The Explicit and the Implicit: Unconscious Mental States

Automatic processes are, at least in traditional views, unconscious, but the mental states they operate on, and generate, are generally held to be conscious. That is to say, with the exception of preconscious processing (which generates unconscious, automatic processes by definition), conscious mental states (percepts, memories, thoughts, etc.) evoke unconscious automatic (or, alternatively, conscious controlled processes), which in turn generate other conscious mental states. This raises the question of whether mental contents -- i.e., percepts memories, thoughts, and the like -- can be unconscious in any meaningful way. That is, can percepts, memories, etc. influence the subject's ongoing experience, thought, and action in the absence of conscious awareness of these contents?

Recently, the philosopher Ned Block distinguished between two kinds of consciousness:

Phenomenal consciousness, or P-consciousness,referring to our experiential states.
Access consciousness, or A-consciousness, in which information interacts with experiential states.

Block argued that you can have P-consciousness without A-consciousness, as when we notice background noise without letting it interfere with what we are doing. And we can have A-consciousness without P-consciousness, in which information can influence our conscious states outside of phenomenal awareness, as in what is called blindsight (about which more below).

The idea that we can have A-consciousness without P-consciousness is generally attributed to Sigmund Freud and his theory of psychoanalysis, in which people are influenced by beliefs, feelings, and motives that are rendered unconscious by means of repression. But it's older than that.

Freud is also commonly credited with the iceberg metaphor for the unconscious: just as only about 1/8Ernest Hemingway, in Death in the Afternoon (1932), of an iceberg is visible above the surface of the water, so consciousness makes up only a small part of mental life. But this attribution is incorrect: the word "iceberg" appears nowhere in the the Concordance to the Standard Edition of the Complete Psychological Works of Sigmund Freud. The iceberg metaphor may have been invented by someone else, but it wasn't Freud.

But the iceberg metaphor does appear in different contexts.

Ernest Hemingway, wrote in Death in the Afternoon (1932), that "If a writer of prose knows enough about what he is writing about e may omit things that he knows and the reader, if the writer is writing truly enough, will have a feeling of those things as strongly as though the writer had stated them. The dignity of movement of an iceberg is due to only one-eighth of it being above water.

Later, in an interview on "The Art of Fiction" published in the Paris Review (1958), Hemingway used the iceberg metaphor again in a discussion of his
Theory of Omission, referring to his novella, The Old Man and the Sea. "Anything you know you can eliminate and it only strengthens your iceberg.... I've seen a school (or pod) of more than fifty sperm whales in that same stretch of water and once harpooned one nearly sixty feet in length and lost him. So I left that out. All the stories I know from the fishing village I leave out. But the knowledge is what makes the underwater part of the iceberg."

In Beyond Culture (1976), Edward T. Hall developed the iceberg theory of culture: observable behavior, in his view, was only a very small part of culture. The task for the anthropologist, in Halal's view, was to see below the surface to identify a culture's core ideas, beliefs, and attitudes..

Even before there was a scientific psychology, the philosopher Immanuel Kant put it concisely:

to have ideas, and yet not be conscious of them -- there seems to be a contradiction in that.... Nevertheless, we may become aware indirectly that we have an idea, although we be not directly cognizant of the same.

At the same time, there is a genuine problem with even considering the possibility of unconscious mental states. As William James put it:

The distinction...between the unconscious and the conscious being of the mental state... is the sovereign means for believing what one likes in psychology, and of turning what might become a science into a tumbling-ground for whimsies.

So that's the problem. It's hard enough to study conscious mental states scientifically. It's going to be even harder to study unconscious mental states -- to convince people that we have thoughts, feelings, and desires that we we don't know we have, because we're not aware of them.

But wait a minute -- we've already established that there are certain mental processes which, once automatized, are executed unconsciously - -outside conscious awareness and conscious control. So what's the problem? The problem is that, under the standard view, unconscious processes create conscious mental states. We may not be aware of Helmholtzian unconscious inferences, but we are certainly aware of the percepts they create. The question is: do we have percepts, and other mental states, that influence our behavior even though we are not consciously aware of them? It's one thing to acknowledge the existence and influence of unconscious mental processes. It's quite another thing to acknowledge the existence and influence of unconscious mental states -- states of knowing, perceiving, remembering, thinking, imagining, feeling, desiring, etc.

To restate the point: the construct of automaticity implies that mental processes can be unconscious, if they have been automatized. But the standard view of automaticity is that automatic processes create conscious mental states.

Now we want to know whether mental states themselves can be unconscious.

And, being unconscious, can they influence the person's ongoing experience, thought, and action?

And how would we know?

Distinguishing States from Processes

How do we know whether something is a state or a process? That's where the philosophical concept of intentionality comes in.

Brentano argued that "Intentionality is the mark of the mental". Intentional states are about something, they refer to something, they represent something, other than themselves.

James made the same point when he asserted that mental states "deal with objects independent of themselves". Mental states are truly cognitive, in that they "possess the function of knowing". We perceive objects, we remember events, we think thoughts, we form mental images, we feel emotions, we desire objects or activities.

And Searle repeated the point when he wrote that mental states have "content", in that they refer to some specific feature of the world.

So, in the cognitive domain, unconscious mental states refer to percepts, memories, thoughts, and pieces of knowledge which, when conscious, obviously have intentionality -- precisely because their "content" is "about" something "other than themselves".

The question is, then, whether intentional states -- percepts, memories, thoughts, and other aspects of knowledge -- can be unconscious and still have this defining feature of intentionality.

Implicit Memory

Initial evidence for unconscious cognitive states was provided by laboratory studies of patients with the amnesic syndrome due to bilateral damage to the medial temporal lobes (including the hippocampus) or the diencephalon (including the mammillary bodies). Such patients show an anterograde amnesia (and may show a retrograde amnesia as well). Theories of the amnesic syndrome have involved:

simple loss of memory or inability to learn;
impaired secondary (long-term) memory but spared primary (short-term) memory;
shallow vs. deep processing at the time of encoding;
impaired episodic memory but spared semantic (and/or procedural) memory.

A major shift in theories of amnesia began with a pioneering study by Warrington and Weiskrantz (1968), who confirmed the anterograde amnesia in a study in which patients studied a list of words (such as ASHCAN or CLEANSER, DARK or FOLLOW) and then received tests of free recall and recognition. Compared to the control subjects, the amnesic patients were much less likely to produce items from the studied list on the tests of recall and recognition. That's pretty much the definition of amnesia.

However, when presented with fragments or word stems (such as ASH___) and asked to guess the corresponding word, amnesic patients were much more likely to produce list items (i.e., ASHCAN), compared to other possible completions (such as ASHTRAY). In fact, their performance on the completion task was no different from that of controls.

It is important to understand that this final test was presented as a sort of guessing game.

In the stem-completion test, the subjects were told something like, "I'm thinking of a word that begins with these three letters. Can you fill in the blanks?

ASH___ CLE__

In the fragment-completion test, the subjects were told something like, "I'm thinking of a word that has these letters in it. Can you fill in the blanks?

D__K F_Ll__W

In the earlier recall and recognition tests, subjects were specifically instructed to remember the words they had studied. In the stem- and _fragment completion tests, there was no reference at all to the prior study episode. The subjects were simply asked to produce the first word that came to mind.

The effect on stem- or fragment completion observed by Warrington and Weiskrantz is known as a priming effect, which comes in positive and negative forms.

In positive priming, the prime facilitates processing of the target. The priming observed by Warrington and Weiskrantz is of this sort, because studying words like ASHCAN and FOLLOW made it more likely that subjects would complete the stem ASH___ with the studied word ASHCAN, as opposed to the word ASHTRAY; and more likely that subjects would complete the fragment F_L__W with the studied word FOLLOW, as opposed to the unstudied word FELLOW.
In negative priming, the prime inhibits the processing of the target. Negative priming is much less studied than positive priming. When researchers refer to "priming", they almost always mean positive priming.

Priming shows that some memory of the items encountered in the study phase was retained, despite the patients' failure of recall or recognition. It shows that some trace of the priming experience has been encoded and retained in memory, so that it interacts with the processing of the target. But the fact that priming occurs in amnesic patients, who can neither recall nor recognize the prime itself, means that priming is not merely a matter of conscious recollection.

Based on results such as these, Graf & Schacter (1985) and Schacter (1987) distinguished between two expressions of memory, explicit and implicit.

Explicit memory is the conscious recollection of a past event, such as recall or recognition.
Implicit memory refers to the effect of a past event on some task which does not require conscious recollection, such as priming effects.

Implicit Memory is Episodic Memory

Implicit memory refers to the effect of a past event on experience, thought, or action in the absence of (or independent of) conscious recollection of that event.

Because the definition of implicit memory refers to priming and other effects attributable to a past event, it should be clear that implicit memory, so defined, is a phenomenon of episodic memory. Thus, there are two different expressions of episodic memory, explicit and implicit.

In principle, it is possible that there is a viable distinction between explicit and implicit expressions of semantic memory as well. For example, a person might be unable to define a word, but might still recognize when it is being incorrectly used in a sentence. In this case, implicit semantic memory would refer to the effect of knowledge on experience, thought, or action, in the absence of (or independent of) the person's conscious awareness of that knowledge.

Implicit semantic memory, like implicit episodic memory, is closely related to metacognition, or one's knowledge about the contents of one's own mind.
Implicit semantic memory is closely related to implicit learning, to be discussed later.

Priming in Amnesia

In priming, a subject is asked to study a list of items, and then is asked to perform some task with those items which does not require conscious recollection, for example:

Stem Completion: "I'm thinking of a word that begins with the letters ASH___; Can you guess what it is?
Fragment Completion: "I'm thinking of a word that has the letters F_L__W: Can you fill in the blanks?
Perceptual Identification: In this task, the subject is presented with a word, or some other stimulus, masked or embedded in a noisy background, and asked to identify the stimulus.
Lexical Decision: In this task, the subject is presented with strings of letters, and asked to indicate whether the string constitutes a legal English word.

Priming occurs when task performance is better (or worse, in the case of negative priming) for studied items than for unstudied items. Priming is an expression of implicit memory because the subject's task does not require conscious recollection of the past, but the fact that priming occurs means that some trace of past experience has been retained, and affects subsequent task performance.

Many early (and, for that matter, more recent) studies of implicit memory compare a task like free recall with a task like stem-completion, and find priming on stem-completion in the absence of, or independent of, recall. But this is not the best way to perform these studies. The reason is that, according to the Cue-Dependency Principle in memory, memory varies as a function of the informativeness of the cues presented at the time of retrieval. In stem-completion, the presented cue, a stem like ASH___ or a fragment like F_L__W, is more informative about the target item than the simple query, "Tell me what words you studied". Although recognition tests provide cues that are even more informative than stem-completion tests, in principle differences in cue value are a potentially important confound in studies of implicit memory: unless they are controlled for, we do not know if the difference in performance between explicit and implicit tests has to do with differences between conscious and unconscious memory, which is what we're really interested in, or whether they have to do with differences in cue-value, which is considerably less interesting.

For that reason, the best studies of explicit-implicit dissociations used In the best comparisons of explicit and implicit memory hold the cues presented during the two memory tests constant, but vary the task to be performed by the subject. For example, in a study of amnesic patients by Graf, Squire, & Mandler (1984), both explicit and implicit tests of memory employed word stems as cues:

In explicit stem-cued recall, the subject is asked to complete three-letter word stems with an item from a previously studied wordlist.
In implicit stem completion, the subject is asked to complete three-letter word stems with the first word that comes to mind.

In either case, memory was indexed by the appearance of items from the studied list in the subject's responses, compared to items from an unstudied control list. Graf et al. showed that implicit memory could be dissociated from explicit memory, even when the cues used in the two tests were identical.

Implicit memory refers to any effect on a subject's experience, thought, and action that is attributable to some past event. Implicit memory is said to be dissociated from explicit memory when this effect occurs in the absence of conscious recollection.

Implicit Memory in Other Forms of Amnesia

A dissociation between explicit and implicit memory can also be seen in the anterograde and retrograde amnesias associated with electroconvulsive therapy (ECT) for depression. In ECT, an electrical current is delivered to one or both cerebral hemispheres of a sedated patient. Because of the sedation, the patient does not experience pain, but the shock does cause an epilepsy-like convulsion and loss of consciousness. Even after the patients regain consciousness, they are in a post-ictal confusional state for a period of time. After they have returned completely to their senses, ECT patients will show two kinds of amnesia:

An anterograde amnesia covering a brief period of time following the delivery of the shock -- including some of the time when the patient has been conscious.
A retrograde amnesia covering a brief period of time prior to the delivery of the shock.

A similar combination of anterograde and retrograde amnesia can be observed in individuals who show concussive blows to the head, as for example in auto accidents or on the football field. However, the conditions under which ECT is delivered allow rigorous study of the two amnesic states -- which, it turns out both dissociate explicit and implicit memory.

In a study of post-ECT anterograde amnesia, Squire et al. (1985) found a severe deficit in explicit memory (recognition), but not implicit memory (stem-completion).

Similarly, in ECT-induced retrograde amnesia, Dorfman et al. (1995) found a severe deficit in explicit memory (stem-cued recall) but not implicit memory (stem-completion).

Amnesia is also a consequence of conscious sedation, often used in outpatient surgery, also known as ambulatory surgery or day surgery, such as arthroscopic knee surgery or colonoscopy. In this case, the anesthesiologist will use a drug such as propofol (the drug that Michael Jackson overdosed on and died) or a benzodiazepine for sedation. The patient remains conscious during the procedure, and in some instances can even control the administration of analgesics for pain relief -- a variant known as patient-controlled analgesia. But after recovering, the patients are typically amnesic for the procedure itself -- they have little or no conscious recollection of what happened. This classifies as an anterograde amnesia, because the amnesia covers events that occurred after the sedative was administered.

Cork and his colleagues studied explicit and implicit memory in patients receiving conscious sedation for routine ambulatory surgery. While the patients were sedated, they were read a list of paired associates, such as ocean-water and butter-knife for a single presentation, immediately before the last suture to close the surgical wound. Then they conducted a variety of memory tests while the patient was in the recovery room.

The patients performed very badly on tests of cued recall, in which they were presented with the first word of the paired-associate (OCEAN or BUTTER) and asked to recall the word it had been paired with (WATER or KNIFE). And they also did poorly when given the whole pairs for a recognition test.
But when given the first item (e.g., OCEAN) and asked simply to generate the first word that came to mind, they were more likely to give the studied associate (e.g., WATER) than they were on control pairs, which they had not studied.

Thus, performance on the word-association test revealed a priming effect. Implicit memory was spared, despite a gross impairment in explicit memory.

The dissociation between explicit and implicit memory has now been demonstrated in a wide variety of amnesias:

General anesthesia.
Conscious sedation.
Hypnosis (i.e., posthypnotic amnesia).
Genuine cases of dissociative disorder (of which there are very few, and mostly transient), including

Dissociative (psychogenic or functional) amnesia;
Dissociative (psychogenic) fugue; and
the inter-personality amnesia associated with dissociative identity disorder (multiple personality disorder).

Implicit Memory in Normal Memory

In normal subjects, who do not suffer from anterograde or retrograde amnesia, the dissociation between explicit and implicit memory takes somewhat different forms.

Nelson (1978) found significant savings in relearning for paired associates that were neither recalled nor recognized on standard tests of explicit memory. First, his subjects learned a list of paired associates of the form 48-BOOK and 36-PARTY. After a 4-week retention interval, they received tests of cued recall and recognition -- which showed, not surprisingly, a fair degree of forgetting. Then Nelson asked his subjects to learn a new list of paired associates. Some of the items in this list consisted of the previously studied, but unremembered items (i.e., "old" items such as 48-BOOK), while others consisted of new pairings of old elements (i.e., "new" items such as, 48-PARTY). Despite their forgetting, the subjects showed significant savings in relearning previously unrecalled items, and even items that had been unrecalled and unrecognized. In this case, Nelson's subjects showed spared implicit memory, in the form of significant savings in relearning, in the absence of explicit memory, in the form of recall or recognition.

In another experiment, Jacoby and Dallas (1981) showed that that performance on recognition (an explicit memory task), but not on perceptual identification (an implicit memory tasks) is affected by the "level of processing" at the time of encoding. During the study phase, some items were presented for relatively "shallow" orthographic processing, in which the subjects had to make a judgment about the visual appearance of the word. Other items were presented for relatively "deep" semantic processing, making judgments about its meaning. Level of processing affected performance on a recognition task, but not on on the level of priming in perceptual identification. Here, of course, explicit memory isn't absent. Rather, performance on the implicit memory task is independent of a factor, depth of processing (the Elaboration Principle), that is known to affect explicit memory. In this case, then, the subjects showed implicit memory, in the form of priming, independent of explicit memory.

Independence cuts both ways. Graf et al. (1985) showed that implicit memory (stem completion) but not explicit memory (stem-cued recall) was affected by a shift in modality (i.e., from oral presentation to written recall) between study and test. Thus, implicit memory was dependent on a factor, that does not affect explicit memory.

Taken together, these results indicate that implicit memory is not just a weak form of explicit memory, but rather is a qualitatively different form of memory.

Terminological Issues

Dissociations between explicit and implicit memory come in several forms.

There are population dissociations, where amnesics and controls show different patterns of performance across explicit and implicit tests.
And there are functional (or experimental) dissociations, in which some experimental variable (e.g., level or processing or modality shift) yields different patterns of performance across explicit and implicit tests.

Dissociations also vary in number. As originally classified by Lashley (1952) and Teuber (1955):

There are single dissociations (Teuber called them simple dissociations), in which a single independent variable, A, affects dependent variable, Y, but not another dependent variable, Z. The literature on implicit memory provides numerous examples where amnesia or level of processing affects explicit memory (e.g., recall or recognition) but not implicit memory (e.g., priming). However, there are occasional findings of the opposite type: for example, modality shift appears to affect implicit memory (priming) but not explicit memory (e.g., recall).

More convincing are double dissociations (I call them twin dissociations, because they're really tantamount two two single dissociations) in which an independent variable A affects one dependent variable Y but not another one, Z, while a second independent variable, B, affects Z but not Y. Double dissociations are the gold standard in implicit memory research, because they seem to indicate that performance on the two tests, explicit and implicit, is mediated by different memory systems. There can also be triple and quadruple dissociations, following along the same lines.

A variant on the double dissociation is the crossover dissociation, in which a single independent variable has opposite effects on the two dependent variables. These are exceedingly rare.

Some theorists refer to the explicit-implicit distinction by other names.

So, for example, Johnson & Hasher (1987) and Richardson-Klavehn & Bjork (1988) distinguish between direct and indirect memory, while Squire and his colleagues distinguish between declarative and procedural memory (Cohen & Squire, 1980) -- or, more recently, declarative and nondeclarative memory (Squire & Knowlton, 1995). However, neither of these distinctions captures the essence of the explicit-implicit distinction -- which is that explicit memories are conscious and implicit memories are not.

Moreover, these alternative vocabularies are somewhat confusing, because they diverge from the established technical meanings of these terms in cognitive psychology.

In particular, Squire's equation of "explicit" with "declarative" seems to confuse Winograd's technical definition of declarative memory (which can be represented in sentence-like propositions) with the ordinary-language use of declarative to refer to "something which can be declared". Squire's usage is based on the observation of intact skill (i.e., procedural) learning in amnesic subjects (e.g., classical conditioning, pursuit rotor learning, mirror tracing, and artificial grammar learning). Squire's usage is not preferred because declarative (propositional) memory can be expressed explicitly, in recall or recognition, or implicitly, in priming effects.

Similarly, Johnson & Hasher's equation of "explicit" with "direct" ignores the historical usage of the direct-indirect distinction in the priming literature. In direct priming, the same item is presented at study and test. So, for example, a subject might study the word "doctor", and then be presented with the word "doctor" on a lexical decision test. In indirect priming, different items are presented at study and test. So, for example, a subject might study the word "doctor", and then be presented with the semantically related word "nurse" on a lexical decision test. Not all priming effects are indirect. J&H's usage is not preferred because priming (implicit memory) effects can be either direct or indirect. In fact, the most widely studied priming effect is repetition priming. which is a type of direct priming.

Repetition Priming and Semantic Priming

There are a wide variety of tests of priming.

Perceptual tests, like repetition priming, can be mediated by a perception-based representation which preserves information about the physical features of the stimulus, and its location in space and time, including its spatio-temporal relations to other stimuli. The item presented at the time of testing is a ful or partial recapitulation of the prime presented during study.

In word-stem completion, the stem ASH___ partially resembles the target word ASHCAN, while the fragment F_L__W partially resembles the target word FELLOW.
In perceptual identification and lexical decision, the entire word, ASHCAN, or FELLOW is presented.

Semantic tests, like free association or category-instance generation, can only be mediated by a meaning-based representation which goes beyond physical appearance, and preserves information about the semantic attributes of the stimulus. There is no physical resemblance between the prime and the target.

In category priming, the prime is a category label, while the target is an instance of that category, such as TREE-OAK.
In associative priming, the prime is a word and the target is a semantic associate of that word, such as TABLE-CHAIR.
Semantic priming can also occur in the case of idiomatic associates, such as SOUR-GRAPES.

Notice, however, that even tests normally associated with repetition priming can be used to examine semantic priming. And tests normally associated with semantic priming can be used to examine repetition priming. It all depends on whether the relationship between prime and target is one of physical resemblance, or denotative or connotative meaning.

Most studies of implicit memory involve repetition priming; very rarely do investigators look at semantic priming, which is too bad. But there are some examples out there.

In one early study, Gardner and his colleagues (1973) asked amnesic patients and control subjects to study a list of category instances. On later tests of memory, the amnesic patients performed very poorly on tests of free and cued recall. But when presented with category labels and asked to give the first instance that came to mind, they showed levels of priming that were equivalent, even greater, than those displayed by the normal controls.

Theories of Implicit Memory

There are a number of theories of implicit memory:

Mandler's activation view
The multiple memory systems views of Squire and of Schacter & Tulving;
Roediger's transfer-appropriate processing view; and
Jacoby's process-dissociation view.

These theories can be organized according to a two-way classification.

At the first level, there are theories that postulate single vs. multiple memory systems.

Multiple memory systems theories argue that explicit and implicit memory reflect the performance of separate memory systems, which in turn have substrates in separate brain systems. So, for example, Schacter and Tulving hold that implicit memory is based on one or another of a set of modality-specific perceptual representation systems which store information about the perceptual structure, but not the meaning, of a stimulus. Similarly, Squire holds that explicit memory is based on a medial-temporal lobe memory system.
Single memory systems theories argue that it is not necessary to postulate multiple memory systems to account for population and functional dissociations between explicit and implicit memory. Within the single memory system theories, there is a further distinction between those that emphasize activation and those that emphasize processing.

The activation view is associated with John Morton (of logogen fame) and George Mandler, and holds that implicit memory reflects the activation and integration of knowledge already available in memory, while explicit memory reflects the elaboration of activated knowledge.
The processing view holds that explicit and implicit memory reflect the operations of different processes. So, for example, Roediger's transfer appropriate processing view begins with the recognition that (most) explicit memory tasks are conceptually driven, while (most) implicit memory tasks are perceptually driven.
On the other hand, Jacoby's process dissociation view seems to adopt the view that (most) explicit memory tasks require conscious processing, while (most) implicit memory tasks require only automatic processing.

Mandler's activation view is based on a two-process theory of memory which holds that encoding an item in memory requires two processes:

the activation and integration of pre-existing knowledge to bind together individual features of an experience; and
elaboration of that knowledge to create a new episodic trace.

On this view, priming is mediated by persisting activation (what Rozin and Diamond called the "hot tubes effect"), while recall (and, to some extent, recognition) requires access to the elaborated memory trace.

The activation view can account for most priming effects fairly easily. For example, Diamond and Rozin (1984) found priming for familiar words, such as candy and number, but not for pseudowords, such as canber and numdy. The interpretation is that semantic memory contains entries for familiar words, which can be activated by perception; but no such activation is possible for non-words, because they don't have pre-existing entries in memory to be activated. By analogy to a radio or TV which stays warm for a while after it is turned off, and so comes on more quickly when powered up, they characterized priming as the "hot tubes" effect.

However, the activation view is troubled by findings of long-term finding, which extends further in time than a "hot tubes" effect probably should. Squire et al. (1987) found that priming in amnesic patients could last for as long as 4 days, while Tulving et al. (1991) reported that the amnesic patient K.C. displayed significant priming over intervals as long as 1 year! Over long delays vacuum tubes cool off; by analogy, activation should dissipate over time. So, findings such as these suggest that activation isn't all that's involved in implicit memory.

The primary competitor to the activation view has been the multiple memory systems view.

According to Squire, declarative (explicit) memory is mediated by a medial temporal lobe memory system, while nondeclarative (implicit) memory is mediated by systems located elsewhere in the brain. Within the procedural domain, separate memory systems may be responsible for priming, mediate associative learning (i.e., conditioning), and skill learning (i.e., genuine procedural learning). Similarly, within the declarative domain, separate memory systems may be responsible for episodic and semantic memory.
Schacter and Tulving, for their part, have proposed that priming (or, at least, repetition priming) is mediated by various perceptual representation systems associated with modality-specific cortical structures, while explicit memory is mediated by a separate episodic memory system associated with subcortical structures like Squire's medial temporal lobe memory system. Schacter & Tulving also postulate separate memory systems mediating procedural memory (conditioning and skill learning), primary (short-term or working) memory, and semantic memory. Furthermore, they have identified (through dissociations) at least three separate perceptual representation systems: a visual word-form system, an auditory word-form system, and a structural description system. Each of these is associated with a different "module" in cerebral cortex.

Of course, there can't really be a word-form system in the brain, because writing was invented only about 5,000 years ago, and that doesn't leave enough time for the brain to have evolved one. However, Stanislas Dehaene, a French cognitive neuroscientist has identified what he calls a "letterbox" in the left hemisphere which has some of the properties attributed to the visual word-form system -- except it deals with letters, not words. Dehaene notes that all writing systems make use of the same basic shapes, vertical and horizontal lines, closed and open curves, and the like, and experiments like those of Hubel and Wiesel show that there are neurons in the visual system that respond specifically to many of those shapes. So, it is possible that, with the invention of writing, an area of the brain that responds to particular features is co-opted to respond to letters as well.

Schacter and Cooper's studies of implicit memory for "impossible" figures illustrate some of the evidence for a structural description system. Impossible figures resemble three-dimensional objects, but cannot actually be given a structural description in three dimensions. The finding that there is priming in an object-decision task for possible, but not impossible objects. is consistent with the idea that there is a module for processing structural descriptions of three dimensional objects that cannot form structural descriptions.

The Squire and Schacter-Tulving views employ overlapping taxonomies of memory, and have a complementary relationship with each other: Squire seems more interested in the system mediating explicit memory, while Schacter & Tulving seem more interested in implicit memory. Both theories, however, are troubled by the potential proliferation of memory systems, one for each dissociation revealed by experimental research. More important, perhaps, they are troubled by the apparent nonexistence of any neurological case in which implicit memory is impaired, while explicit memory is spared. If the brain systems mediating explicit and implicit memory are truly independent, then it should be possible to selectively destroy (or damage) each system without affecting the other.

For that reason, single-memory system views remain viable, although many contemporary single-memory theorists distance themselves from Mandler's activation theory.

Roediger's transfer-appropriate processing view begins with the observation that most tests of explicit memory depend on conceptually driven processing, while most (but not all) tests of implicit memory depend on data driven processing. Therefore, the observed dissociations between explicit and implicit memory may be more a matter of the types of processing required by the test than the involvement of any special memory system. "Transfer" of memory between tests is good when they require the same kind of processing, and poor when they do not. The TAP predicts that dissociations will occur between tests of implicit memory that require different kinds of processing, and in fact Roediger and his colleagues have found such dissociations.

Jacoby's process-dissociation framework also is predicated on the existence of only a single memory system. Jacoby argues that explicit tests emphasize conscious processing while implicit tests emphasize unconscious processing; but every task has its unconscious (controlled) and conscious (automatic) components. He attempts to separate these by means of his method of opposition, in which subjects are instructed to deliberately exclude remembered items from what would ordinarily be an implicit memory task (e.g., stem completion). Jacoby's process dissociation procedure takes data generated by the MOP to provide estimates of automatic (unconscious) and controlled (conscious) contributions to task performance. When this is done, explicit tests are found to be largely controlled by conscious processes, implicit tests by unconscious processes. (Actually, if truth be told, Jacoby's process-dissociation view can be construed as consistent with a multiple-memory-systems view, on the assumption that controlled and automatic memory processes are mediated by different memory systems in the brain. Sometimes Jacoby writes like a multiple-memory-system theorist, but in general he aligns himself with single-memory-systems theorists, and they with him. For example, Roediger and Jacoby are colleagues at Washington University, St. Louis.)

Implicit Memory for Novel Information

There is another viable classification of implicit memory theories, based on their predictions concerning implicit memory for novel events.

The activation (sometimes called the modification) view holds that implicit memory reflects only the activation (or modification) of pre-existing knowledge. Therefore, it cannot explain implicit memory for novel events.
All the other views hold that implicit memory reflects the encoding of some sort of new representation of experience. Accordingly, they predict that implicit memory will occur for novel events, which do not have any prior representation in memory, as well as familiar events, which obviously do have such a representation. These new representation views then divide according to single- and multiple-memory system views

The MMS views include those of Schacter and Squire
The SMS processing views include those of Roediger and Jacoby. (Recall, however, that Jacoby's process-dissociation view can be construed as consistent with a multiple-memory-systems view, on the assumption that controlled and automatic memory processes are mediated by different memory systems in the brain.)

In the current environment, most controversy is between the multiple memory systems and the processing views, giving the impression that the activation view is dead and can be safely ignored. However, in my view reports of the death of the activation view are greatly exaggerated. To begin with, the activation view seems to give a better account of implicit memory for novel stimulus materials.

Tests of the activation view began with a paper by Diamond & Rozin, which showed priming for actual two-syllable words (e.g., CANDY and NUMBER) which have a pre-existing lexical representation, but not for disyllabic pseudowords (e.g., CANBER and NUMDY) which do not. By itself, this finding was consistent with the activation view. However, there also appeared evidence contradicting this finding, and showing that implicit memory was possible for nonwords as well as for words.

Jeff Bowers (1994) found priming for words (e.g., KITE) and for both legal (KERS) and illegal (XYKS) nonwords, although there was less priming for illegal than for legal nonwords. He argued that priming for nonwords was inconsistent with activation theory, but it turned out that the priming observed for illegal nonwords, which is the really big test of the new representations, was affected by level of processing at the time of study; therefore, technically, it was not dissociated from explicit memory. In any event, Bowers agreed that his experiment was not the last word, not least because we need a principled way of distinguishing between familiar and novel items. That's true: after all, because we make sense of new events in terms of what we have learned from prior experience, nothing is ever completely novel.

Jennifer Dorfman (1994) asked the deeper question of what makes an illegal word illegal. Based on Mandler's dual-process theory of memory, she argued that presentation of a word automatically activates the sub-lexical components contained in a word, and then integrates them to form the word itself. From this point of view, then priming of novel materials like nonwords depends on how they are constructed: if they are constructed of existing sub-lexical components, you should get priming; if they aren't, you shouldn't. So she constructed novel words of three types: morphemic, composed of actual morphemes (e.g., GEN-VIVE); syllabic, composed of syllables which are not morphemes (e.g., FAS-NEY), and pseudosyllabic, composed of elements which are legal, but neither morphemes or syllables (e.g., ERK-TOFE). She got highly consistent priming for morphemic items, less consistent priming for syllabic items, and rarely got priming for pseudosyllabic items. Moreover, in Dorfman's experiments, the priming for morphemic and syllabic pseudowords was independent of level of processing (there was little or no priming for pseudosyllabic pseudowords, rendering the question moot). So, sub-lexical structure is an important determinant of priming for nonwords, which is what the activation view holds (and which the perceptual representations view can't explain). In any event, some version of Mandler's activation view offers the best available explanation for why truly novel materials don't produce priming.

In the final analysis, my own view is that it is not necessary to choose between these theories. The memory systems view can incorporate the processing view, although the processing view has difficulty explaining outcomes like conceptual priming in amnesics or normals. More important, I think that theory is now well served by an almost exclusive focus on repetition priming, and experimental procedures which don't favor semantic processing. Such experiments never encourage investigators to look for conceptual effects, and don't allow you to get it even if you looked for it. Moreover, theory has not been well served by the general disinterest shown by investigators in the relations among various forms of implicit memory (e.g., priming, skilled performance), or by what appears to be a studied avoidance of computational models of memory.

Intentionality and Automaticity in Implicit Memory

When implicit memory is dissociated from explicit memory, the implicit memory is an unconscious memory. The subject shows a priming effect, but does not remember the priming event. Still, the fact that priming occurs shows that some trace of the event has been encoded and stored in memory, even if it cannot be consciously retrieved. This memory trace is "about" something: it refers to, and represents, the prime.

The representation may be perception-based, in which case it is "about" what the prime looked like (or, in the case of auditory priming, what it sounded like, etc.), and it will mediate repetition priming.
Or it may be meaning-based, in which case it is "about" the meaning of the prime, and it will mediate semantic priming.

It has to be said that repetition priming isn't all that interesting (unless you are a perception theorist, and you want to know what perception-based representations look like). The memory in repetition priming is the memory of a wire paper clip, which, having been bent once, is easier to bend a second time (metallurgists use the term memory in just this sense). The memory in semantic priming is "about" so much more, and comes closer to what we ordinarily mean by memory.

Implicit memories are unconscious memories, and thus are unconscious mental states. But the processes that create the priming effect are presumably automatic in nature (and turn out that way, when investigators have explored them with techniques like the process-dissociation procedure). But implicit memory doesn't just reflect automatic processing of the prime. The subject was conscious of the priming event at the time it occurred, but subsequently lost conscious access to it. It's one thing for priming to occur automatically and unconsciously; it's another thing entirely for the person to be unaware, in memory, of the event that gave rise to the priming effect.

Still, whether it's manifested in repetition priming or semantic priming, implicit memory is unconscious memory, because it can be dissociated from explicit, or conscious, memory. The next question is whether the explicit-implicit distinction can be extended beyond memory, to other domains of cognition, and for that matter the entire range of mental life.

In the rest of these lectures, semantic priming will serve as the strong test of an unconscious mental state. That is because the representation underlying semantic priming carries meaning, it doesn't just represent physical structure.

Implicit Perception

Research on implicit memory lays the foundation for extensions to other domains of cognition, and beyond cognition to emotion and motivation. In particular, the same kinds of priming effects that provide evidence of implicit memory can also provide evidence of implicit perception.

Implicit Perception Defined

By analogy to implicit memory, we may define implicit perception as the effect of a current event on experience, thought, or action, in the absence of (or independent of) conscious perception of that event.

(By "current event" we also mean an event in the very immediate past -- what William James referred to as "the specious present"). The immediate past would include the temporal boundaries of primary, "short-term", or "working" memory.

Implicit perception is typically demonstrated by the same sorts of priming tasks used to demonstrate implicit memory. The essential difference between the two domains has to do with the subject's awareness of the event to be remembered, at the time it occurred:

In implicit memory, the subject was consciously aware of the event at the time it occurred, but cannot consciously remember it later. Thus, it is memory that is implicit in priming.
In implicit perception, the subject was not consciously aware of the event at the time it occurred. Because priming implies that the event was processed anyway, albeit outside of conscious awareness, it is perception that is implicit in priming.

Subliminal Perception

Implicit perception is epitomized by the controversy over so-called subliminal influence.

In the 1950s, Vance Packard argued in The Hidden Persuaders that corporations and advertising agencies employed subliminal advertising to alter people's attitudes and interests without their knowledge.

During World War II, a message that Americans should buy War Bonds was embedded into a cartoon shown in movie theaters.
In the 19950s, a market researcher claimed that subliminal messages to "Buy Popcorn" and "Drink Coca-Cola" increased sales among moviegoers. However, this "proprietary" research was never subjected to peer review -- and might never have been done in the first place.

Packard's assertions were generally discounted, but Issues of subliminal influence were revived in the famous 'Judas Priest" trial in the 1980s.
More recently, the 2000 presidential election campaign included the charge that Republicans subliminally presented the word "RATS" during an ad concerned with Gore's support for the Democrats' plan for insurance coverage of prescription medicines.

Although very much a 20th-century idea, the idea of subliminal perception goes way back.

In the 17th century, Leibniz (1704) argued that conscious percepts were the product of petites perceptions lying below the threshold of perception:

[A]t every moment there is in us an infinity of perceptions, unaccompanied by awareness or reflection.... That is why we are never indifferent, even when we appear to be most so.... The choice that we make arises from these insensible stimuli, which... make us find one direction of movement more comfortable than the other.

In the 19th century Herbart (1816) asserted that conscious and unconscious ideas compete at the threshold (German, limen) of consciousness.

One of the older ideas can... be completely driven out of consciousness by a new much weaker idea. On the other hand its pressure there is not to be regarded as without effect; rather it works with full power against the ideas which are present in consciousness. It thus causes a particular state of consciousness, though its object is in no sense really imagined.

These ideas raised the question of whether perceptual representations did, indeed, exist below the threshold for conscious awareness.

In the late 19th century, the American pragmatist philosopher C.S. Peirce (pronounced "Perss") and his graduate student Joseph Jastrow attacked Herbart's concept of the limen in the first published studies of subliminal perception. In these studies, they showed that forced-choice judgments of heaviness and brightness were more accurate than chance, even though the observers had zero confidence in their judgments. On the assumption that the observers' confidence levels were a reflection of their conscious (explicit) perception, it is apparent that stimuli below the (relative) threshold were perceived outside conscious awareness.

Over the subsequent decades, a number of theorists were extremely critical of arguments and evidence for subliminal perception. Among the most important criticisms of this work had to do with the procedures for setting the "threshold" against which "subliminal" stimuli were calibrated. I call this criticism the threshold bugaboo.

Put bluntly, if the thresholds are not set appropriately, then ostensibly "subliminal" stimuli might not really be subliminal at all.
Even with adequate threshold-setting procedures, the threshold is defined as the level of stimulation at which a stimulus is detected on 50% of trials -- or, alternatively, by 50% of the subjects. But even below this level the stimulus will be detected sometimes, by some people, and these conscious detections may be sufficient to produce the illusion of subliminal detection.

Both points are well taken, but it's also true that, if you want to dismiss someone's claim to have documented subliminal perception, it's pretty easy just to complain about his threshold-setting procedures.

Underlying many of these criticisms was a behaviorist (or, at least, positivist) approach to consciousness, which identified consciousness with discriminative behavior (you wouldn't want to depend on self-reports, would you?). But consider the implications of this definition for the Peirce and Jastrow experiment. Their evidence for "subliminal" perception was discriminative behavior in a forced-choice test. Therefore, the fact that their observers were able to make valid discriminative choices indicates, to the critics, that they were conscious of the stimuli in question. In this way, subliminal perception is ruled out of existence by fiat.

Later, under the influence of cognitive approaches to perception and attention, some theorists were willing to concede that some subliminal processing might be possible after all. However, they usually confined the scope of subliminal perception to analyses of the physical characteristics of stimuli -- the sorts of perceptual analyses that might be accomplished preattentively, and thus preconsciously, as in the "early selection" theories of attention. However, because it was generally thought that semantic analyses of the meaning of a stimulus could not be accomplished preattentively, or preconsciously, the possibility of subliminal perception was not extended beyond the perceptual domain to the semantic domain.

In the 1980s, the increasing popularity of priming as an experimental paradigm led to a revival of subliminal perception research, asking the question of whether priming was possible even if the primes were presented subliminally. In particular, the occurrence of subliminal semantic priming would provide convincing evidence for subliminal perception.

In this newest research, then implicit perception is revealed by the same sorts of tasks that reveal implicit memory, but with the following important difference: in implicit memory, the subject consciously perceived the event at the time it occurred; in implicit perception, the subject did not do so.

Masked Priming

Most traditional research on "subliminal" perception, involving the presentation of weak stimuli, as in the study by Peirce and Jastrow -- where the "weak" stimulus is a very small difference between two weights or two lights.

Later research on subliminal perception shifted away from an emphasis on "weak" stimuli, or at least a redefinition of what it means to be weak, to briefly presented stimuli, employing a mechanical device known as a tachistoscope (first introduced in the 1850s) -- a kind of slide projector that permits very briefly controlled exposures; or, more recently, by means of computers. The "T-scope", as it quickly came to be called (college students of a particular age were taught that, so long as they could pronounce the phrase tachistoscopic episcotister, they could be sure that they weren't suffering from tertiary syphilis!), permitted experimenters to expose visual stimuli for extremely brief durations, as little as 1 msec. At those durations, the stimuli themselves are invisible. Yet, they gave rise to what we would now recognize as priming effects, and other effects interpreted as subliminal perception.

But there's a problem with this method, which is known as the persistence of vision (Segner, 1740). That is to say, a visual percept does not terminate when the visual stimulus terminates. In the classic multi-store model of memory, the very first stage of human information processing consisted of a set of modality-specific sensory registers that held a veridical representation of a stimulus, analyzed for physical structure but not for meaning, for a very brief period of time.

The sensory register for vision came to be called iconic memory, and the sensory register for audition came to be called echoic memory (Neisser, 1967). The representation then faded briefly over time, or was displaced by other, incoming information arriving more recently. But still, unless actively displaced, a stimulus presented for only 1 msec persisted in memory for much longer than that -- long enough, perhaps, that "subliminal" stimuli was not quite as "subliminal" as investigators thought they were.

The problem was solved by re-rigging the T-scope to control the duration of the icon by following the stimulus with yet another stimulus, which would effectively displace it from the sensory register. This procedure, which is called masking, is now done with computers, and hardly anyone uses T-scopes anymore.

In masking, the priming stimuli are presented at intensities, and for durations, that would be normally sufficient to allow them to be perceived under ordinary circumstances, but embedded in a "mask" that effectively renders it invisible. The interval between the presentation of the stimulus and the presentation of the mask is known as the stimulus-onset asynchrony (or SOA). Stimuli are effectively masked, and rendered invisible, with an SOA of approximately 50 msec.

There are basically four kinds of masking procedures:

energy masking, in which the stimulus is followed by a high-energy flash of light;
backward pattern masking, in which the stimulus is followed by a visual pattern -- perhaps a string of letters or hashmarks like #####;
forward pattern masking, in which the pattern mask precedes the stimulus (this works too);
forward and backward pattern masking, in which the mask both precedes and follows the stimulus (it's a little like wearing both a belt and suspenders).

Marcel (1983) carried out a series of studies of "masked" semantic priming, in which masked primes (e.g., DOCTOR) facilitated lexical decisions of semantically related targets (e.g. NURSE). Marcel claimed that his experiments showed that stimuli could be processed preconsciously for meaning. Of course, this upset proponents of "early selection" theories of attention, who believed that preattentive processing was limited to physical features, and that semantic processing had to wait until attention -- and, thus, consciousness -- had been devoted to the stimulus.

Marcel also showed that the Stroop color-word effect could be obtained with subliminal presentations of the color words, embedded in color patches.

The subliminal Stroop effect was confirmed by Cheesman & Merikle (1984). In their experiment, the Stroop effect was weakened as the prime became less and less detectable, but it was still apparent at very low levels of detectability, approaching the "threshold" level of 50%. However, the Stroop effect disappeared when prime detectability fell well below chance levels of 50%.

On the basis of findings such as these, Cheesman & Merikle further distinguished between two different thresholds:

the subjective threshold, or "the detection level where subjects claim not to be able to discriminate perceptual information at better than chance level"; and
the objective threshold, or "the level of detectability where perceptual information is actually discriminated at chance levels".

In their view, "subliminal" perception occurs in the gap between the subjective and objective thresholds.

In a further experiment, Cheesman and Merikle put the distinction between objective and subjective thresholds into practice. They examined the subliminal Stroop effect under three conditions:

supraliminal stimulation, when the prime-mask SOA was long enough that the mask was ineffective, and the color word was clearly visible;
subliminal stimulation, when the prime-mask SOA was shortened until the subjects could not detect the color word at better than chance levels; and
what might be called "sub"-subliminal stimulation, when the prime-mask SOA was shortened still further, to the point where all discriminative response to the stimulus disappeared.

The Stroop effect was clearly obtained in the supraliminal condition, but it was also apparent in the subliminal condition, when the duration of the color word was below the subjective threshold. But when the duration of the color word was below the objective threshold, the subliminal Stroop effect disappeared.

Cheesman and Merikle also introduced a second manipulation in their experiment. In addition to varying the prime-mask SOA, moving around in the window between the subjective and objective thresholds, they also varied the prime-target SOA -- that is, the length of the interval between the presentation of the prime and the presentation of the target. The subliminal Stroop effect was reduced as the prime-target SOA lengthened. In other words, subliminal perception doesn't last very long.

The Limits of Subliminal Perception

Which brings us to the question of the limits of subliminal perception. Were the Vance Packards of the world right, that subliminal presentation of a message like "Drink Coke" or "Vote Republican" could actually encourage people to do such things? Were the plaintiffs in the "Judas Priest" trial right, that the masked presentation of a message, played backwards yet, could induce listeners to commit suicide? Are the marketers of subliminal self-help tapes right, that the masked presentation of suggestions to stop smoking or cut down on eating actually help people?

Probably not, if a series of studies by Greenwald and his colleagues are right.

Studies of masked semantic priming focus on denotative aspects of meaning. In other recent work, Greenwald and his colleagues (e.g., 1989) have focused on connotative meaning, or emotional valence (Greenwald is a social psychologist, so he's interested in topics like emotion; he's also an extremely careful methodologist, and an extremely dogged researcher who spent the better part of a decade documenting subliminal perception in a manner that is beyond dispute). In their experiments, subjects make judgments about the emotional valence of target words, preceded by a masked prime word whose emotional valence is either congruent or incongruent with the target. Affective priming occurs when judgments about prime-congruent targets are faster than judgments about prime-incongruent targets. Using this sort of paradigm, Greenwald et al. have shown subliminal affective priming.

Greenwald and his colleagues (1996) have also developed a new statistical technique for identifying subliminal perception, in which performance on a subliminal perception task like affective priming (plotted on the Y-axis) is regressed onto performance on a conscious perception task like detection (plotted on the X-axis). Where X=0, any positive Y-intercept provides evidence for subliminal perception: an undetected stimulus has given rise to a priming effect. Their general finding, replicated across a large number of studies, is that the Y-intercept is almost always positive.

So subliminal perception actually occurs. But just because subliminal perception occurs, that doesn't mean that subliminal perception is particularly powerful. There are clear limits on the effect.

For

example, the impact of subliminal perception decreases with decreasing prime duration and prime-mask stimulus-onset asynchrony (SOA), and with increasing prime-target SOA as well. The shorter the SOA, the more degraded the prime. Thus, highly degraded subliminal primes produce less priming. Put another way, the more degraded the prime, the less subliminal perception is possible.

Moreover, Greenwald is able to get subliminal affective priming with one-word primes, but not two-word primes. In his usual experiment, single-word primes like enemy and friend prime judgments about affectively congruent targets like loses and wins -- this is the basic subliminal affective priming effect. But in one study, Greenwald and Liu (1985) composed two-word phrases consisting of positive and negative words:

Hero Wins is a positive phrase, composed of two positive words.
Enemy Wins is a negative phrase, composed of a positive and a negative word.
Hero Loses is also a negative phrase, composed of a positive and a negative word.
But Enemy Loses, composed of two negative words, is a positive phrase.

If subjects can extract the meaning of a two-word phrase, then a positive phrase like ENEMY LOSES should prime affective judgments of positive targets. But it doesn't. Two-word phrases like HERO WINS prime the processing of positive targets, but phrases like ENEMY LOSES do not. Thus, it appears that subliminal semantic processing is able to analyze the connotative meaning from individual words, but cannot synthesize the connotative meaning from a two-word phrase.

In personality and psychopathology, there is a remarkably large literature ostensibly documenting the effects of "subliminal symbiotic stimulation", or the subliminal presentation of the message Mommy and I are one (the psychologist who discovered this effect, Lloyd Silverman, was a psychoanalyst). Based on everything we know about the limits of subliminal semantic priming, whatever the effects of symbiotic stimulation are, they are almost certainly not subliminal in nature. The analytic limitations of subliminal perception are so severe, that it is almost certainly not possible to process the meaning of a sentence as complicated, and as subtle, as Mommy and I are one.

As a hypothesis, I suggest that the limits of subliminal perception -- whether, and to what extent, subliminal perception can extend to semantic as well as physical analysis --depend on where the prime is located in the space between the subjective and objective thresholds. If the prime is very close to the subjective threshold, some degree of subliminal semantic processing will be possible. But if the prime is very close to the objective threshold, only perceptual analyses will be possible.

So there are limits to subliminal perception.

The prime-mask SOA can't be too short (bringing us perilously close to the objective threshold).
The prime-mask SOA can't be too long (because apparently subliminal effects don't last too long).
The subliminal stimulus can't be too complicated (or it will tax the capacity of automatic processing)

But however limited it might be, subliminal perception seems to be a valid construct. In order to circumvent arguments over threshold-setting procedures, I prefer to think of subliminal perception as the paradigm case of implicit perception. The priming observed in experiments on subliminal perception occurs in the absence of conscious perception of the prime.

The Ethics of Subliminal Advertising

Subliminal advertising doesn't work -- or, at best, it doesn't work very well. But that doesn't mean that the idea of subliminal advertising doesn't raise interesting ethical questions. Consider this one, posed to "The Ethicist" in the New York Times Magazine (02/02/2014).

Our subconscious beliefs, which we cannot easily understand or change, motivate many of the decisions we make. Some companies use manipulative branding techniques that create untrue associations between their products and some deep human desires -- like sex or social status -- with the goal of having the viewer establish this connection unconsciously. The desired result, presumably, is that the consumer will then buy the product with the misguided intent of fulfilling said desire. Is this practice unethical? ADAM GIDDING, NEW YORK

When I first received this question, I did not take it too seriously. But I've found myself thinking about it a lot, and so have a lot of other people. The Federal Trade Commission and the Bureau of Consumer Protection have each considered this issue, including situations in which the problem isn't even visible.

In the eyes of the government, clarity of purpose is paramount: An ethical commercial seeks to distinguish a product from its competition, while an unethical advertisement tries to cloud that distinction and purposefully confuse the consumer. But you're asking about something significantly more complex. Your question is closer to the plot of John Carpenter's "They Live."

Everyone knows that advertisers try to create unreal associations between products and lifestyles. Beer commercials show beer-drinkers living dynamic lives, so the implication is that consuming a certain type of beer will make your life more exciting. There is a kind of unreal message there. But this isn't the same as subliminal advertising. The fact that the commercial's erroneous relationship can be described and mocked proves that a consumer can recognize and reject the ad's message.

This is not the case with actual subliminal messages, which are embedded into other mediums and virtually impossible to recognize with the conscious mind. Imagine a beer company that shoots a commercial of two hunters enjoying brews in a duck blind but secretly includes a single frame of an orgy. The single frame would be imperceptible to the casual viewer but, in theory, would still imprint on the viewer's unconscious and tap into a base desire he will unknowingly connect to beer consumption (and if this strikes you as impossible, go to YouTube and punch the words "subliminal advertising" into the search field).

Now, the general scientific consensus is that this advertising strategy doesn't work particularly well. Yet if it works to any degree whatsoever, it creates a unique ethical dilemma. Subliminal advertising is an optical trick. As such, it would be easy to argue that it's unethical (because deception is unethical). That argument is valid, but it draws an awkward distinction between "lifestyle" advertising and subliminal messaging. I say this because, within reason, adults should be expected to understand the implicit agreement they enter into by watching advertising.

When a commercial for Tide laundry detergent comes on TV and the viewer elects to continue watching, the viewer is essentially saying: "O.K., pitch me on Tide. I am open to this attempt. Try to persuade me to buy Tide instead of Cheer." We know what we are seeing and what Tide is trying to do. The only inflexible ethical expectation is that Tide won't lie to you directly. They can't fabricate factual data; Tide can't claim it makes you taller or that it’s safe to eat as a snack. But I don't think it's incumbent on the creator of the commercial to pitch the product for the "right" reasons (whatever those may be). It's not as if they must exclusively argue that Tide is an effective solvent or that it's a bargain (and even if they did, few would take those messages at face value).

The viewer knows he's seeing a biased 30-second attempt to promote the consumption of a certain product. The important thing is that this remains the ad's singular intent: that it's selling the product it purports to be selling. The motives need to match. So if Tide commercials are embedding subliminal messages solely to sell more Tide, I don't think it would be necessarily unethical; it potentially could be, but that would depend on what the message was. To me, the larger risk with subliminal messages is that they could (in theory) be used to sell something else entirely.

Here's what I mean: Let's say somebody made a tampon commercial and employed subliminal advertising techniques, but the hidden messages promoted cigarettes. This would mean people were being pitched an unrelated product without their knowledge. They could not choose to avoid (or even consider) the embedded message, because they would not know the message was being delivered. That destroys the tacit relationship between the advertiser and the consumer. It does not allow the viewer to contextualize what they are experiencing.

It would likewise be unethical if a commercial used subliminal advertising to besmirch a competitor (Tampax shouldn’t use subliminal messaging to claim that Kotex was racist) or to directly target children. But viewed purely as an advertising technique for mature consumers — and assuming the creator's unconscious messages reflect the intent of its conscious messages -- I don’t see subliminal ads that differently from regular ads. The traditional ethical guidelines would still apply.

Beyond Subliminal Perception

It turns out, however, that implicit perception is not confined to the subliminal case. It also encompasses conditions where the stimuli are clearly supraliminal (in terms of energy levels, duration, and absence of masking), but nevertheless are not consciously perceived.

In the neurological literature, there are a number of such syndromes, including

blindsight;
visual neglect; and
prosopagnosia.

And in normal, neurologically intact subjects, there are a number of phenomena in which the stimulus may be supraliminal but not consciously perceived, by virtue of the subject's attention being directed elsewhere:

presentation in the periphery of the visual field; or
presentation over the unattended channel in a dichotic listening task.

In other cases, a stimulus may not be consciously perceived despite the fact that attention is directed right at (or, at least, near it):

repetition blindness;
inattentional blindness;
the attentional blink;
change blindness.

Blindsight

Weiskrantz's classic study of blindsight illustrates implicit perception with supraliminal stimulation. Following surgical ablation of his left occipital cortex as a radical treatment for intractable migraine headaches, patient D.B. suffered a blind spot, or scotoma, initially covering his entire left visual half-field (it subsequently shrank to the his lower-left quadrant of this field). He had no visual experience of stimuli presented in this portion of his visual field. Nevertheless, instructed to guess, D.B. could make above-chance judgments about stimuli presented in his scotoma: the presence or absence of a stimulus, its location, form, movement, velocity, orientation, and size.

D.B. is able to point his finger at objects presented in his scotoma, and do so almost as accurately as he does when objects are presented in his intact visual field.

D.B. is able to discriminate between horizontal and vertical lines, between vertical and diagonal lines, and between Xs and Os.

D.B. is able to discriminate between coarse and fine gratings presented in his scotoma. With very fine gratings, perception in his left-field scotoma is not as good as in his intact right field, but with coarser gratings, left-field perception is just about at ceiling.

Blindsight is not to be confused with the Riddoch phenomenon, in which otherwise blind individuals are able to perceive motion. In the Riddoch phenomenon, the visual cortex in the occipital lobe has been destroyed, rendering the patient unable to see objects. However, in rare cases the "motion area" of the brain, M1, is spared, allowing the patients to perceive objects that are in motion. Apparently, M1 has a connection to a perceptual processing area in the middle portion of the temporal lobe that do not go through V1, allowing patients to perceive objects that are in motion -- or static objects when the patients themselves are moving. But this is not blindsight. Patient DB has no visual experience of objects presented in his scotoma. But Riddoch's patients consciously perceive moving objects. They don't need to rely on implicit, or unconscious, perception.

A syndrome similar to blindsight can also be found, on occasion, in patients with other forms of cortical sensory defects.

For example, Patient IA lost his hearing after damage to his primary auditory cortex, A1. Nevertheless, he is able to detect auditory stimuli with the same accuracy as controls, and also has some ability to localize those stimuli in space. He doesn't do so well on other tasks, but the preserved abilities, in the absence of the conscious experience of hearing, has been dubbed Deaf hearing.

There is also a patient who suffers from tactile anesthesia following damage to his primary somatosensory cortex, S1. Nevertheless, he shows some preserved tactile function, despite the fact that he cannot consciously feel anything - -a syndrome called Numbsense.

Prosopagnosia

A special form of visual agnosia known as prosopagnosia is associated with damage to an area in the temporal lobe known as the fusiform gyrus. Patients suffering from this form of brain damage do not recognize, and cannot identify, faces that are, objectively, familiar to them. Nonetheless, prosopagnosic patients show face-name and name-face priming effects, indicating that some degree of recognition and identification goes on unconsciously.

Neglect

In another neurological syndrome, hemispatial neglect, the patient ignores regions of space contralateral to the site of the lesion -- meaning that they will neglect objects in their left visual field. Patients who show hemispatial neglect tend to have brain damage localized near the junction of the temporal and parietal lobes.

In the current context, the really interesting question is whether neglect patients can process information from the neglected area of space -- preattentively, and preconsciously. A clever experiment by Marshall and Halligan (1988) suggests that they can. These investigators presented patients with two versions of a figure, such as a house and a house with smoke billowing out a window. When the smoke was on the right side of the house, the patients noticed this immediately. But they did not notice the smoke when it was on the left side -- the one they neglected. Nevertheless, when asked which house they would rather live in, the patients chose the one that was not on fire! Evidently, information from the left side of the house was processed unconsciously.

Parafoveal Vision and Dichotic Listening

Similar sorts of findings have been obtained in neurologically intact subjects.

In parafoveal vision, subjects may be asked to detect the appearance of targets presented in the center of a screen. At the same time, words can be presented on the periphery of the screen. Typically, subjects do not notice these peripheral targets. But they can induce semantic priming effects. One such experiment was conducted by Bargh and PIetromonaco (1982), and discussed in the Lecture Supplement on "Automaticity and Free Will".

An interesting example of parafoveal vision is the parafoveal Stroop effect. In this paradigm, subjects are asked to name the color of a patch presented in the center of their s visual field. Off in the periphery, a color word is presented. If the word is presented far enough in the periphery -- not very far at all, in fact -- the subject does not notice the word. Yet presentation of an unnoticed incongruent word nevertheless disrupts the subject's color-naming performance.

There is a little trick in this experiment, which is that it's hard to read words that are presented even a little bit off the center of vision. So, it's necessary to present the words in bigger fonts, to compensate for the subject's reduced visual acuity.

In an experiment by Merikle &Gorewich (1979), color words printed in small type, presented relatively close to the fixation point, yielded a significant Stroop effect, but none when presented just a little farther out. But when the words were presented in a somewhat larger type, these investigators found Stroop effects at both angles of displacement.

Another experiment by Gatti & Egeth (1978) employed an even larger type face, and obtained significant Stroop effects even further out in the periphery

In a dichotic listening experiment similar to those described in the Lecture Supplement on "Attention and Automaticity", Eich (1984) asked subjects to shadow a demanding prose passage. Over the unattended channel, he presented them with paired associates consisting of a homophone (words that have the same sound but two different meanings) plus a disambiguating word -- such as TAXI-FARE (as opposed to FAIR-GAME) and WAR-PEACE (as opposed to PIECE-CAKE). As would be expected, the subjects did not notice these words, and they didn't remember them on a later recognition test. But when given a spelling test and asked to spell the homophones, they spelled them in line with the disambiguating context. So, again, the subjects were not conscious of the stimuli coming over the unattended channel, but they processed them anyway, to at least some degree, outside of conscious awareness.

Implicit Perception in Inattentional Blindness

In the classic instances of preattentive processing, the subject's attention is deliberately focused elsewhere -- on one channel of a dichotic listening task rather than the other, or to some portion of the visual field rather than somewhere else. However, there are also other circumstances in which people do not consciously see or hear things, even when their attention is not distracted by a competing task. Although some of these phenomena appear to reflect the fact that subjects' attention is focused on one aspect of a stimulus rather than another, in every case the subjects appear to be blind to some object or feature, despite the fact that they are staring right at it. This phenomenon is called inattentional blindness (IB).

IB was initially

Or, perhaps we should say, IB was rediscovered.

Once the phenomenon of IB itself was established to their satisfaction, Mack and Rock (1998) conducted an extensive series of studies designed to discover its properties and limits). For present purposes, the most interesting of Mack and Rock's experiments attempted to examine implicit perception during inattentional blindness. That is to say, what is the cognitive fate of a stimulus for which the subject is inattentionally blind? Some hint of perception without awareness came from the early studies, which found that subject's guesses concerning the location of the object were about equally accurate, regardless of whether subjects saw the square. But the most interesting evidence comes from studies of priming.

In one set of studies, for example, familiar words such as flake, and grace served as the new objects. In one version of the experiment, the words were presented in the fovea, while in another version the words were presented parafoveally; in either case, however, the subjects' attention was focused on the cross. After the inattention trial, the subjects were asked to complete word stems such as fla___ or gra___ (alternative completions include flame and flash, grant and grasp). Under full attention conditions, fully 96% of subjects saw the target word, and every one of these subjects used the target as a stem completion. On the inattention trials, however, almost two-thirds of the subjects showed IB. Of these subjects, more than one third produced the target as a stem completion; slightly fewer subjects showed priming when the words were presented parafoveally. Although this rate of priming is lower than that shown by the subjects who were not inattentionally blind, it is significantly greater than the base rate assessed in a control group who were simply asked to generate stem completions off the top of their heads.

A companion experiment substituted a recognition test for the stem-completion test, in which subjects were forced to choose among five alternatives, each of which began with the same stem as the target. This procedure yielded 47% correct recognition of the target word, against a chance expectation of 20%. Although this might be taken as evidence that the ostensibly blind subjects actually saw the words after all, it is also consistent with the hypothesis that the perceptual and conceptual fluency that comes with priming can serve as a cue for recognition by a feeling of familiarity.

The stem-completion studies showed that repetition priming, based on physical similarity, could occur in inattentional blindness. But what about semantic priming? In another set of experiments, Mack and Rock followed the critical trial with an array of five pictures, and asked the subjects simply to pick one -- any one, arbitrarily, with no constraints. For example, if the stimulus word was flake the pictures were of a snowflake, a flame, a flag, a flash camera, or a flat car tire. All of the subjects saw the stimulus word in the full attention condition, and all but one of them chose the picture corresponding to that word. By comparison, a yoked control group of subjects who never were presented with the stimulus at all chose the "correct" picture only about 12% of the time. The difference between the full-attention and control subjects is a priming effect, and because the priming is from a word to a picture, the priming is not perceptual, but rather semantic, in nature. Most subjects (72%) also saw the target word in the divided attention condition, and most of these (80.6%) also chose the picture corresponding to the word. The remaining subjects failed to see the target word, despite the fact that they were looking for it, but 43% of them still chose the "correct" picture when asked to choose. Of course, the most interesting data involves the inattention trials, where almost half of the subjects, 48%, missed the target -- showing the IB effect. Nevertheless, 48% of these subjects chose the "correct" picture, compared to 77% of those who actually saw the word. In another experiment, in which the target word was masked on the inattention trial, choice rates were 42.1% for those who were inattentionally blind, as opposed to 85.7% who were not. The "hit rate" is substantially reduced for inattentionally blind subjects, compared to subjects who detected and identified the stimulus in any of the experimental conditions -- full attention, divided attention, or inattention. But it is still substantially greater than the chance performance of the control subjects who were never presented with the stimulus at all. In other words, inattentionally blind subjects give evidence of semantic priming.

Mack and Rock (1998) also offer preliminary evidence for inattentional deafness and tactile insensitivity, roughly analogous to deaf hearing (Garde & Cowey, 2000) and numbsense (Rossetti et al., 2001), but do not report any studies of priming or other aspects of implicit perception in these domains.

Inattentional Blindness and the Attentional Bugaboo

Mack and Rock (1998) began their program of research with the view that conscious perception requires attention, and thus that inattention produces a kind of blindness -- a state of "looking without seeing" (p. 1). From their point of view, attention is "the process that brings a stimulus into consciousness... that permits us to notice something" (p. 25). Accordingly, there is "no conscious perception at all in the absence of attention" (p. 227). Objects may capture attention involuntarily, or attention may be deliberately directed to an object (by virtue, for example, of expectations, intentions, or mental set); in either case they enter into phenomenal awareness. But even in the absence of attention, some degree of perceptual processing appears to take place unconsciously, outside the scope of conscious attention, before attention is directed to the object, while attention is directed elsewhere. "[B]ecause attention requires an object", they write, "some percept, if only a minimal one, must exist prior to the engagement of attention". This preattentive, preconscious representation of the stimulus, revealed by repetition and semantic priming, is another facet of implicit perception.

But priming isn't evidence of implicit perception unless Mack and Rock are right that attention is required for consciousness, and unless the ostensibly blind subjects really failed to see the stimuli. And there's the rub: neither postulate is necessarily true. Writing in the tradition of Eriksen (1960) and Holender (1986), Dulany (2001) emphatically denies both points. In the first place, he questions the intimate association between attention and consciousness asserted by standard information-processing views of cognition (see also Dulany, 1997). As he points out, when listeners focus their attention on the soloist in a concerto, the rest of the orchestra doesn't vanish from consciousness. It fades into the background, to be sure, but it does not disappear entirely: they are still aware of it (not least because a concerto involves a complex interplay between the soloist and the orchestra). So, even though Mack and Rock's subjects may have been paying attention to the crosses that comprised their manifest task, they may well have been conscious of the unattended stimuli as well.

But what about the subjects' reports that they did not, in fact, see the unattended stimuli. Here, frankly, Dulany simply doubts that these reports are correct -- or, perhaps more precisely, he finds no reason to think they are correct, because Mack and Rock's methods for assessing conscious awareness are "deeply flawed" (p. 3). After all, in their very first experiments, a solid three-quarters of the subjects noticed the interloping stimulus: maybe the rest did, too, but simply didn't report it. Perhaps they thought they shouldn't have noticed it: if they were supposed to be paying attention to the cross, admitting that they also saw the square might mean that they weren't doing their job. Or perhaps they thought that the experimenter didn't want them to have noticed it: if the experimenter is really interested in the square, and not the cross, then reporting the cross means that the subjects have seen through the experimenter's deception. As Dulany points out, the demand characteristics of the experimental situation (Orne, 1962) clearly imply that the subjects shouldn't pay attention to the extraneous stimulus, even if they do.

The term "demand characteristics" was introduced into psychological discourse by Orne (1962), as part of a broad critique of methodology in social-psychological research (see also Orne, 1969, 1972, 1973, 1981). By "demand characteristics", Orne meant the totality of cues available in the experimental situation that communicate to the subject the experimenter's design, hypotheses, and predictions {for an analysis, see Kihlstrom, 2002). Many social psychologists employ deception to hide their real intentions, but as Mack and Rock's experiments show, social psychologists are not the only ones to do so. By fabricating the cover task of comparing the lengths of crossed lines, when they are really interested in identification of the extraneous stimulus, Mack and Rock engaged in no less a deception than did Milgram (1963) in his famous experiments on obedience to authority. And just as Milgram's subjects may have caught on to his deception (1968), so Mack and Rock's subjects may have caught on to theirs.

Certain details of Mack and Rock's procedures and results lend weight to Dulany's critique. For example, in a variant of the stem-completion priming experiment, Mack and Rock asked inattentionally blind subjects to choose the presented word (flake or short) from a set if possibilities, all of which began with the same three letters (pp. 179-181). The rate of stimulus recognition, 47%, was virtually the same as the rate of priming observed in the earlier stem completion experiments. Perhaps priming gave rise to a feeling of familiarity that enabled subjects to correctly guess the stimulus, in the absence of conscious recognition, as occurs in cases of amnesia (see Chapter 6). But it is also possible that the subjects consciously noticed the stimulus, despite their claims to the contrary, and that this conscious attentional processing gave rise to both the priming effect and the recognition performance.

Perhaps the most troubling aspect of Mack and Rock's results is that they found that the rate of inattentional blindness more than doubled (from roughly 25% to as much as 80%) when the location of the extraneous stimulus was moved from the parafoveal region to the point of fixation. Although selective attention can be centered on objects as well as on regions of space (Duncan, 1984; Humphreys et al., 1996), the claim that subjects can fail to notice a stimulus that is literally right in front of their eyes strains credulity. At the same time, Mack and Rock obtained the same rate of inattentional blindness, 48%, and roughly the same rate of semantic priming (42% vs. 48%), when the stimulus array was followed by a mask intended to eliminate conscious perception of the stimulus. Either the mask was ineffective, or the subjects in the unmasked condition also failed to consciously perceive the stimulus word.

Mostly, however, Dulany's problems with Mack and Rock's demonstrations of inattentional blindness, and of implicit perception in inattentional blindness, are conceptual:

As in the other experiments, on preattentive processing, it always remains possible that subjects may have shifted their conscious attention from the cross to the stimulus. If Mack and Rock had imposed controls to prevent such attentional shifts, the subjects might have failed to notice the stimulus, but priming would have disappeared as well.
Even without deliberately shifting their attention, subjects may have experienced "a fleeting literal awareness of the unattended critical stimulus, or only a fragment of it" (p. 4), and then immediately forgotten it. Along these lines, Wolfe (1999) has suggested that inattentional blindness would be better called inattentional amnesia (see also Coltheart, 1999; Moore, 2001). Under these circumstances, the priming effect might serve as an index of implicit memory, because the stimulus was consciously processed to some degree, and then subsequently forgotten; but it would not serve as a memory of implicit perception.
Even if subjects devoted only a small amount of attention to the extraneous stimuli, as might be the case with a brief attentional shift or fleeting perceptual awareness, this might well be enough to produce priming effects -- especially priming effects on stem completion, which are relatively independent of the level of processing at the time of encoding. Because stem completion is in some sense an "easier" memory task than cued recall or even recognition, it would be possible for this minimal attentional processing to show up later in a priming effect; but strictly speaking, the priming effect would not be evidence of inattentional processing.
The procedures for assessing subjects' perceptual awareness may have been insufficient for the task. As Dulany notes, Mack and Rock merely asked, on the critical inattention trial, whether subjects saw "anything on the screen on this trial that had not been there on previous trials?" (Mack, 1998, p. 13). If subjects saw the stimulus on the inattention trial, they may well have inferred that it had been present on the previous trials as well. If so, they would have said "No", and appeared to be inattentionally blind. But if subjects had been asked whether they saw anything on the current trial that they had not seen before, their answers -- and our conclusions about priming in inattentional blindness -- might be different.

As attractive as the notion of inattentional blindness might be, it has to be said that there are as yet no good answers to many of Dulany's criticisms. To a great extent, the intransigence of Dulany's arguments stems from the fact that, like Eriksen before him, he resists the identification of conscious awareness with introspective self-report. (In fact, Dulany was a graduate student of Eriksen's at the University of Illinois before joining him on the faculty there. In their one joint publication, Dulany and Eriksen (1959) argued that verbal self-reports were no more than accurate than covert psychophysiological responses as indices of stimulus discrimination.)

Dulany expresses a positivist's reserve about introspection, but he is no radical behaviorist. Far from denying the existence or importance of consciousness, Dulany has asserted a thoroughgoing mentalism that identifies consciousness with symbolic representation, and he has defined symbolic processing so broadly -- encompassing everything that happens after sensory transduction and before motor transduction (Dulany, 1991, 1997). Some symbolic codes are literal, close to a perceptual representation, while others are symbolic, more abstract and full of meaning, but none of them are any more or less conscious than the others because from his point of view all symbolic codes are represented in consciousness.

In the end, Dulany adheres to the "grand principle" {Dulany, 1997, p. 184} that "mental episodes consist of nonconscious operation upon conscious intentional states, yielding other conscious intentional states". In other words, mental processes may be unconscious, but mental contents are conscious. This position comports well with the traditional view, in cognitive psychology, that procedural knowledge is unconscious, inaccessible to introspective conscious access in principle, but that declarative knowledge is accessible to conscious awareness. But a major thrust of this course is that mental contents -- percepts, memories, and thoughts -- can also be unconscious, in the sense that they can interact with ongoing experience, thought, and action in the absence of, or independent of, conscious awareness. That is what implicit memory, implicit perception, and other, cognate phenomena (yet to be discussed) are all about.

As it happens, Dulany disputes much of this evidence as well. Although he has little to say about implicit memory, he is critical of evidence for both subliminal perception and blindsight (Dulany, 2001). And he has been critical of evidence for implicit learning as well (Dulany, 1991; Dulany et al, 1984, 1985). In lieu of the explicit-implicit dichotomy, Dulany (1997) prefers to contrast deliberative and evocative "mental episodes" (p. 179), but this is not a quarrel over terminology, like the one we encountered over explicit vs. implicit, direct vs. indirect, or declarative vs. procedural (or nondeclarative) memory. The proponents of all these terms accepted some sort of distinction between conscious and unconscious memory, but Dulany does not: "We should identify the explicit with deliberative mental episodes and the implicit with evocative mental episodes..., but one is no more conscious or nonconscious than the other (p. 179). Deliberative mental episodes are propositional in nature, representing the belief that some object or feature exists in reality; evocative episodes, by contrast, are non-propositional in nature, conveying only "a sense of" something (p. 185). Evocative episodes occur as a result of involuntary activation and association, while deliberative episodes reflect inference, decision, and judgment; but in Dulany's view both types of mental episodes are conscious.

It is possible that Dulany is right, that only mental processes are unconscious, and that mental states, as symbolic representations, are always conscious. But there is a danger here, and it is of the same kind that Eriksen's arguments posed for studies of subliminal perception. Eriksen identified consciousness with discriminative behavior; but if the evidence for unconscious percepts, memories, and the like consists in observations of discriminative behavior in the absence of self-reported awareness, he has ruled the psychological unconscious out of existence by definitional fiat. Similarly, Dulany identifies consciousness with symbolic representations; but if the evidence for unconscious percepts, memories, and the like consists in observations like priming outside of awareness, then he too has ruled the psychological unconscious out of existence. We are left with the traditional view in cognitive psychology, which identifies the unconscious with automaticity and procedural knowledge, and consciousness with all mental states and contents and states.

Varieties of Attentional Blindness

Like Eriksen's (1960) before them, Holender's (1986) critique of preattentive semantic processing and Dulany's (1997) criticisms of inattentional blindness are not trivial, and should have the effect of forcing researchers to clean up their methodologies, putting evidence for preattentive, preconscious processing on ever-firmer empirical grounds. This exactly what happened with research on subliminal perception (Greenwald et al.,1996; Draine & Greenwald, 1998). In the meantime, other investigators have opened several new lines of research on the general questions of attention, conscious awareness, and preattentive, preconscious processing. The studies described previously in this chapter examined processing when subjects' attention was directed elsewhere -- to one channel in dichotic listening, to foveal rather than parafoveal regions, to some objects but not others. By contrast, these studies document anomalies of awareness even when subjects are directing their attention right at the target. For this reason, I classify these studies as concerned with attentional blindness -- deficits in conscious visual processing that occur despite the subject's appropriate deployment of attention, and accurate expectations (see also Kanai et al., 2010).

Paralleling the conclusions of Mack and Rock (1997) concerning inattentional blindness, we can conclude from the phenomena of attentional blindness that Attention does not guarantee conscious perception. But again, in the present context, we are interested not so much in the blindness itself -- though that's interesting. What we're interested in is whether there is unconscious, implicit perception of the stimuli in question. Can you get priming from a stimulus for which a subject is attentionally blind?

Repetition Blindness

The phenomena of attentional blindness are most commonly observed in laboratory conditions employing a technique called rapid serial visual presentation, or RSVP (Potter, 1969; Forster, 1970), in which subjects view a series of visual stimuli presented one after another in very quick succession -- about 100 milliseconds each. In an early experiment by Kanwisher and her colleagues, subjects viewed sentences such as It was work time so work had to get done, presented one word at a time, and then were asked to report verbatim what they had seen. Note that the sentence contains a repetition of the word work. We can call the first instance of the repeated word R1, and the second instance R2. Under RSVP conditions, streaming about 10 words per second, a large number of subjects -- about 60% -- omit R2: they report "It was work time so had to get done", despite the fact that the sentence is obviously incomplete (Kanwisher, 1987). When R2 is not the same as R1, it is missed only about 10% of the time.

The procedure here is a little tricky: because most skilled readers do not read every word in a sentence, there is a tendency for them to fill in the gaps with meaningful words -- just as we generally ignore misspellings, missed words, and other typographical errors. To counteract this tendency, Kanwisher carefully instructed subjects to report what they saw verbatim. And she also included a number of sentences that actually had words missing, so that subjects would not be surprised when they reported an incomplete or anomalous sentence.

A later experiment by Kanwisher and Potter(1990, Experiment 1) makes the effect even clearer. In this study, subjects saw three different kinds of sentences.

In the repeated condition, the sentence was a standard RB stimulus, such as We were anxious for autumn well before autumn arrived.
In the synonym condition, R1 was replaced by a synonym, as in We were anxious for fall well before autumn arrived.
In the unrepeated condition, R1 was replaced by a semantically unrelated word that made sense in the context of the sentence, such as
We were anxious for apples well before autumn arrived.

Across the three conditions, subjects correctly reported R1 approximately 94% of the time; however, they reported R2 only 40% of the time in the repeated condition, compared to an average of about 88% for synonyms and unrepeated condition. The fact that RB occurs for a repetition, but not a synonym, indicates that RB occurs at the level of a perceptual (or possibly lexical) representation, but not at the level of meaning.

RB occurs when one instance of a word -- a token of the type -- appears fairly quickly after the first -- within one or two words, in fact, corresponding to a lag of about 200-300 milliseconds at a presentation rate of eight words per second. If the repeated word is delayed by about 500 milliseconds, either by positioning the word later in the sentence or by slowing the rate of presentation, RB does not occur.

Such findings led Kanwisher (1987, 2001) to propose a type-token individuation hypothesis of RB. According to this hypothesis, conscious awareness requires both the activation of a memory representation of an event (the type) and the individuation of that activated structure as the representation of a discrete event (a token of the type). The first time the subject sees the word autumn, the perceptual representation of that word is activated and a token individuated; but when autumn is repeated so quickly, individuation does not occur, and so the subject does not see it as a different, discrete event -- as another token of the type. The event need not be a word: RB has also been observed for digits and letters, pictures, and simple and complex shapes -- including impossible figures

According to Kanwisher, the lack of individuation creates a failure of conscious perception: the subject is not consciously aware of the token's repetition. But the claim that the second token is not consciously processed begs the question of whether it is unconsciously processed. Unfortunately, this is not an easy question to answer. In principle, we could find evidence of unconscious processing in the form of a priming effect of the second token -- as in Mack and Rock's work on inattentional blindness. (Of course, repetition blindness is itself a form of negative priming, in which processing the first token impairs processing of the second token [Neill & Mathis, 1998]. But what is at issue in the present context is priming (positive or negative) from the repeated token to a third stimulus.) In practice, however, the extremely short delays between the first and second token would seem to make it impossible to distinguish priming by the second token from priming by the first. For example, subjects might miss the second token of autumn in the first sentence above, and then show a priming on perceptual identification for yet a third token of autumn, or for an orthographically related word like author or a semantically related word like summer, but it would seem difficult to know whether the priming effect stemmed from the that token, as opposed to the first one, which had been processed consciously.

A clever attack on this question was mounted by Morris and Harris (2004), who made use of the finding that RB extends to orthographic neighbors of the initial word (Morris & Harris, 1999).

So, for example, in the sentence When my sister eats beef all beer tastes awful, substantial RB occurs for the word beer -- only 44% recognition, compared to more than 91% for beef.
By contrast, in the sentence When my sister eats rice all beer tastes awful,
recognition of beer is at 81%, compared to the 92% for rice.

Although RB is usually thought to occur at the level of the lexical representation, orthographic RB indicates that it can also be mediated at the sub-lexical level of individual letters or letter clusters. RB occurs to the extent that letters, or clusters of letters, overlap between R1 and R2.

In their experiments, Morris and Harris asked their subjects to read RB-inducing sentences, followed by a lexical decision task in which the target was an orthographic neighbor of a word covered by RB.

For example, the subjects were asked to make lexical decision about the letter string term after reading a "repeating" sentence like I hope our team next term is good, which induces RB for term,
and after reading a "nonrepeating" sentence like I hope our judge next term is good, which does not induce RB.

The same amount of priming occurred for repeating sentence, in which team induces RB for term, as in the nonrepeating sentence, in which there is no RB. Morris and Harris cite unpublished research showing similar priming effects on stem-completion.

How can we be sure that the priming occurs from the orthographic repetition term, and not from the initial team? Morris and Harris argue against such a "cohort activation" account based on the pattern of error rates observed in their study. But the error rates are very low to begin with, ranging from 1-3%, and the differences between them are not always significant. Still, in this sort of situation priming effects will dissipate quickly with time and the interference caused by intervening words. If the priming were a product of R1 and not R2, we would expect weaker effects in the repeating than in the nonrepeating condition. And, by the same token, if the priming were a product of activation from R1 combining with activation from R2, we might expect even stronger effects in the repeating condition. In fact, the levels of priming are equivalent, so the priming seems to be coming from R2 alone. RB abolishes explicit perception of the repeated item, but it does not abolish implicit perception, as indexed by priming effects.

The Attentional Blink

Another phenomenon of attentional blindness is known as the attentional blink (AB), -- a phenomenon initially noticed by Broadbent and Broadbent (1987), but given its name by Raymond, Shapiro, and their associates (Raymond et al., 1992; Shapiro et al., 1994; for a review, see Shapiro, 1994, 1997}. In AB experiments, subjects view a series of stimuli (such as letters) presented in rapid succession -- say, 30 items over a period of a three seconds, or a rate of about 100 milliseconds per item (this is roughly the same rate as in RB experiments). On each trial, the subject's task is to report the identity of a distinctive target letter -- for example, the one printed in red as opposed to black. However, the subjects are also asked to detect whether a second probe -- for example, the letter "R" -- also appeared in the letter series. Of course, the presence of the primary target (T1) and secondary probe (T2), and their locations within the string of letters, varied from trial to trial. Control subjects performed the probe-detection task only, without the target-identification task. In a typical experiment (Shapiro et al., 1994), control subjects showed a high level of accuracy, approximately 85%, regardless of where in the sequence T2 was placed. However, the experimental subjects missed T2 about half the time when it was presented within about 500 milliseconds (i.e., within five items) after T1. If T2 was presented six to eight items later, outside the 500-millisecond interval, detection rates for experimental subjects were essentially the same as those of the controls.

Of course, the attentional blink is not an actual eyeblink: subjects rarely blink their eyes during an RSVP trial. Nor is the attentional blink a phenomenon of visual masking, because it does not occur if the subject is not instructed to identify (i.e., pay attention to) the preceding T1. Instead, it is as if the subject has blinked, by analogy to the actual eye-blinks that occur after subjects shift their gaze from one location to another. Shapiro and Raymond suggested that T1 is identified preattentively, on the basis of its physical characteristics -- e.g., that it is red as opposed to black. Attention is then directed toward T1 in support of the identification process (e.g., that it is a T rather than an L), but it takes time for attention to shift back to the stream of RSVP stimuli, giving T2 time to be lost through decay or interference. If T2 occurs soon enough after the target, then, it is likely to be missed. However, T2 probe is not missed if the subject is instructed to ignore T1 -- thus supporting an attentional interpretation of the effect.

Although AB is not an actual eyeblink, it is nonetheless a real blink. Sergent and Dehaene (2004) asked subjects to rate the visibility of the T2 at various lags after T1. After presentation of T1, the visibility of T2 decreased immediately at a lag of 2 (that is, with only a single stimulus between T1 and T2, and returned to baseline immediately at a a lag of six (that is, with five stimuli between the two targets). The disappearance and reappearance of T2 occurred in a discontinuous, all-or-none fashion, with not even a hint of continuous degradation and recovery. This pattern was quite different from that observed with backward masking, which produced an immediate degradation of the preceding stimulus (that is, at a lag of 1), and a gradual increase as the SOA lengthened. Sergent and Dehaene suggest that their results are most compatible with a two-stage model of the AB in which each stimulus in the RSVP display is subject to automatic, preconscious pickup, but conscious processing of T1 consumes cognitive resources that are necessary for conscious required for conscious processing of T2 (Chun & Potter, 1995).

If attention blinks, as it were, so that the probe is not consciously perceived, then the obvious question arises: is it perceived unconsciously, or is it just not processed at all? For example, can we observe priming by the probe despite the attentional blink? An experiment by Shapiro and his colleagues addressed this question by combining the attentional blink with repetition blindness (Shapiro et al., 1997). Subjects viewed a sequence of letters and numbers presented via RSVP, with instructions to identify the digit printed in white, among a stream of digits printed in black. They were also asked to indicate whether they saw a probe letter appearing in the sequence of digits. In fact, two probe letters appeared on each critical trial: the first was printed in uppercase, such as E; the second was either a matching (e) or non-matching (n) letter printed in lowercase. The uppercase probe was positioned with respect to the target so as to be within the attentional blink, while the lowercase probe was positioned with respect to the uppercase probe so as to be subject to repetition blindness.
The logic of the research depends on the conclusion, from studies by Kanwisher and her colleagues, that repetition blindness is mediated by representations that are somewhat more abstract than the orthographic level -- that is, the two tokens need not be physically identical. Recall that, in Kanwisher's analysis, RB occurs by virtue of conscious perception of the first token; in the absence of conscious perception, RB should not occur for the second token. Thus, conscious perception of the first token E would produce RB (negative priming) for the second token e, because both are lexical representations of the letter E, even though they are physically dissimilar. However, if attentional blindness blocks conscious perception of the first token E, then RB should not occur for the second token e. But if attentional blindness permits some processing of the first token E, albeit outside of perceptual awareness, then subjects should show positive priming in identification of the second token e. Thus, in the context of attentional blindness, priming provides evidence of implicit perception.

And that is what Shapiro et al. (1997) found. In their first experiment, subjects correctly reported the identity of the target digit on 84.2% of the trials. But the first token of the letter probe was correctly identified on only 67.5% of the trials -- a reduction that presumably reflects the attentional blink. When the first probe was correctly identified, matching probes were identified less frequently than mismatching probes: this is an instance of repetition blindness. But when the first probe was not correctly identified, presumably due to the attentional blink, matching probes were identified more frequently than mismatching probes. The occurrence of this priming effect the priming effect means that despite the attentional blink, the first probe was processed to at least some extent outside of conscious awareness. In other words, priming from the attentional blink is another example of implicit perception.

Note that the effect observed by Shapiro et al. (1997) goes beyond repetition priming, because the lowercase target was not physically identical to the uppercase prime. This means that the priming had to be mediated by a higher-level, more abstract, lexical representation (e.g., of the letter e) than by a perceptual representation of the letter printed in some particular case or font. But priming in the attentional blink can be mediated by semantic representations as well. For example, in a second experiment, Shapiro et al. substituted words for letters. The target was a word such as river, the first probe (covered by the attentional blink) was a word such as doctor, and the second probe was a word such as nurse, semantically related to the first probe but not to the target. In this experiment, targets were correctly identified on 87.8% of the trials, but the initial probes were identified on only 46.3% of trials -- a strong effect of the attentional blink. In this experiment, because the target and first probe are not tokens of the same type, no repetition blindness would be expected in those instances where the first probe was correctly identified. In fact, there was a strong semantic priming effect. The second probe was much more likely to be correctly identified when it was semantically related to the first probe. But semantic priming also occurred when the first probe was not correctly identified due to the attentional blink. The effect was weaker, but it was significant, and shows that implicit perception in the attentional blink can extend to semantic as well as lexical representations.

Shapiro et al. (1997) also discovered that the attentional blink disappeared when the probe consisted of the subject's own first name, compared to other names or common nouns. However, AB did occur for subsequent other-name probes when the subject's own name was the target.

Additional evidence for semantic priming during the attentional blink was provided by Maki and his colleagues (Maki et al., 1997), in a series of experiments employing a variant of RSVP with words}. The variant was to include a prime, semantically related to T2, among the distractors interposed between T1 and T2 -- for example, nest-bird. In some experiments, the prime appeared immediately after T1; in others, it appeared immediately before T2; control trials employed semantically unrelated primes (e.g., chill-bird). Figure 11. shows the basic results of the experiment (Maki and his associates analyzed their data in a number of different ways, always with the same outcome). Recall of T1 in their experiments was approximately 75%. When T2 was presented outside the boundaries of the AB, with 6 intervening items (thus a lag of 600 milliseconds), recall of T2 was very close to this figure, approximately 74%. But when T2 was presented with only 2 intervening items (200-300 msec), recall of T2 dropped to about 16%: this is the attentional blink. (he exact figures are 74.63%, 74.37%, and 15.70%, and respectively. For the record, recall of the prime was uniformly poor, 3.33%. Figures calculated from the 2- and 6-lag conditions of Experiments 2, 4, and 5, collapsed across the associated and unrelated conditions, as presented in Tables 2, 4, and 5.) Nevertheless, Maki et al. found an associative priming effect, such that presentation of a semantically related word inside the attentional blink actually increased the likelihood that T2 would be recalled. Note that the effect is actually greater for primes presented well within the temporal boundaries AB, than for primes presented on its edges or outside it.

The priming effect observed by Maki et al. is small, and it is smaller than the corresponding effect that occurs when the clearly identified T1 is semantically related to T2 (which is what they found in Experiment 1). And it is also short-lived, lasting only about 200 milliseconds. But it is statistically significant. The fact that associative priming occurs at all indicates, as the authors put it, that "Word meaning does survive the AB" (p. 1028) -- if only by the skin of its teeth. Thus, the attentional blink does seem to be another phenomenon in which implicit perception occurs. Despite the blink, which prevents the subject from being perceptually aware of whatever occurs while the attentional gate is shut, information about those events is processed, at least to some degree, nonetheless, and influences subsequent task performance at least for a short period of time.

Change Blindness

Another variant on attentional blindness is change blindness (CB) , which occurs when subjects fail to notice rather large alterations in familiar scenes (Rensink, 1997; for comprehensive reviews, see Rensink, 2004; Simons, 2000; Simons, 2005; Simons, 1997). In the flicker paradigm, subjects are presented with a picture of a natural scene rapidly alternating with a slightly modified picture in which there has been some change in the presence or absence of some object or feature, its color, orientation, or location -- about 240 msec per stimulus. In this case, subjects will readily detect the change, apparently because the alteration creates a "luminance transient" that draws attention to the spatial location in which the change has occurred. But if the two images are separated by a blank gray field, presented for only 80 msecs, images the situation is a little different. In the laboratory, the difference between the two stimuli is typically restricted to only a single modification in appearance (i.e., the presence or absence of some object or feature), color, or location. The two versions alternate back and forth on a random schedule, each version appearing for only a very short time (typically 240 msec), and separated by a gray field that appears for an even shorter period of time (typically 80 msec). On a single trial, presentation continues in this manner for 60 seconds, or until the subject detects the change. The purpose of this procedure is to simulate the situation in real-world vision, in which the observer must integrate information across saccadic eye movements -- in this case, simulated by the gray field interposed between the two versions of the stimulus. In the absence of the gray field, subjects will immediately notice the change

Change blindness falls under the rubric of attentional blindness because the subject is not aware that the second image is different from the first. That is a lapse of consciousness. But in the present context, the really interesting question is whether subjects can show priming from the unnoticed element of the second image. As of April 2013, to my knowledge, no such demonstration has been reported (hint, hint).

Persisting Issues

I prefer the term implicit perception to the older term subliminal perception because of the definitional parallels with implicit memory, but also because the domain of unconscious perception clearly extends beyond the traditional concept of subliminal perception:

Stimuli which, by virtue of "subthreshold" energy levels, are truly subliminal.
Stimuli which have been rendered notionally "subliminal" by virtue of presentations, for extremely brief periods, of stimuli that would have been clearly supraliminal if presented at longer intervals.
Stimuli which, by themselves, are supraliminal in terms of both the intensity and duration of presentation, but which have been rendered notionally "subliminal" by virtue of some form of masking.
Stimuli which are "supraliminal" by every normal standard, but are presented to neurological patients with disorders in the same sensory or attentional system, as in cases of blindsight and neglect.
Stimuli which are "supraliminal" by every normal standard, but are presented to neurologically intact subjects under conditions of degraded attention, as in cases of inattentional blindness or the attentional blink.

By far most of the experimental literature involves either subliminal or unattended stimuli, and there are two big issues here.

One issue, initially raised by Erikson and Goldiamond in the late 1950s and early 1960s (in the wake of a spate of claims of subliminal perception produced by the "New Look" theorists), has to do with the nature of thresholds. Merikle and his colleagues have responded by distinguishing between two thresholds, the subjective and the objective. The subjective threshold is the point at which the subject's ability to consciously detect a stimulus falls to chance levels; the objective threshold is the point at which all discriminative response to stimulation falls to chance levels. In general, implicit perception effects are strongest when levels of stimulation are closer to the subjective than to the objective threshold.

Another big issue is the nature of implicit perception -- that is, whether any priming effects observed are confined to perceptual processes (e.g., repetition priming), or can extend to the conceptual (e.g., semantic priming). Critics like Holender and Shanks & St. John argue that semantic processing without conscious identification is not possible. However, other investigators, such as Greenwald and Merikle, regard this as an open question. Greenwald has been particularly vigorous in asserting that semantic processing is possible in the absence of conscious identification, but at the same time he is quite clear that this processing is extremely limited. So, for example, Greenwald has shown masked priming based on the connotative meaning (i.e., affective valence) of single words (e.g., ENEMY primes BAD) but not of two-word phrases (e.g., ENEMY LOSES, a positive phrase composed of two negative words, does not prime GOOD). The analytic limitations on masked priming imply that more dramatic demonstrations of subliminal influence, such as subliminal suggestions to stop smoking, or the subliminal effects, claimed by Silverman and his associates, of the "symbiotic stimulus" MOMMY AND I ARE ONE are -- frankly -- dubious.

It should be noted that while subliminal perception may be analytically limited, this need not be true for other cases of implicit perception, such as those involving unattended stimuli or hypnosis. In these cases, especially the latter, there are more cognitive resources available to process the stimulus, and consequently the priming effects may extend to the complexly semantic.

In considering the relations between explicit and implicit memory and explicit and implicit perception, priming during general anesthesia may be a sort of limiting case. Memory in the present requires perception in the past, and so any evidence that surgical events are retained in memory means that they were processed perceptually at the time they occurred. By definition, adequately anesthetized patients lack all explicit memory for surgical events; however, there is some (and slowly mounting) evidence that, at least under some circumstances, implicit memory can be spared. At the same time, it appears that, to all intents and purposes, the adequately anesthetized patient is unconscious. So, whatever implicit memory persists after anesthesia must have its source in implicit perception during anesthesia (this does not logically follow, as the patient's unconscious appearance may be deceiving, and in fact there are competent anesthetic procedures, known generically as _conscious sedation_, in which the patient is consciously aware of surgical events at the time they occur, but does not consciously remember them afterward).

Still, assuming (even just for the sake of argument) that implicit perception is possible during anesthesia, the effect must be analytically limited in the extreme. Currently available demonstrations of implicit memory (implying implicit perception) for surgical events are largely limited to effects such as repetition priming, which can be mediated solely by perceptual or other presemantic representations. It remains to be seen whether semantic priming can occur during adequate general anesthesia, but such an effect would seem to be extremely unlikely given the extremely limited cognitive resources available to the adequately anesthetized patient.

Implicit Thought

The phenomenon of Implicit memory is well established within cognitive psychology, and the phenomenon of implicit perception only less so. However, the explicit-implicit distinction can also be extended to other domains of cognition.

Along with Dorfman and Shames, I have extended the explicit-implicit distinction beyond perception and memory to the domain of thinking and problem-solving (Dorfman, Shames, & Kihlstrom, 1996; Kihlstrom, Shames, & Dorfman, 1996).

Implicit Thought Defined

Again by analogy to implicit memory, we may define implicit thought as the effect of an idea on experience, thought, or action, in the absence of (or independent of) conscious awareness of that idea.

(By "idea" we also mean a mental state such as a thought or an image, itself neither an episodic memory or a percept).

As is the case with implicit memory and implicit perception, implicit thought is exemplified by priming and related effects. However, in this case, the priming is not attributable to an unconscious memory (a representation of an event in the person's past experience) or to an unconscious percept (a representation of an event in the person's present environment). Rather, the priming is attributable to a thought -- an idea, a concept, an image, etc. -- that is "on the person's mind", even though he or she is not consciously aware of it.

As with implicit memory and implicit perception, implicit thought is typically manifested in priming effects.

Priming and Free Association

Consider, first, the Remote Associates Test, introduced by Mednick & Mednick (1962, 1967) as a test of individual differences in creativity. In the test, subjects are presented with three cues words, and are asked to generate a target word which all three cues have as a common associate. For example:

DEMOCRAT

GIRL

FAVOR

The answer is, of course, PARTY. There is the Democrats who are members of the Democratic political party, there are people known as "party girls", and there are gifts given as party favors.

The RAT is not necessarily a good test of creativity, though you can see why the Mednick's thought it might be. But it's a great deal of fun at parties, and serves as the basis for Taboo, a party game sold by Parker Brothers.

Anyway, K.S. Bowers and his coworkers (1990) used the RAT as the basis for a "Dyads of Triads" task. In the DOT, subjects are presented with two triplets of words, one of which -- the coherent triad -- is united by a common associate and the other one -- the incoherent triad -- not. Their task is to respond with the common associate that unites the coherent triad; if they cannot do so, they are asked to guess which triad actually is the coherent one.

For example:

BLANK

WHITE

LINES

STILL

PAGES

MUSIC

The answer, of course, is Triad A. There is blank paper, and paper with lines on it, and white paper. Barring psychotically loose associations (or, perhaps, a bout of extremely high creativity), there does not appear to be any word that units Triad B.

The coherent triads used in the DOT are all soluble, but they are also fairly difficult, and so many otherwise intelligent people, with good verbal abilities, fail to solve many of them. Nevertheless, Bowers et al. found that subjects could reliably discriminate between coherent and incoherent triads, even when they could not generate the solution itself. Much as Peirce and Jastrow (1884) found that observers could distinguish which of two weights were heavier, and which of two lights were brighter, even though they could not detect any difference between them, Bowers et al.'s subjects were able to discriminate between coherent and incoherent triads, even though they were unaware of the solution to the coherent set.

In a followup study, Shames (1994) showed priming effects in lexical decision for the solutions to coherent triads, even though the subjects were unaware of the solutions. For example, the solvable unsolved triad

PLAYING

CREDIT

REPORT

primed lexical decisions for CARD, regardless of whether subjects actually found the solution itself.

The findings of Bowers et al. (1990) and of Shames (1994) seem to illustrate implicit thought. An idea -- namely the solution to the coherent triad influences both choice behavior and lexical decision, even though the idea itself is not represented in conscious awareness. Note that this is not a phenomenon of either implicit memory or of implicit perception.

The target, such as CARD, has not been presented to the subjects sometime in the past, giving rise to priming effects in the present. Because the target was not presented, it cannot have been forgotten, so the priming is not an implicit expression of memory for some past event.
Moreover, the triads themselves are completely visible to the subjects at the time of presentation, so the priming is not an implicit expression of perception, either.

The mental state that influences subjects' performance is not a memory, and it's not a percept, so there's only one thing left: it's an implicit thought.

Intuition in the Stages of Thought

The priming effects uncovered by Bowers and Shames have much in common with intuitive feelings -- where something "rings a bell", or "just seems right", even though we cannot say just why this is the case.

These intuitive feelings, in turn are related to the "Stages of Thought" described by Wallas (1926):

Preparation, including two somewhat different processes:

the accumulation of knowledge and the mastery of the rules which govern the particular domain in which the problem resides, as well as
the adoption of a problem-solving attitude, including the awareness that there is a problem to be solved, and the deliberate analysis of the problem itself.
If it seems that the problem can be solved by a well-known algorithm, the problem may be solved at this point, and the thinker moves straight to the verification stage. Otherwise, the thinker proceeds to work actively on the problem. But if the problem is very difficult, the thinker may give up, or at least take a break. Here is where things get interesting.

Incubation, in which the thinker is not consciously thinking about the problem, but problem-solving activity goes on instead, outside of consciousness.
Intimation, "the state of rising consciousness which indicates that the fully conscious flash of success is coming" -- otherwise known as intuition.
Illumination, in which the solution to the problem appears in the consciousness of the thinker -- otherwise known as insight.
Verification, in which the solution is confirmed and refined, or shown to be incorrect, in which the problem-solver returns to the preparation stage.

Still, intuition appears to be what is going on in Bowers' DOT task and Shames' priming task. The solution to the coherent triad has not yet risen to the level of conscious awareness, but unconsciously -- influences the subject's discrimination between coherent and incoherent triads, as well as the subject's performance on the lexical decision task. Intuitions arise in the period of incubation, and persist there as well, until they emerge into consciousness in the form of a full-blown insight.

These intuitive feelings are also controversial, with some theorists arguing that they are misleading more often than not, and other theorists believing that too much conscious thought can actually make for bad decision-making, and that people are better relying on their automatically generated intuitions. This more recent emphasis on the "power" of intuitive thinking reflects, to a great degree, the "automaticity juggernaut" discussed in the Lecture Supplement on "Automaticity and Free Will", and also the "affective counter-revolution" discussed in the Lecture Supplement on "Implicit Motivation and Emotion".

Incubation, or unconscious problem-solving, plays an important role in the mythology of problem-solving. Poincare famously claimed that the solution to a difficult mathematical problem came to him unbidden when, after working on the problem intensely, he gave it up to spend a few days at the beach. And Kekule famously claimed that not one but two important discoveries -- the general theory of chemical structure and the specific structure of the benzene molecule -- came to him in dreams (see A.J. Rocke, Image and reality: Kekule, Kopp, and the Scientific Imagination (2010).

Incubation is also one of the most controversial topics in the literature on thinking, reasoning, and problem-solving, and many researchers and theorists deny that it occurs at all. In the first place, the critics note, almost all the evidence for incubation is anecdotal. In fact, the standard view of incubation is that it doesn't exist. Rather, the critics argue, subjects "forget" inappropriate solutions over the retention interval, making it more likely that they will achieve the correct solution. Alternatively, they argue that, after a rest, subjects renew work on the problem with a more appropriate set, or enter the problem space at a new entry point. In none of these cases is there any sense of an unconscious solution gradually gaining strength until it emerges into consciousness.

Still, evidence for incubation effects in problem-solving has begun to accumulate in the laboratory. For example, subjects who are asked to solve the nine-dot problem often report that the correct solution occurs to them suddenly, after they have taken a break from trying to solve the problem.

In fact, a meta-analysis of incubation effects by Sio and Ormerod (2009) revealed a significant effect of incubation on solving a wide variety of problems: linguistic (like the RAT), visual (like the nine-dot problem), and creative (e.g., "How many uses can you think of for a brick?"). Incubation effects were especially strong when the subjects solved problems under low cognitive load, or none.

A particularly striking example of implicit thought in problem-solving was presented by Siegler and Stern (1998; see also Siegler, 2000) in a study of second-graders learning to solve "inversion" problems of the form

A + B - B = _____.

At the beginning, the children employed a "computational" strategy, adding and subtracting B mechanically. But eventually, they achieved the insight into the problem, and simply announced that the solution was A.

Siegler and Stern also asked the children, after each problem, how they had solved it. These reports represented their explicit problem-solving approach. But Siegler could also infer the children's implicit problem-solving approach by examining the time it took them to solve the problem. On average, it took the children more than 4 seconds to solve the problems computationally, but less than 4 seconds using the shortcut. What was of interest to them was that, sometimes, the children reported performing a mechanical computation even though their problem-solving time was very short. These students were clearly using the shortcut, but they were doing so unconsciously -- they were using the shortcut, but didn't know they were doing so.

In this graph, the trial on which each subject first used the unconscious shortcut, as indicated by a sudden drop in response latency coupled with a self-report of using computation, is labeled as Trial 0. Then Siegler looked at reported strategy use in the trials leading up to the insight, and in subsequent trials.

By definition, all the children reported using computation before Trial 0.
After Trial 0, most of the children continued to report using computation for several trials.
In fact, only by Trial 4 did 50% of the children report using the shortcut. Between Trial 0 and Trial 4, then, most children were using the shortcut unconsciously.
By Trial 5, 80% of the children reported using the shortcut. It was only at this point that the majority of the children were aware of how they actually solved the problem.

A parallel analysis focused on the first conscious use of the shortcut by these children -- that is, the trial where they actually reported using the shortcut. This trial is now labeled Trial 0.

In the trials leading up to Trial 0, about 80% of the children were actually using the shortcut unconsciously. The remainder correctly reported that they were using computation.
Once the children reported using the shortcut, they continued to use it.

To explain these effects, Siegler and Stern offered an unconscious activation hypothesis, that increasing activation of a strategy leads to its unconscious use, while further increases in activation lead to its conscious use. They found that presenting problems in blocked format, in which all problems of the same sort were grouped together, permitted children to discover the shortcut faster, compared to a mixed condition in which different types of problems were interspersed. Blocked presentation also reduced the delay between unconscious and conscious use of the shortcut, more consistent use of the shortcut, and also greater generalization of the strategy to other types of problems.

Implicit Thought and Metacognition

Metacognition may be defined as cognition about cognition, and has two aspects (Flavell, 1979; Nelson, 1992; Metcalfe & Shimamura, 1994; Reder, 1996):

knowledge of the contents of one's mind;
knowledge of the rules by which the mind operates.

For example metamemory includes the person's knowledge of the information stored in memory, and of the principles that guide the encoding, storage, and retrieval of memories.

In problem solving, intuitions are studied in the form of feelings of warmth (FOWs; Newell, Simon, & Shaw, 1962/1979), in which problem-solvers feel like they are getting close to the solution to a problem. In the General Problem Solver, FOWs increase as the distance between the current state and the goal state decreases.

In memory, intuitions take the form of feelings of knowing:

In the tip of the tongue state (Hart, 1965), subjects are presented with the definition of an obscure word, such as The secretion from the sperm whale used in the manufacture of perfume and must retrieve the word from semantic memory. Even subjects who do not generate the correct answer (ambergris), can often identify such acoustic features as the first letter of the word, the number of syllables, and where the accent is placed; when asked to list the words that come to mind while trying to generate the correct word, they often come up with words, such as asparagus, which sound like the target word. They are aware of features of the word, even if they are not aware of the word itself.
In the feeling of knowing (Nelson & Narens, 1980), subjects are presented with "Trivial Pursuit" questions such as What is the highest mountain in South America? Even though hardly anyone can answer this question, many subjects will say that they would recognize the answer (Aconcagua) if they saw it, and these judgments tend to be correct. They know they know the answer, even if they are not aware of the answer itself.

Implicit memory and implicit perception are often described as failures of metacognition, because people do not seem to know what they've perceived, or what they remember. But in a sense, implicit thought represents a success of metacognition: people know that they know something, even though they can't consciously think of it.

Unknown Knowns

Donald Rumsfeld, who as Secretary of Defense in the George W. Bush administration ("Bush 43"), helped lead America into the Iraq War, was famous for pithy sayings, some of which constituted a sort of poetry. They were collected as such by Hart Seeley in Pieces of Intelligence: The Existential Poetry of Donald Rumsfeld and set to music by composer Bryan Kong. Link to a story aired on National Public Radio. Among the most famous of these was a ditty called "the Unknown", from a news conference Rumsfeld held on February 12, 2002.

As we know,
There are known knowns.
There are things we know we know.
We also know
There are known unknowns.
That is to say
We know there are some things
We do not know.
But there are also unknown unknowns,
The ones we don't know
We don't know.

This epistemological adage gave the title to Rumsfeld's memoir, Known and Unknown: A Memoir (2013), and also to The Unknown Known, a film about Rumsfeld directed by Error Morris. The film's title is derived from an exchange between Morris and Rumsfeld, in which Morris points out that there's a fourth possibility: that there are things we know but we don't know that we know.*

As Maureen Dowd reported ("A Mad Tea Party", New York Times, 10/12/2013):

Rumsfeld doesn't even seem to understand his signature phrase. reading from a 2004 memo, he says "There are known knowns. There are known unknowns. There are unknown unknowns." He tells Morris that there are also unknown knowns. Things that you possibly may know that you don't know you know.

Morris challenges him: "But the memo doesn't say that. It says that we know less, not more, than we think we do."

Rumsfeld finally admits a boo-boo: "Yeah, I think that memo is backwards."

Actually, Morris has it wrong. He thinks that the concept of unknown knowns means that you know things that you really don't. But it doesn't. The concept of unknown knowns means, precisely, that there are things that we don't know that we actually know. This is exactly the idea behind the notion of unconscious thought. There are ideas which we have, and which influence our experience, thought, and action, but of which we are not consciously aware.

That's what implicit thinking is: thoughts -- knowledge and beliefs -- that we're not consciously aware of, but which nonetheless influence of conscious experience, thought, and action.

*Actually, Morris's probe was anticipated by Slavoj Zizek, a philosopher and psychoanalyst, in an opinion piece critiquing Rumsfeld's response to the revelations of abuse at Iraq's Abu Ghraib prison ("What Rumsfeld Doesn't Know That He Knows About Abu Ghraib", In These Times, 05/21/2004):

In March 2003 [apparently another occasion in which Rumsfeld used his "signature phrase"], Rumsfeld engaged in a little bit of amateur philosophizing about the relationship between the known and the unknown: "There are known knowns. These are things we know that we know. There are known unknowns. That is to say, there are things that we know we don't know. But there are also unknown unknowns. There are things we don't know we don't know." What he forgot to add was the crucial fourth term: the "unknown knowns," the things we don't know that we know -- which is precisely, the Freudian unconscious, the "knowledge which doesn't know itself," as Lacan [another psychoanalytically oriented philosopher] used to say.

Zizek is wrong to ascribe "unknown knowns" to the "Freudian unconscious". There's nothing Freudian about it -- no sexual and aggressive "monsters from the Id" (quoting a phrase from Forbidden Planet, my all-time favorite science-fiction movie). If there's an unconscious in operation there, it's precisely the non-Freudian unconscious -- just like implicit memory and implicit perception, except that what's unconscious isn't a memory or a percept -- it's a thought, a piece of knowledge or belief.

Implicit Learning

Actually antedating Schacter's distinction between explicit and implicit memory is Reber's distinction between explicit and implicit learning. Traditionally, learning is defined as a change in behavior which occurs as a result of experience (thus excluding behavioral changes which occur following maturation, injury, or the ingestion of drugs). And traditionally, learning has been classified as intentional rather than incidental, but this distinction occurs to the learning experience, not the subject's awareness of what s/he has learned. Following Schacter, we can define explicit learning as the acquisition of knowledge, skills, and rules accompanied by conscious awareness of what has been learned. Similarly, implicit learning can be defined as the acquisition of knowledge (etc.) in the absence of awareness of what has been learned, or even that the subject has learned anything at all.

Implicit Learning Defined

Traditionally, learning is defined as any relatively permanent change in behavior or knowledge that occurs as a result of experience. From the cognitive point of view, this behavioral change is mediated by the acquisition of new knowledge about the world. For cognitive psychology and cognitive science, the behavior is not as interesting as the knowledge that gives rise to it. So, in cognitive terms, we define learning as the acquisition of knowledge that occurs as a result of experience. The default assumption is that this learning is explicit -- that is, the person is consciously aware of what has been learned. Again by analogy to implicit memory, we may define implicit learning as the acquisition of new knowledge through experience, in the absence of conscious awareness of that knowledge, or of the learning experience, or both.

Now, it can happen that the person is also unaware of the learning experience. Because such learning experiences are episodes in the subject's personal history, a failure to remember a learning experience would count as a disruption of implicit memory.

Implicit Learning of Artificial Grammars

Implicit learning is commonly studied in an "artificial grammar learning" paradigm developed by Reber (1967), in which subjects are asked to memorize strings of letters which have been generated by an artificial 'Markov process" finite-state grammar which specifies which letters can appear in the string, how many times, and and in what order. The subjects are not informed of the existence of this grammar, however.

Over repeated study-test trials, Reber found that subjects could, in fact, learn such lists, and that they learned lists of grammatical letter strings better, and faster, than they did random strings. The subjects were obviously exploiting the grammatical structure of the strings to facilitate learning; but when queried, the subjects could not indicate what the grammar was.

In another version of the experiment, After reaching a criterion of learning, the subjects are informed that the studied strings had been generated by a grammar, and they are then asked to discriminate between new letter strings which were also generated by the grammar, and others which have not. Typically, subjects are able to make this distinction between grammatical and ungrammatical strings reliably above chance -- even though, again, they are unable to specify the grammar according to which the strings have been constructed.

Reber claims that while they were consciously trying to memorize the letter strings, they unconsciously induced the grammar which generated them.

Other Domains of Implicit Learning

Similar findings have been obtained in other paradigms involving the control of complex systems, sequence learning, and matrix scanning. In each case, subjects acquire new knowledge through experience, but are unaware of the knowledge they've acquired.

Implicit Learning in Amnesia

Implicit learning can also be demonstrated in amnesic patients as well.

The famous amnesic patient H.M. learned the pathway through a finger maze, and retained this knowledge over a period of up to 14 days.

Other amnesic patients learned to use a "pursuit-rotor" apparatus.

And still other amnesic patients learned to read mirror-reversed type.

Amnesic patients can also learn artificial grammars, in much the same way that normal subjects do. But in all of these situations, the patients do not remember their learning experiences. Nor do they display any awareness that they know what they know. Even after many learning trials, they don't recognize the apparatus; and when they sit down and try their hand at it, they are sometimes amazed that they perform so well.

In fact, Squire based his original equation of "explicit" and "implicit" with "declarative" and "procedural" on the observation of intact skill (i.e., procedural) learning in amnesic subjects (e.g., classical conditioning, pursuit rotor learning, mirror tracing, and artificial grammar learning). The problem is that priming is not necessarily procedural in nature, and may be mediated by declarative memory.

What is Learned in Implicit Learning?

Sometimes, implicit learning is characterized as procedural, as opposed to declarative. But while some implicit learning paradigms (like artificial grammar learning) can be represented in the "if-then" format characteristic of production systems, some of the knowledge acquired during implicit learning can be represented in declarative (propositional) format.

There has been considerable controversy over claims of implicit learning (Dulany, 1968, 1995; Shanks & St. John, 1994). Interestingly, many of the critiques of implicit learning seem to echo those applied to implicit perception. So, for example, it is possible that subjects' performance on the artificial grammar task is mediated by partial knowledge of the grammar. Alternatively, it may be that subjects distinguish between grammatical and non-grammatical test stimuli by comparing them to consciously remembered items from the original study list, without reference to any grammatical structure at all.

Implicit Cognition and Polanyi's Paradox

We know more than we can tell. That's Polanyi's paradox in a nutshell. Michael Polanyi (1891-1976), a Hungarian-British physicist, economist, and philosopher, was much concerned with the nature of human knowledge -- and, for that matter, with the limits of empirical science. In Personal Knowledge (1958), he argued that the scientific method was not guaranteed to yield truth, because all scientific investigations involve subjective judgments about the issue being studied, including whether and how to study it (you can read this as aDen anticipation of Thomas Kuhn's notion of scientific paradigms). More important, in the current context, in The Tacit Dimension (1966), argued that our knowledge and abilities are based on skills and rules that operate outside of conscious awareness, and which were acquired without conscious awareness through evolution and, especially, social learning. In this respect, Polanyi was being critical of the British empiricists, who argued that all knowledge begins with sensory data -- by which they apparently meant consciously perceivable sense data.

Polanyi had in mind the distinction that cognitive psychologists make between declarative and procedural knowledge. There's considerable agreement that procedural knowledge can guide behavior outside conscious awareness and conscious control. But the literature on dissociations between explicit and implicit cognition show that "tacit" knowledge extends beyond the procedural. In implicit perception, we perceive objects and events outside of conscious awareness; the priming effects that epitomize implicit memory involve declarative, not procedural memory. The knowledge acquired in implicit learning, likewise, is both declarative and procedural in nature. And the priming effects observed in implicit thought are also mediated by declarative knowledge structures.

Cognition is about knowledge: how it's acquired, how it's represented, and how it's used in the course of adaptive (and maladaptive) behavior. When it comes to knowledge, "We know more than we can tell" about so many things, and in so many different ways.

Beyond the Cognitive Unconscious

The cognitive unconscious consists of (1) automatic processes invoked in perception, mesmory, learning, and thought; and (2) unconscious percepts, memories, thoughts, and knowledge revealed by studies documenting dissociations between explicit and implicit memory, perception, thought, and learning.

But cognition isn't all there is to the mind. According to Kant, "knowledge, feeling, and desire" are separate and irreducible faculties of mind. Hilgard also writes of the "trilogy of mind": cognition, emotion (affection), and motivation (conation). If emotion and motivation are somehow separate from cognition, then we need to consider the possibility of unconscious emotion and motivation, separate from the possibility of unconscious cognition.

So having firmly established the existence and influence of unconscious memories and percepts, probably unconscious knowledge, and possibly unconscious thoughts, we now want to ask the question of whether we have unconscious emotions (feelings) and motives (drives). And that takes us to

Unconscious Motivation and Emotion.

This page last updated 09/29/2018.