Content uploaded by Olivier Corneille
Author content
All content in this area was uploaded by Olivier Corneille on Jul 17, 2020
Content may be subject to copyright.
Behavioral and Physiological Evidence Challenges the Automatic Acquisition of
Evaluations.
Olivier Corneille, UCLouvain.
Gaëtan Mertens, Tilburg University & Utrecht University.
Manuscript in press in Current Directions in Psychological Science
Olivier Corneille (Contact author). Psychological Sciences Research Institute, UCLouvain, 10
Place du Cardinal Mercier, 1348 Louvain-la-Neuve, Belgium. Email:
olivier.corneille@uclouvain.be
2
Behavioral and Physiological Evidence Challenges the Automatic Acquisition of
Evaluations.
1. The Automatic Acquisition of Evaluations.
An influential view pervading psychological research (e.g., in social, consumer,
health, and clinical psychology) is that evaluations can be acquired through an
associative/affective learning mode that automatically registers mere stimuli co-occurrences
encountered in the environment. In attitude research, associative learning is thought to allow
for the unconscious, efficient, involuntary, and uncontrollable formation of evaluations and to
be insensitive to the relational meaning of stimuli co-occurrences (e.g., Gawronski &
Bodenhausen, 2014). Likewise, influential theories of fear learning posit that fear can be
automatically acquired (i.e., without consciousness, effort, or intention, and uncontrollably),
resulting in non-conscious and unqualified associative representations that automatically elicit
fear when activated (e.g., LeDoux & Pine, 2016; Öhman & Mineka, 2001).
This automatic learning view implies a high susceptibility of individuals to social and
environmental influences. For instance, people may fall prey to a negative political
advertisement if the information it contains creates, uncontrollably or unconsciously, a
negative evaluation of the targeted candidate. It also often assumes a difficulty to change
automatic evaluative responses based on mere verbal information. For instance, a fear of dogs
may be difficult to change using verbal instruction if the fear response reflects the automatic
activation of an inaccessible or unconscious mental association between a dog and having
been bitten.
These questions have been mostly investigated using conditioning procedures. In
attitude research, a neutral stimulus (the conditioned stimulus, or the CS; e.g., a neutral face)
is paired with a positive or negative stimulus (the unconditioned stimulus, or the US: e.g., a
pleasant or unpleasant sound or picture). An evaluative conditioning effect is said to occur
3
when the evaluation of the CS changes after the CS-US pairing. In fear-conditioning research,
typically aversive USs (e.g., an electric shock, a loud noise) are used, and subjective,
behavioral, and physiological responses related to fear and arousal (e.g., distress ratings,
avoidance responses, skin conductance responses) are measured (e.g., Lipp, 2006). Evaluative
and fear conditioning procedures involve a simple associative procedure (i.e., pairing CSs
with USs) that is considered ideally suited to the investigation of associative/affective
learning.
2. Testing the Automatic Learning of Evaluations Using “Explicit”, “Implicit” and
Physiological Measures.
The automatic learning of evaluations has been tested using “explicit”, “implicit”, and
physiological measures. Explicit measures are self-reported evaluations, such as good/bad
judgments or direct scale ratings about an attitude object. Implicit measures are less clearly
defined (see Corneille & Hütter, 2020), but generally rely on indirect behavioral tasks that
reduce participants’ control and deliberation during measurement. The Implicit Association
Test is considered the gold standard of “implicit” evaluative measures. In this task,
respondents may sort faster positive stimuli alongside category A (e.g., thin people) and
negative stimuli alongside category B (e.g., obese people). The extent to which they do so (or
the opposite) indicates their implicit preference for one category over the other. Measurement
outcomes of this sort are thought to reflect biased mental associations, commonly referred to
as “implicit biases” or “unconscious biases.” Physiological measures record participants’
physiological responses to stimuli. Skin conductance responses and potentiation of the startle-
reflex are examples of such measures, and they capture a mix of arousal and evaluation
(Lonsdorf et al., 2017). Whereas “explicit” and “implicit” evaluative measures are frequently
used in social psychological, social cognition, and consumer research, physiological measures
are more common in clinical and neuroscience research.
4
Dual-learning theories assume that less controllable and deliberate measures of
evaluations (i.e., “implicit” and physiological measures) provide a better window into
associative/affective learning (e.g., Gawronski & Bodenhausen, 2014; Ledoux & Pine, 2016).
This is because these measures reduce the influence of deliberate and controlled post-learning
processes on task performance.
3. Growing Evidence Challenges the Automatic Acquisition of Evaluations on all
Categories of Measures.
Research on evaluative conditioning has largely failed to obtain conclusive evidence
for automatic evaluative learning (for a comprehensive review, see Corneille & Stahl, 2019).
No evaluative learning effect is observed in procedures preventing a conscious encoding of
the stimuli, such as when using short-timed, parafoveal, or visually suppressed presentations.
Evaluative learning is also resource demanding. It is fully disrupted when participants are
engaged in a concurrent task during the stimuli presentation. Finally, research indicates that
evaluative learning is sensitive to participants’ processing goals. For instance, evaluative
learning effects are largely weakened when participants are distracted from processing the
affective quality of the information.
Of importance too, evidence that may be considered supportive of automatic
evaluative learning is unrelated to “implicit” versus “explicit” measurement. The mere
exposure effect, often considered the hallmark of “preferences-that-need-no-inferences”, is
typically demonstrated on self-reported evaluations (e.g., “Which of these two stimuli do you
prefer?”) Likewise, preliminary evidence for uncontrollable evaluative learning (for instance,
participants presumably unable not to start liking a stimulus paired with positive information)
has been best observed on “explicit” liking measures (Hütter & Sweldens, 2018). Hence,
contrary to a widespread dual-process view, that an evaluation is observed on an “implicit
5
task” or on a self-reported measure is unrelated to how it was acquired in the first place
(Corneille & Hütter, 2020).
Besides the automaticity of evaluative learning, dual-learning models of evaluations
posit that associative knowledge captures raw associations between events, unqualified by the
relational meaning of these associations. For instance, people may harbor negative
associations about a pain-relieving medication because of its co-occurrence with pain (i.e., the
raw contingency), despite the medication relieving the pain (i.e., the relational implication of
the medication). Preliminary evidence supporting the possibility of such unqualified
representations or processes, however, has again been found on “explicit” (e.g., Kükken,
Hütter & Holland, 2019) measures, and these effects are highly sensitive to task structure
when measured with “implicit” tasks (e.g., Bading, Stahl, & Rothermund, 2020).
As it appears, differences in performance on “explicit” vs. “implicit” tasks cannot be
univocally interpreted in terms of dissociations in learning modes. Post-learning processes
may be responsible for these differences. For instance, a smaller amount of information is
likely retrieved from memory when completing a speeded (e.g., an Implicit Association Test)
than a non-speeded (e.g., an evaluative rating) task. Hence, divergences in outcomes on
“implicit” versus “explicit” tasks may reflect divergences in retrieval, despite originating in a
unitary learning process that is neither unconscious nor immune to verbal information. As a
matter of fact, “implicit” attitudes can be created in a snap based on mere verbal information
(e.g., De Houwer, 2009). And this is also true when it comes to changing evaluations - for
instance, merely telling participants that Gandhi prevented his wife from taking a medical
treatment that could have saved her life results in a quick reversal of scores on “implicit”
measures (Van Dessel, Ye, & De Houwer, 2018). This runs counter a prevalent view that sees
“implicit biases” as the result of unconsciously and slowly acquired associations.
6
So far, we have discussed behavioral measures. As explained above, physiological
measures may provide a more sensitive test for the existence of a distinct learning mode
grounded in more affective systems. Here too, however, recent research questions automatic
learning and supports the sensitivity of physiological indicators of evaluations to verbal
instruction (e.g., Mertens et al., 2018). For instance, a recent meta-analysis has concluded in a
weak fear acquisition effect, plausibly driven by publication bias and methodological
artifacts, when using subliminal stimuli (see Mertens & Engelhard, 2020). And, similar to
explicit and implicit evaluative measures, fear responses can also be modulated by verbal
instruction (Atlas, 2019; Mertens et al., 2018). For example, conditioned physiological
responses can be rapidly abolished when participants are told that no more USs will be
delivered (Luck & Lipp, 2016). It is telling that these effects have also been demonstrated
using the startle-reflex (Luck & Lipp, 2015; Mertens & De Houwer, 2016), a basic defensive
reflex evident within 20-120 ms (Blumenthal et al., 2005) that is considered an index of
amygdala activity (Hamm & Weike, 2005).
4. Implications for theory and practice.
Implications of these findings for psychological theories are straightforward. There is
less evidence for and more evidence against the automatic learning of evaluations than one
may think. There is also more evidence than is commonly assumed for the role of verbal
information in the acquisition and change of evaluations. Alternative single learning models
should, therefore, be considered a more parsimonious alternative to dual-learning models.
Examples of such models are propositional models (De Houwer, 2009), goal-directed models
(e.g., Boddez, Moors, Mertens, & De Houwer, 2020) or retrieval-based models (e.g., Stahl &
Aust, 2018). Perhaps even more important are the practical implications of these recent
findings, which we now briefly discuss in the social, consumer, and health and clinical
psychology domains.
7
Social psychology: An influential social-psychological view that found its way to the
general public, and that is currently inspiring large-scale social interventions, is that “mental
associations” automatically formed about social groups (coined “unconscious bias” or
“implicit bias”) influence people’s judgment and behavior. Drawing on automatic evaluative
learning may facilitate people’s acceptance of their prejudice. Unfortunately, this may also
normalize prejudice (Eberhardt & Banks, 2019) and undermine accountability for social
discrimination (Daumeyer, Onyeador, Brown, & Richeson, 2019). That is, if “mental
associations” about social groups build up unconsciously and uncontrollably, one may
consider that people are not so much responsible for holding them.
As discussed here, however, the evidence is scarce that evaluations may be acquired
automatically. Likewise, it is unclear whether “implicit” measures assess unconscious
evaluations (e.g., Gawronski, 2019). As provocative as this may seem, even assuming
“implicit” tasks do assess unconscious and automatically acquired associations, we do not
even know whether they have any behavioral impact at all. Because outcomes on “implicit”
measures can be neither coherently nor distinctly induced, their causal role remains
unestablished (for a recent discussion, see Corneille & Hütter, 2020).
Again, this conclusion runs against a popular view holding that “unconscious bias
infiltrates every arena of life” and “can be just as devastating as the harm done by explicit
racism, sexism or homophobia » (Eberhardt & Banks, 2019). Admittedly, “unconscious bias”
may be understood here in terms of automatically enacted behaviors as opposed to
automatically acquired and unconsciously held “mental associations” presumably underlying
these behaviors (e.g., Gawronski, Ledgerwood & Eastwick, in press). However, this
alternative understanding is uncommon, and it would be highly problematic if the
“unconscious bias” construct were applied both to the cause and to its consequences.
8
Similar conclusions apply to personality research that has theorized discrepancies in
so-called “implicit” and “explicit” self-esteem. If “implicit” measures of the self are not
diagnostic of distinct learning processes and representations (see also Schimmack, 2019), one
is left to wonder how discrepancies in these measures should be interpreted, as well as how
and why “implicit” self-esteem should be changed at all.
Problematic theorization on the “implicit” construct is widespread in social
psychological research and this can be consequential (for a comprehensive discussion, see
Corneille & Hütter, 2020). During the Covid-19 outbreak, the Society for Personality and
Social Psychology featured research inviting people to indulge in comfort food to feed
“primitive and implicit feelings of being cared for and loved” (SPSP, 2020). This surprising
recommendation (literally: eat high-caloric food to please your unconscious inner self) may
contribute to another epidemic: the obesity one.
Consumer psychology: Consumers may be upset about the use of subliminal
conditioning (such as when briefly pairing Al Gore with “Rats” in a political advertisement).
In all likelihood, however, subliminal conditioning is largely ineffective for changing
people’s evaluations (see Corneille & Stahl, 2019). Ironically, people generally feel at ease
with blatant social influence techniques (e.g., a presidential candidate featured against the
American flag) that are much more influential. This is not to say that automatic processes,
including unconscious ones, play no role in consumer or social behavior. They certainly do.
Yet, as we have already pointed out, many of these effects may occur at a post-learning stage.
Consistent with associative/affective learning views, companies have relied on
“implicit” measures to assess “unconscious” preferences in their consumers. The company
Pizza Hut, for instance, has used eye-tracking technology to assess unconscious pizza topping
preferences in consumers. There is, however, no indication that eye-tracking is relevant to the
study of “unconsciously” acquired or “unconsciously” held pizza topping preferences (not to
9
speak of whether eye-tracking provides a valid measure of preferences at all). And, even if it
was, there is no evidence either that this measure of preference would allow for tastier
experiences. It is largely assumed, when using “implicit” measures, that testers know more
about respondents than respondents know about themselves. This, however, is a
philosophically and empirically tricky assumption. And, if it is incorrect, it may have ethically
questionable consequences (e.g. feeding consumers with pizza topping they don’t like,
persuading individuals that they hold conflicting conscious and unconscious evaluations about
themselves and about others).
Health and clinical psychology: That fear responding is acquired and may be changed
on the basis of mere verbal instructions could be inspirational for new clinical interventions.
For instance, in therapy, patients can be asked to formulate their expectations (e.g., if I walk
in the park, a dog will attack me) and asked how these expectations may be tested (e.g., take a
walk in the park during a busy afternoon). This approach can help identify patients’
idiosyncratic expectations and allow for more targeted exposure therapy. Interventions of this
sort have been successfully developed in cognitive-behavioral therapies (Hofmann, 2008).
Surprisingly enough, research on these types of interventions remains surprisingly scarce in
fundamental research on fear conditioning (see Carpenter, Pinaire, & Hofmann, 2019). A
broader appreciation for the role of consciously accessible representations could lead to a
more widespread application of such cognitive interventions to challenge dysfunctional
beliefs about contingencies in the world.
5. Clarification and limitations.
We want to emphasize that we fully endorse the view that automatic processes
influence people’s judgments, and behaviors. Clearly, for instance, people may show efficient
behavioral and physiological responses when exposed to disliked stimuli. Likewise, it would
be unreasonable to assume that people are constantly aware of the determinants of their
10
perceptions, judgments and behaviors. Instead, we are pointing to a growing body of
empirical evidence challenging the view that these responses are automatically learned and
are insensitive to verbal information. We also question the view that some measures (e.g.,
“implicit” and physiological measures) would best indicate an associative/affective learning
mode or qualitatively different representations. Although our analysis was mainly based on
conditioning procedures (for the reasons we have explained), similar conclusions have been
reached in related paradigms, such as approach-avoidance and the mere exposure effect (for
an in-depth discussion, see Corneille & Stahl, 2019).
Our analysis did not include studies with brain imaging techniques and brain-lesioned
populations. With regard to studies on brain-lesions and process-specific impairments, some
studies have shown that region-specific lesions (e.g., amygdala damage) result in process-
specific impairments (e.g., successful acquisition of fear conditioning with skin conductance
responses, but no acquisition of declarative knowledge), while the reverse being true for
damage in another region (i.e., hippocampal damage and unsuccessful fear conditioning, but
the successful acquisition of declarative knowledge; Bechara et al., 1995).
Such findings may be seen as providing strong evidence for dual-learning theories of
evaluations. However, other researchers have pointed out that double dissociations of this sort
do not necessarily need to be interpreted as evidence for modularity and independent
processes (e.g., Plaut, 1995). Furthermore, conflicting findings have been reported in the
literature. For instance, Coppens et al. (2009) reported that patients with unilateral amygdala
damage can in fact acquire conditioned fear responses after acquiring explicit knowledge of
CS-US contingencies.
With regard to brain imaging studies, we believe that brain activation should not be
considered a direct measure of attitude or fear. Instead, brain imaging data reflect a neural
implementation level of analysis and can be logically consistent with different models of how
11
evaluations are acquired, retrieved, and expressed. Additionally, it can be noted that
associative mental representations are not necessarily neuro-anatomically, -chemically, or –
biologically more plausible than non-associative (e.g., propositional) representations (e.g.,
Langille & Gallistel, 2020). Given their different levels of analysis, however, psychological
and neurocognitive evidence can be mutually informative to approximate a more accurate
psychological and neurobiological model of how evaluations and fears are acquired and can
be changed (e.g., Amodio & Ratner, 2011).
6. Conclusion
A growing body of research challenges the existence of an associative/affective formation of
attitudes and fears when this learning mode is defined as automatic and impervious to verbal
information. As discussed in this article, the acquisition of evaluations and fear is much less
automatic than often assumed, and both can be modulated using verbal instruction, even when
using “implicit” and physiological measures. Acknowledging more broadly this recent
evidence is important for both theory and practice. The unconscious and resource-free
learning of attitudes and fears should be questioned. In addition, the potential of verbal
instructions for changing “deep-rooted” evaluations offers a promising avenue for social and
clinical interventions.
12
References:
Amodio, D. M., & Ratner, K. G. (2011). A memory systems model of implicit social
cognition. Current Directions in Psychological Science, 20, 143-148.
Atlas, L. Y. (2019). How instructions shape aversive learning: higher order knowledge,
reversal learning, and the role of the amygdala. Current Opinion in Behavioral
Sciences, 26, 121-129.
Bading, K., Stahl, C., & Rothermund, K. (2020). Why a standard IAT effect cannot provide
evidence for association formation: The role of similarity construction. Cognition and
Emotion, 34(1), 128-143. doi: 10.1080/02699931.2019.1604322.
Bechara, A., Tranel, D., Damasio, H., Adolphs, R., Rockland, C., & Damasio, A. R. (1995).
Double dissociation of conditioning and declarative knowledge relative to the amygdala
and hippocampus in humans. Science, 269(5227), 1115-1118.
Blumenthal, T. D., Cuthbert, B. N., Filion, D. L., Hackley, S., Lipp, O. V., & Van Boxtel, A.
(2005). Committee report: Guidelines for human startle eyeblink electromyographic
studies. Psychophysiology, 42(1), 1-15.
Boddez, Y., Moors, A., Mertens, G., & De Houwer, J. (2020). Tackling fear: Beyond
associative memory activation as the only determinant of fear responding. Neuroscience
& Biobehavioral Reviews.
Carpenter, J. K., Pinaire, M., & Hofmann, S. G. (2019). from extinction learning to anxiety
treatment: Mind the gap. Brain sciences, 9(7), 164.
Coppens, E., Spruyt, A., Vandenbulcke, M., Van Paesschen, W., & Vansteenwegen, D.
(2009). Classically conditioned fear responses are preserved following unilateral
temporal lobectomy in humans when concurrent US-expectancy ratings are
used. Neuropsychologia, 47(12), 2496-2503.
Corneille, O., & Hütter, M. (2020). Implicit? What do you mean? A comprehensive review of
the implicitness construct in attitude research. Personality and Social Psychology Review.
Corneille, O., & Stahl, C. (2019). Associative attitude learning: A closer look at evidence and
how it relates to attitude models. Personality and Social Psychology Review, 23(2), 161-
189.
13
Daumeyer, N. M., Onyeador, I. N., Brown, X., & Richeson, J. A. (2019). Consequences of
attributing discrimination to implicit vs. explicit bias. Journal of Experimental Social
Psychology, 84, 103812.
De Houwer, J. (2009). The propositional approach to associative learning as an alternative for
association formation models. Learning & Behavior, 37(1), 1-20.
Eberhardt, J.L, Banks, S. (2019; July 12). Implicit bias puts lives in jeopardy. Can mandatory
training reduce the risk? Los Angeles Times. Retrieved from https://www.latimes.com
Gawronski, B. (2019). Six lessons for a cogent science of implicit bias and its criticism.
Perspectives on Psychological Science, 14, 574-595.
Gawronski, B., & Bodenhausen, G. V. (2014). The associative-propositional evaluation
model: Operating principles and operating conditions of evaluation. Dual-process
theories of the social mind, 188-203.
Gawronski, B., Ledgerwood, A., & Esatwick, P. (in press). Implicit bias and anti-
discrimination policy. Policy Insights from the Behavioral and Brain Sciences.
Hamm, A. O., & Weike, A. I. (2005). The neuropsychology of fear learning and fear
regulation. International Journal of Psychophysiology, 57(1), 5-14.
Hofmann, S. G. (2008). Cognitive processes during fear acquisition and extinction in animals
and humans: Implications for exposure therapy of anxiety disorders. Clinical Psychology
Review, 28(2), 199-210.
Hütter, M., & Sweldens, S. (2018). Dissociating controllable and uncontrollable effects of
affective stimuli on attitudes and consumption. Journal of Consumer Research, 45, 320-
349.
Kukken, N., Hütter, M., & Holland, R. W. (2019). Are there two independent evaluative
conditioning effects in relational paradigms? Dissociating the effects of CS-US pairings
and their meaning. Cognition and Emotion, 34, 1-20.
Langille, J., & Gallistel, C. (2020). Locating the engram: Should we look for plastic synapses
or information-storing molecules? Neurobiology of Learning and Memory, 169.
LeDoux, J. E., & Pine, D. S. (2016). Using neuroscience to help understand fear and anxiety:
a two-system framework. American Journal of Psychiatry, 173, 1083-1093.
Lipp, O. V. (2006). Human Fear Learning: Contemporary Procedures and Measurement.
Lonsdorf, T. B., Menz, M. M., Andreatta, M., Fullana, M. A., Golkar, A., Haaker, J., ... &
Drexler, S. M. (2017). Don’t fear ‘fear conditioning’: Methodological considerations for
14
the design and analysis of studies on human fear acquisition, extinction, and return of
fear. Neuroscience & Biobehavioral Reviews, 77, 247-285.
Luck, C. C., & Lipp, O. V. (2015). A potential pathway to the relapse of fear? Conditioned
negative stimulus evaluation (but not physiological responding) resists instructed
extinction. Behaviour Research and Therapy, 66, 18-31.
Luck, C. C., & Lipp, O. V. (2016). Instructed extinction in human fear conditioning: History,
recent developments, and future directions. Australian Journal of Psychology, 68(3), 209-
227.
Mertens, G., & De Houwer, J. (2016). Potentiation of the startle reflex is in line with
contingency reversal instructions rather than the conditioning history. Biological
Psychology, 113, 91-99.
Mertens, G., Boddez, Y., Sevenster, D., Engelhard, I. M., & De Houwer, J. (2018). A review
on the effects of verbal instructions in human fear conditioning: Empirical findings,
theoretical considerations, and future directions. Biological Psychology, 137, 49-64.
Mertens, G., & Engelhard, I. M. (2020). A systematic review and meta-analysis of the
evidence for unaware fear conditioning. Neuroscience & Biobehavioral Reviews, 108,
254-266.
Öhman, A., & Mineka, S. (2001). Fears, phobias, and preparedness: toward an evolved
module of fear and fear learning. Psychological Review, 108(3), 483.
Plaut, D. C. (1995). Double dissociation without modularity: Evidence from connectionist
neuropsychology. Journal of Clinical and Experimental Neuropsychology, 17(2), 291-
321.
Schimmack, U. (2019). The implicit association test: a method in search of a construct.
Perspectives on Psychological Science, 1745691619863798.
Stahl, C., & Aust, F. (2018). Evaluative conditioning as memory-based judgment. Social
Psychological Bulletin, 13(2), 1-30.
SPSP (2020, March 25)."Preparing and eating [comfort] foods activates a primitive and
implicit feeling of being cared for and loved”. [Tweet]
https://twitter.com/SPSPnews/status/1242837564156936194?s=20
Van Dessel, P., Ye, Y., & De Houwer, J. (2018). Changing Deep-Rooted Implicit Evaluation
in the Blink of an Eye: Negative Verbal Information Shifts Automatic Liking of Gandhi.
Social Psychological and Personality Science, 10, 266-273.
15
Recommended readings:
Corneille, O., & Hütter, M. (2020). Implicit? What do you mean? A comprehensive review of
the implicitness construct in attitude research. Personality and Social Psychology Review.
Corneille, O., & Stahl, C. (2019). Associative attitude learning: A closer look at evidence and
how it relates to attitude models. Personality and Social Psychology Review, 23(2), 161-
189.
De Houwer, J. (2009). The propositional approach to associative learning as an alternative for
association formation models. Learning & Behavior, 37(1), 1-20.
Gawronski, B. (2019). Six lessons for a cogent science of implicit bias and its criticism.
Perspectives on Psychological Science, 14, 574-595.
Mertens, G., & Engelhard, I. M. (2020). A systematic review and meta-analysis of the
evidence for unaware fear conditioning. Neuroscience & Biobehavioral Reviews, 108,
254-266.
