An old problem: How can we distinguish between conscious and unconscious knowledge acquired in an implicit learning task?

Abstract

A long lasting debate in the field of implicit learning is whether participants can learn without acquiring conscious knowledge. One crucial problem is that no clear criterion exists allowing to identify participants who possess explicit knowledge. Here, we propose a method to diagnose during a serial reaction time task those participants who acquire conscious knowledge. We first validated this method by using Stroop-like material during training. Then we assessed participants’ knowledge with the Inclusion/Exclusion task (Experiment 1) and the wagering task (Experiment 2). Both experiments confirmed that for participants diagnosed as having acquired conscious knowledge about the underlying sequence the Stroop congruency effect disappeared, whereas for participants not diagnosed as possessing conscious knowledge it only slightly decreased. In addition, both experiments revealed that only participants diagnosed as conscious were able to strategically use their acquired knowledge. Thus, our method allows to reliably distinguish between participants with and without conscious knowledge.

Introduction

In the last three decades, this question regarding the existence of learning without awareness is one of the hottest debated topics in the field of implicit learning. Basically, the term ‘implicit learning’ refers to learning without acquiring (verbally) expressible knowledge. In tasks like, for instance, the serial reaction time task (SRT task, Nissen & Bullemer, 1987), participants see several marked positions on the computer screen. Each position is associated with one certain key. Participants’ task is to react at fast as possible with the respective key whenever an asterisk occurs at the marked screen-positions. Unbeknownst to the participants the locations of the asterisk follow an underlying regular sequence. After several blocks, participants receive a transfer block without further announcement. In this block, the asterisk appears randomly on the screen or follows a second, different sequence. If participants have learnt something about the regular sequence their response speed should decrease in this transfer block. Consequently, the deceleration of response times indicates learning of the sequence. After training, participants’ knowledge is assessed. The crucial finding is that even though participants show a decrease in response speed in the transfer block, the post-experimental knowledge tests do not reflect any knowledge. Many researchers interpret this dissociation between performance and expressible knowledge as an indicator that task performance and expressible knowledge rely on different learning systems, an implicit and an explicit system (e.g., Cohen et al., 1990, Frensch, 1994, Frensch, 1998, Frensch and Rünger, 2003, Haider and Frensch, 2005, Haider and Frensch, 2009, Keele et al., 2003, Nissen and Bullemer, 1987, Rünger and Frensch, 2008, Willingham, 1998).

However, albeit this dissociation between task performance and expressible knowledge has been replicated in thousands of experiments, a large debate regarding this dissociation has emerged and still goes on. The reasons are twofold: As elaborated below, the older and more methodological debate doubts whether this dissociation reflects any functional dissociation between different learning or memory systems. The more recent debate about the existence of implicit learning processes concerns the conceptualization of consciousness.

With regard to the methodological reasons the main points of critique are that the two methods used to assess participants’ knowledge differ in (a) their sensitivity or decision criteria (e.g., Shanks and St. John, 1994, Snodgrass, 2004), (b) in their exhaustiveness (e.g., Shanks & St. John, 1994), and/or (c) in their reliability (e.g., Buchner and Brandt, 2003, Buchner and Wippich, 2000, Meier and Perrig, 2000). In short, the first point refers to the problem that differences in the amount of knowledge assessed during task performance, on the hand, and in knowledge tests, on the other hand, might be due to lower sensitivity of post-experimental knowledge tests. The second point is related to the problem that differences in the amount of knowledge can occur when participants are not willing to express in a knowledge test all knowledge they possess, for instance, because they are afraid to make an error. According to the third point, a dissociation between the amount of knowledge assessed during training and that assessed in post-experimental knowledge tests might result from differences in reliability between these tests.

Overall, these arguments have challenged the assumption that the dissociation between performance and post-experimentally assessed knowledge reflects any functional dissociation between different learning or memory systems. Moreover, these arguments are hard or even impossible to disprove. And, they now occur whenever a researcher tries to provide evidence for task processing having taken place without conscious awareness of the stimuli (e.g., Bechara, Damasio, & Damasio, 2000 versus Hannula et al., 2005, Maia and McClelland, 2004, for examples from other fields).

Consequently, researchers have to consult all their creativity in order to develop new methods which are assumed to be better suited to assess potential explicit or conscious knowledge. And even if they develop new methods, this does not at all convince other researchers (see, for example the most recent discussion concerning the wagering task of Persaud, McLeod, and Cowey (2007; and the critiques of e.g., Clifford et al., 2008, Dienes and Seth, 2010, Seth, 2008). Thus, it seems as if research in the field of implicit learning (or even unconscious processing, at all) has run in a methodological impasse.

The second more recent debate about the existence of implicit learning processes concerns the conceptualization of consciousness. Due to the huge complexity of this phenomenon, it is hard to find any agreement among researchers with regard to a definition of consciousness. This leads to the second problem in the field of implicit learning: Evidence of implicit knowledge depends on the respective method to assess the knowledge which itself is inextricably connected to the particular conceptualization of consciousness (see, e.g., Frensch and Rünger, 2003, Seth et al., 2008, for reviews). As long as researchers do not agree with regard to the crucial question of how to best define consciousness, the main question whether implicit learning or implicit task processing is a basic phenomenon cannot be solved. Thus, beyond the methodological impasse a theoretical impasse seems to exist as well.

A legitimate claim put forward by, for instance, Seth (2008) therefore is to measure consciousness in implicit learning tasks with more than one test in order to cope with the complexity of the phenomenon of consciousness (or unconsciousness). Accordingly, our goal, here, is to show that the development of consciousness within an implicit learning task leads to qualitative differences in task performance which can be diagnosed during training. This online diagnosis of consciousness then provides an external criterion allowing to test for analogous qualitative differences in post-experimental knowledge tests.

For this purpose, we start by adopting a perspective on consciousness that defines consciousness in terms of the generation of a meta-representation, as is assumed, for instance, in Higher-order Thought theories (e.g., Dienes & Perner, 1999) or in the Global Workspace theory (e.g., Baars, 2003, Dehaene et al., 2006). Global Workspace theory assumes that consciousness requires that information which is otherwise encapsulated in separate modules becomes available through this workspace and as such constitutes subjective experience (e.g. Dehaene & Naccache, 2001). Likewise, in Higher-order Thought theories consciousness about a fact in the world refers to a state when someone knows that he/she knows. That means a person having such a conscious representation is able to answer questions about his/her state of knowledge. By contrast, not knowing that one knows implies that a person in such a state of pre-consciousness can only guess, or has at least the subjective feeling that he/she is guessing, even when the answer is correct (e.g., Dienes, 2008; for a more elaborated review of this position, see Rünger & Frensch, 2010).

Importantly, this perspective implies that consciousness is not simply an epiphenomenon with ‘unconscious’ task processing reflecting a weaker form of the same kind of representation. Rather, it assumes that conscious and unconscious task processing differ functionally, in the sense that only conscious representations of facts in the world enables someone to develop or pursue long-lasting goals concerning these facts. This, in turn, provides the possibility for voluntary control of (or top-down triggered) performance (e.g., Baars, 2003).

Accordingly, we suspect that when a participant develops a conscious (meta-) representation of the underlying sequence within a SRT task, he/she probably will rely on this representation and, thus, will be able to switch from stimulus-driven to a top-down triggered task processing. If so, it should be possible to diagnose this qualitative change in performance. Thus, the diagnosis of such a qualitative change during training should enable us to distinguish between participants who develop conscious knowledge during training and those who do not.