Kiani & Shadlen, 2009; Zylberberg, Barttfeld & Sigman, 2012)? or on the contrary, does it continue to integrate information after the decision (see Post-Decisional Locus Models (PDLM): Pleskac & Busemeyer, 2010; Resulaj, Kiani, Wolpert, & Shadlen, 2009)? Recently, Fleming et al. (2015) contributed evidence supporting the second alternative. The authors, through transcranial magnetic stimulation (TMS) altered the process of information accumulation, in certain cases prior to the participants’ decision to a visual discrimination task or after the decision to the same task. They showed similar effects on introspective judgments when the TMS pulse was applied before, or after the decision, suggesting that the accumulation process is extended after the participants’ response (PDLM).
Both the multiplicity of sources of information that form an introspective judgment, and the temporal extension of the accumulation process itself, allows us to suggest that participants could be able to introspectively recover information from different moments during first order processing. We hypothesize the presence of both early and late introspective integration. Thus, it could be possible to imagine that participants can flexibly activate an early recovery process of the information related to perceptual load during a cognitive task, but also a late recovery process related to information on performance in the same task. Previous authors have implicitly assumed that introspection can be guided to different points in the development of the task (Goldman, 2004). Hurlburt and Heavey (2004), in the context of experience sampling protocols, suggest the use of a marker (the beep procedure) to signal to participants the moment when they must introspect. This method would specify the cognitive stage where the hypothetical mental content of interest is generated and to which introspection must be applied. According to this view, the distinction between DLM and PDLM could be understood in terms of a flexible introspective focus, either on early or late task processes. In addition, some theorists (Schwitzgebel, 2011) have recently suggested that there is little control from researchers and participants regarding how different sources of information impact introspection. Indeed, if the focus of introspection is labile, it is also possible to imagine that “confusion of the source” may occur: participants may have difficulty in discriminating the mental content of interest from irrelevant information that would lead to confabulations. Ericsson and Simon’s (1980) think aloud procedure was precisely meant to track and follow as closely as possible the evolving content of the stream of consciousness, so that at each time point, the content of the report would exactly match the mental state. But of course, the temporal resolution of verbal discourse is poor compared to the stream of thought, notably because of the syntactic constraints in language. Therefore, in order to advance on this issue, the introspective recovery process towards certain mental content is manipulated experimentally in this research (regardless of the time when these were generated in the experimental task).
Another important constraint on introspection that has been put forward recently is the high inter-individual variability in the accuracy of introspection. These inter-individual differences have been linked to structural variability in brain regions that subserve introspection. Indeed, recent studies associate both medial and lateral regions of the anterior PFC (Baird et al., 2013; de Martino et al., 2013; Fleming et al., 2010; Fleming et al., 2012; Rounis et al., 2010; Yokoyama et al., 2010), as well as the dorsal premotor cortex (Fleming et al., 2015), with the precision of metacognition. Importantly, variations in participants’ introspective sensitivity have been shown to covary with morphological variability in these regions. The importance of these findings is that if adequate control over the introspection mechanisms is achieved, there would still be a structural modulator that determines the precision of introspection.

Experimental studies

The main aim in this thesis was to investigate whether one could have introspective access to cognitive processes, as opposed to introspective access to the cognitive states or to the behavioral consequences that they generate. The NWC model will thus be assessed; but we shall do so by means of quantified introspection over well studied and constrained simple cognitive tasks. The objective of the study has not been re-evaluated recently (however, see Petitmengin et al., 2013), as most recent advances on metacognition has focused on the mechanisms of confidence judgments. However, confidence would clearly be classified as a “cognitive state” in Nisbett & Wilson’s classification. In addition to confidence judgments (Fleming et al., 2010; Pleskac & Busemeyer, 2010), judgments of duration of perceptual decisions (Corallo et al., 2008; Marti et al., 2010; Miller, Vieweg, Kruize & McLea, 2010, Bryce & Bratzke, 2014) have also attracted some attention. In order to go beyond reports on cognitive states, a new introspective measure, the « Subjective Number of Scanned Items » (SNSI) was devised, which is applicable whenever the first order task asks for the processing of more than one item. Then, the SNSI consists in the number of items that the participant consciously processed before the decision. The methodology is applied in visual searches (e.g. “how many items did you scanned before you found the target?”) and in working memory searches (e.g. “how many items of the list did you review before deciding whether the target was present in the list?”). The hypothesis of this research is that this number is an index of participants’ access to the processes of the searches themselves.
A problematic aspect of subjective reports is that they are highly dependent on the experimental context (Goldman, 2004). Recent conceptualizations suggest that access to different mental states would require different species of introspection with different forms of measurement (Prinz, 2004). It has even been suggested that different experimental approaches (e.g., think aloud procedure: Ericsson, 2003; Ericsson & Simon, 1980; experience sampling: Hurlburt & Heavey, 2004; quantified introspection method: Corallo et al., 2008; Marti et al., 2010; script-report procedure: Jack & Roepstorff, 2002, etc.), could be associated with different levels of metacognitive access (Overgaard & Sandberg, 2012; Sandberg et al., 2010). Taking into account this introspective pluralism (Schwitzgebel, 2011), the procedure in this research (SNSI), was designed to reduce the possibility of obtaining confabulatory introspective responses: Firstly, my research is not focused on introspection of higher level processes (as are almost all experiments examined by Nisbett and Wilson (1977). The first order tasks that we study are complex in the sense that they are potentially multi-step tasks (a search might be fully serial, with examination of one item at a time). But they are not high level tasks, in the sense that the signal that contributes to the decision is precisely defined and controlled. Secondly, verbal re-coding of the experience was not required, as it has been suggested that this could favor over-interpretations (Ericsson & Fox, 2011; Fox et al., 2011; Schooler, 2002a; Schooler, 2011): SNSI reports are given on quantitative scales. Thirdly, this research procedure evaluates introspection on a trial-by-trial basis, an important aspect criticized in the studies examined by Nisbett and Wilson (White, 1988). Finally, the experimental strategy, inspired by the Jack and Roepstorff (2002) procedure, investigated whether introspection is capable to detect changes in the cognitive process strategically generated by the experimental conditions. The general prediction in these experiments was that participants’ introspection will be sensitive to such changes, suggesting that this ability can access the complexity of the cognitive process deployed in each case.
In the first paper, the investigation focused on whether introspection can access and distinguish two different cognitive processes that underlie two kinds of visual searches. A series of experiments were designed which contrasted “pop-out” and “serial” visual searches. The strategy was to evaluate whether introspection could track whether the search had been serial or parallel. The results showed that participants were able to introspect the search processes, provided that they had enough time to deploy their introspection on the decision. In addition, indications showed that participants were in fact reporting how many attentional switches they performed before one decision.
In the second paper, the hypothesis that participants had access to the attentional guidance during visual searches was specifically tested. In order to do so, a brief cue was introduced, unbeknownst to participants, before the search array, so as to manipulate exogenously their attention. Importantly, our cue was so subtle that participants did not spontaneously notice its presence. The important point here is that it is possible to avoid the impact of potential contaminants related to the experimental setting on the introspective judgments. As participants did not know that there was a cue, attentional guidance was manipulated during the search, while participants’ perception was not affected. The findings demonstrated that while participants were unaware of the cause of the modification of their search processes, their introspective reports showed that they accurately access the processes themselves.

Table of contents :

1. Introduction 
1.1. Historical context
1.1.1. Before Cognitivism
1.1.2. Cognitivism
1.1.3. Neuro-Cognitivism
1.2. Introspection of high-order cognitive processes
1.2.1. The Nisbett and Wilson Canon (NWC)
1.2.2. A new framework for the research on Introspection
1.3. Experimental studies
2. Experimental Studies 
2.1. Introspection during visual search
2.1.1. Introduction
2.1.2. Results
2.1.3. Discussion
2.1.4. Paper
2.1.4.1. Experiment 1
2.1.4.2. Experiment 2
2.1.4.3. Experiment 3
2.1.4.4. Experiment 4a-b
2.1.4.5. Conclusion
2.2. Introspective access to an implicit shift of attention
2.2.1. Introduction
2.2.2. Results
2.2.3. Discussion
2.2.4. Paper
2.2.4.1. Experiment
2.2.4.2. Conclusion
2.3. Introspection during working memory scanning
2.3.1. Introduction
2.3.2. Results
2.3.3. Discussion
2.3.4. Paper
2.3.4.1. Experiment 1
2.3.4.2. Experiment 2
2.3.4.3. Conclusion
2.4. Self-knowledge dim-out: Stress impairs metacognitive accuracy
2.4.1. Introduction
2.4.2. Results
2.4.3. Discussion
2.4.4. Paper
2.4.4.1. Experiment
2.4.4.2. Conclusion
2.4.4.3. Pilot Experiment

GET THE COMPLETE PROJECT