1 Introduction Learning and Memory are usually considered separately. Definitions Learning = “…a relatively permanent change in behavior due to experience.” (Note: By definition, maturation and injury would not be learning.) Memory = “…the representation of experience that is stored in the mind.” The definition of learning is quite “behavioristic” and that of memory is quite “cognitive.” That makes sense historically: The two subject matters were studied under different paradigms. Learning was the focus for Behaviorism. Memory is an important area within Cognition. Paradigmatic assumptions and implications: Behaviorist approach Organism is “empty” (merely a switchboard) There is no “mind” Environmentalism No “free will” Strict adherence to British Empiricism principles Experimental approach The central problem of psychology is learning Learning follows laws that apply across all species 2 Cognitive approach Organism is not empty Mind is the “author” of the majority of behavior Anti-environmentalism There is free will More moderate empiricism approach Also takes experimental approach Central problem of psychology is explaining the “higher mental processes” Every species has different cognitive system Forms of learning Habituation = a decrease in responding (i.e., orienting or alerting) that occurs when a stimulus is presented repeatedly and the stimulus is not related to any biologically significant event Sensitization = an increase in responding to a mild stimulus after an intense stimulus These two processes are related reciprocally and probably help to maintain a biological imperative, homeostasis. Classical conditioning = A previously neutral stimulus comes to elicit a reflexive or involuntary response. Instrumental (and also Operant) conditioning = The rate of a freely emitted response is changed based on the consequences of that response. 3 Forms of memory Short-term memory = Sometimes called “working memory” and is roughly equivalent to “consciousness” – limited capacity – maintained by rehearsal Long-term memory = More or less permanent store – large capacity – no maintenance seems to be required Important memory terms Encoding = Forming an internal code (representation) that reflects one’s experience Storage = Maintaining the internal code within the memory system over time Representation = deals with what a stored memory “looks like” (i.e., its nature) Retrieval = locating and activating a stored memory so that it can be used. (Note: Memory may be quite “constructive” and, consequently, retrieval may be a very creative process.) Relationship Between Learning and Memory An interactive relationship: Learning is guided by past experience (stored as memory); and, of course, learning becomes stored as memory. 4 Learning and memory are potentially biologically adaptive because they cope with a changing environment. (If the environment didn’t change, hardwired “instincts” would be more efficient.) But, to be adaptive, learned behaviors must be retrieved at the appropriate time. But, even though learning and memory are related, we usually keep them conceptually separate. One reason is that unique variables can apply to each: For example, learning efficiency is influenced by the practice schedule during acquisition; but once acquisition has occurred, memory performance is influenced by length of the retention interval. Most important perhaps: Historically, learning and memory have been studied separately, usually under separate paradigms, by different psychologists. Separate literatures. Biological Basis of Learning and Memory In higher organisms, the brain seems to be involved. But the picture is less than clear. Hippocampus seems to be involved in memory registration. The hippocampus is part of the limbic system (a subcortical part of the forebrain). Note that the limbic system is also involved with emotion and motivation. “H.M.” is a famous case (reported by Milner in 1950s): 5 H.M.’s hippocampus was removed bilaterally (both sides) to relieve epilepsy. (This would not be done today.) Result: Some retrograde amnesia (going back in time from present) with worst effects for more recent past; and almost complete anterograde amnesia (going forward in time from the present) with nothing new being remembered after a distraction. Principle of Mass Action (Lashley, 1950s): Rat’s performance on a complex task (e.g., a maze problem) depended more on the amount of cortex remaining in the brain, not the location of the remaining cortex. As more cortex is removed, errors go up but the basic task remains intact. Other possibilities for representation: changes in firing rates of neurons; changes in brain chemistry; altered pathways (e.g., changes in patterns of excitation and inhibition across large numbers of neurons) We know that some areas of the brain become “active” during certain cognitive tasks, but we are not sure what that means. 6 Learning might even have a basis outside of what would normally be thought of as “brain.” Consider the classical conditioning of the planarian flatworm (Dugesia species). The planarian is capable of “regeneration” after sectioning. A cartoon from McConnell, Jacobson, and Kimble (1950s) illustrates the classical conditioning procedure: Note: The posterior portion of the sectioned planarian retained just as much of the response as the anterior portion. Finally note structure of planarian nervous system below (ganglia are only in the anterior part of the body): (Ant.) (Post.) 7 Learning can be biologically constrained—this is an important exception to the Behaviorist notion that learning principles apply across species uniformly. Wilcoxon study (1970s): Rats and quail drank water that was both blue and sour. Later a drug injected to make them sick. Subsequently, rats avoided sour (but not blue) water; but quail avoided blue (but not sour) water. Note: In rats, olfactory and gustatory senses dominate; in quail, visual sense dominates. Species-Specific Defense Reactions (Bolles, 1970s): Rats easily learn to escape from a shock grid by jumping over a barrier to a non-shock area (called a “shuttle” response), but rats have difficulty learning to press a bar to end shock. The Role of Learning in the Higher Mental Processes This is a tough subject. Wolfgang Köhler (early 1900s), a Gestalt psychologist, studied problem solving in chimps. Do chimps learn to do things incrementally (i.e., trial and error via conditioning -- Behaviorism) or all at once (i.e., suddenly via insight -- Cognitivism)? In one study, he placed a chimp in a cage along with a stick; outside the cage was a bunch of bananas out of direct reach. A given chimp would grasp the stick, play with it for a while, then suddenly rush to the bars using the stick to claim his prize. But Pavlov objected: Maybe prior experience with sticks biased the animal toward using the stick to get the bananas (i.e., the use of the stick was already high – so a random encounter with the bananas using the stick is likely). Who is right? Hard to tell. 8 Classical Conditioning Ivan Pavlov (early 1900s) did the basic experiments. He was a physiologist interested in digestion. Basic idea: A previously neutral stimulus comes to signal that a reflexive stimulus-response sequence is about to occur. In effect, the organism learns to anticipate the reflexive sequence. 9 Definitions of terms: CS = Conditioned Stimulus = The previously neutral stimulus = The clicking sound (or bell, etc.) US = UCS = Unconditioned Stimulus = The stimulus that naturally elicits the reflexive response = The meat powder UR = UCR = The reflexive response that is naturally elicited = Salivation CR = The conditioned response that comes to be elicited by the Conditioned Stimulus = Salivation (similar to UR) Classical Conditioning is Ubiquitous Whenever we have a reflexive or emotional response in the presence of a stimulus that does not naturally elicit it, we have evidence for classical conditioning. Classical conditioning and drug addiction To understand this, we need an important theory: Opponent-Process Theory of Motivation (Solomon & Corbit, 1974): A pleasurable stimulus (e.g., drug) produces an “aprocess” which is fast-acting and decays quickly when the stimulus is removed. The body reacts with a “b-process” (to counteract the aprocess). The b-process is sluggish in its rise and decay. 10 The purpose of this opponent process is to maintain homeostasis. With repeated presentations, the a-process remains the same, but the b-process grows (starts sooner, becomes more intense, and lasts longer). So with repeated presentations of the drug, we see a decrease in the primary emotional response and an increase in the after-reaction. This is the foundation of the “drug-tolerance effect.” Figure from Solomon and Corbit (1974): So the body’s need for homeostasis produces the drugtolerance effect. Question: Suppose you run out of regular coffee (which you usually drink in the morning) and substitute de-caf, what will happen? 11 Siegel et al. (1980’s) – an “animal model” for heroin addiction Two groups of rats were injected with heroin every other day for 30 days – on non-heroin days, they received dextrose injections. Both groups received injections in two distinct rooms: the rat’s home room or an alternate room. For one group of rats, heroin only in home room and dextrose in alternate room; for the other group of rats, dextrose only in home room and heroin in alternate room. Amount of heroin was increased across days to adjust for tolerance effects. A third (control) group of rats received no heroin, but received dextrose in both home and alternate rooms. After 30 days, all rats were given a double dose of heroin: The experimental rats either received the overdose in their normal heroin room or their normal dextrose room; the control rats received their first (and only) experience with heroin as a double dose. What should happen with regard to overdose fatalities? Consider the three groups: controls, OD in “heroin” room, and OD in “dextrose” room 12 Results: Note: Different room = dextrose room; Same room = heroin room. The customary heroin room served as a CS for a compensatory CR (an opponent process, or b-process) that helped to counteract the effects of the overdose. Summary: controls (no opponent process); different room (opponent process, but not well elicited); same room (opponent process well elicited). Contingency – the dependency between CS and US A contingency means there is a correlation between CS and US occurrence 13 A positive contingency means that the CS reliably predicts the delivery of the US; a negative contingency means that the CS reliably predicts the absence of the US. Experiments using positive contingency are called classical excitatory conditioning experiments; experiments using negative contingency are called classical inhibitory conditioning experiments. If there is no contingency between CS and US, conditioning cannot occur (i.e., there is nothing to learn). Common Classical Conditioning Paradigms Eye-Blink Reflex (Human) US = air puff UR = eye blink reflex CS = tone CR = eye blink Nictitating Membrane Reflex (Rabbit) Same as above except inner eye membrane blinks Conditioned Emotional Response (CER) A really important paradigm for developing animal models of pathologies such as phobias US = shock UR = fear (operationalized as startle or freezing) CS = tone CR = fear (operationalized as startle or freezing) 14 To measure freezing, we present the CS while the subject is freely emitting a high-rate operant behavior (such as bar pressing for food) and measure the amount of suppression of the operant behavior. (more later) Variables that affect excitatory conditioning Temporal relationship between CS and US: Best: Forward Delayed Some: Forward Trace None: Simultaneous, Backward CS-US Interval Stimulus onset asynchrony (SOA) = the time between onsets of two stimulus events (a cognitive term really) Here we are interested in the SOA between CS and US 15 Smith et al. (1969) studied conditioning as a function of SOA (using 50-msec tone CS and 50-msec aversive US) Note that 50 msec is only 1/20th of a second, so these are very brief stimuli Results: The backward (-50 msec SOA), simultaneous (0 msec SOA), and slightly forward (+50 msec SOA) CS-US arrangements do not result in conditioning. The best arrangement here seems to be an SOA of +200 msec (which is really a trace conditioning procedure due to short stimulus durations involved). Spooner and Kellogg (1947) used somewhat longer stimulus durations (200 msec) in human study. CS = tone, US = finger shock, UR = finger withdrawl, CR = finger withdrawl on CS-alone trials. 16 Results: Here we see, once again, that the backward and simultaneous procedures don’t work. But now the maximum conditioning occurs at an SOA of 500 msec (1/2 sec). This is the more usual finding. There is a sharp roll off after 500 msec. Note: As CS-US SOAs become very long, we would start to experience a negative contingency between the CS and US, and have a classical inhibitory conditioning procedure (i.e., the CS would signal a US-free period). 17 Learned Taste Aversion Studies (e.g., Garcia and his colleagues in the 1960s and 1970s) At first glance, it seems to be the exception to the fairly short positive SOA rule. In LTA studies, the animal eats or drinks a novel substance with distinctive taste or odor. If the animal is made sick by a drug hours later, they will avoid the novel ingested substance in the future. This is one-trial learning of a very special kind. LTA probably does not represent normal conditioning – therefore, it should not be listed as an exception to the usual short positive SOA rule. 18 The Role of Contingency (i.e., the correlation of CS and US) Background: Traditional conditioning theory emphasized psychology’s traditional philosophical viewpoint, associationism: By this view, the more pairings of CS-US, the better the conditioning. Today, we call this a contiguity view. In other words, what is learned is an association based on the appearance of CS-US together in time (i.e., contiguously). Robert Rescorla challenged that view as a graduate student at Pennsylvania (late 1960s). Because he emphasized the CS-US correlation, his was a contingency view. In other words, learning is the subject’s realizing that the CS predicts the US. Note that in the traditional excitatory conditioning experiment, contiguity and contingency are confounded. We need to take special steps to deconfound them. After Rescorla (1968), consider two conditions: Cond A CS US Time 1 + + Time 2 + Time 3 + + Time 4 + Cond B CS US Time 1 + + Time 2 - Time 3 + + Time 4 - The number of CS-US pairings (contiguity) is equal for the two conditions; but in Condition B, the US always appears when the CS is present (and never alone). 19 Contingency is quantified by the following conditionalprobability equation: Contingency = P(US | CS) – P(US | no CS) For Condition A (no contingency), ContingencyA = 2/2 – 2/2 = 0.0 For Condition B (perfect contingency), ContingencyB = 2/2 – 0/2 = 1.0 Rescorla (1968) taught rats to press a bar for food pellets. Then he carried out a conditioning procedure in which the CS was a tone and the US was shock. Several conditioning groups were formed based on the probability of shock occurring without the tone (i.e., P(US | no CS)) while holding the number of pairings constant P(US | CS). Rescorla (1968) evaluated the quality of conditioning by the amount of suppression of the operant (bar press) response when the CS was presented. This is called a Conditioned Emotional Response (CER) procedure. Suppression is a measure of the amount of fear produced by the CS, which in turn is a measure of the quality of the prior conditioning. 20 The CER procedure uses the suppression ratio as its dependent variable: The more the animal freezes when the CS is presented during bar pressing, the lower the suppression ratio; conversely, the less the animal freezes, the higher the suppression ratio. Note: Range: 0.0 ≤ SR ≤ 0.5 Results: If P(US | no CS) = 0, the suppression is at maximum (suppression ratio = 0) indicating maximum fear. Conclusion: Conditioning also depends on contingency, not just contiguity. Both are important. 21 Number of CS-US Pairings All other factors held constant, the more CS-US pairings, the greater the likelihood that a CS will come to elicit a CR (until an asymptote is reached). Figure 2.11: This finding accords with our intuitions concerning the “learning curve” and seems to go nicely with a pure contiguity view of conditioning (but recall that the contiguity and contingency views overlap a lot). CS and US Intensity In general, as either CS or US intensity increases, strength of conditioning increases. This effect seems to be due to salience (intensity relative to background) rather than absolute stimulus magnitude. Prior Experience with CS or US In general, pre-exposure to a CS alone hinders subsequent conditioning using that CS (called latent 22 inhibition). Same thing with pre-exposure to US (called pre-exposure effect). Relevance of CS and US In defiance of a “pure associationism” doctrine, not just any CS goes with a given US. Gorgeous experiment by Garcia and Koelling (1966): Rats trained to drink a solution from a tube. The solution had a distinctive taste and drinking was accompanied by a light and a clicking sound. Note: Both taste and A/V stimuli were combined into single CS complex. After drinking, ½ rats were given shock US and the other ½ rats given lithium chloride US (which produces nausea). Then all rats (now thirsty) were given access to drinking tubes again. The tubes either had the original taste CS or the original A/V CS (but not both). For each tube type, ½ rats had been given the shock US previously and ½ rats have been given the lithium chloride US previously. Results (Licks per minute): Test CS Taste A/V Previous US Li-Cl Shock 140 300 280 70 23 Conclusion: Biological relevance dictates the ease with which specific CS-US combinations result in conditioning. CS as a Compound Stimulus Overshadowing. If some stimulus of a CS compound is more salient than the other stimuli, the more salient stimulus will become the effective CS and the less salient stimuli will receive little or no conditioning. Blocking. If one stimulus in a CS compound has undergone conditioning alone before it became part of the compound, that stimulus will prevent conditioning of the other stimuli in the compound. Correlation. If only one stimulus in a CS compound is correlated with the US, only it will be conditioned; however, if no stimulus in a CS compound is correlated with the US, but one is a constant component of the CS, then the constant component will be conditioned. Example: Wagner et al. (1968) 24 Results: Light (L) in Uncorrelated condition serves as a “background” stimulus and, because nothing else is correlated with US, becomes the focus of the subject’s fear response. Conditioning Without an Explicit US Higher Order Conditioning Once a CS is conditioned to a US-UR sequence, another CS can be conditioned to the previouslyconditioned CS. So we have a chaining effect: If And Then US (shock) UR (fear), CS (tone) is paired with (US-UR) CS (tone) CR (fear) Now if Then CS2 (light) is paired with CS (tone) CS2 (light) CR (fear) Essentially CS2 predicts fear, so it comes to elicit it. 25 Classical Inhibitory Conditioning A really murky area: Little has been done, and even less is coherent. However, a couple of things emerge: CS Discriminations. Subjects readily learn that a CS+ is always followed by a US and that a CS– is never followed by a US. If the US is aversive, CS+ produces fear and CS– probably produces a hedonically opposite effect (relief?). Long CS-US Intervals. The subject can learn that the US will follow the CS after a long delay. Here the CS (which is supposed to be excitatory) takes on the properties of a CS–. 26 Extinction A really important concept in conditioning. The CS ceases to be a reliable predictor of the US-UR sequence; as a result, the CR diminishes in magnitude as the CS is presented alone repeatedly. Although it is analogous to forgetting (in memory), extinction is different: Forgetting occurs because a memory is not used; extinction occurs because a contingency no longer exists. Three things suggest that the CS-CR association remains intact during extinction, even though the CR diminishes: 1) The CS-CR can be reconditioned in fewer trials than the original conditioning required; 2) a change in context can allow the CS-CR to return at full strength; 3) “spontaneous recovery” can occur in which the CR re-emerges in the old context (although at less than full strength). Note: One of the best measures of CS-CR conditioning strength is the number of trials to extinction required for the CS presented alone. Variables That Affect Rate of Extinction 1. Number of CS-US acquisition trials. 2. Partial reinforcement (no US on some conditioning trials) results in more resistance to extinction than continuous reinforcement (US on every trial). But continuous reinforcement results in faster conditioning. 27 Clinical Applications of Extinction (treating phobias) The main problem is to allow extinction to occur instead of avoidance (which prevents extinction). Techniques: Flooding. Subject is immersed in phobic CS’s. Systematic Desensitization (Wolpe). Relaxation techniques are learned first, then the client confronts CS’s in a “fear hierarchy” (beginning with the least fear-inducing CS). Relaxation is incompatible with fear. Because fear (due to original US-UR) does not occur in presence of a given CS, we have extinction. 28 The pursuit of truth and beauty is a sphere of activity in which we are permitted to remain children all our lives. --Albert Einstein Classical Conditioning: Theoretical Issues What theory serves as an adequate explanation of classical conditioning (i.e., what conditions are necessary and sufficient to explain it)? There are three basic theoretical positions: contiguity, contingency, and hybrid approaches. Contiguity = temporal relationship between CS and US-UR sequence is what matters Contingency = the predictive relationship (i.e., correlation) between CS and US-UR sequence is what matters Rescorla-Wagner model = a hybrid approach Contiguity theory (Pavlov, 1900’s) Basic explanation: The temporal overlap of neural activity of CS and US-UR provides the basis for association; nature of CS and US don’t matter (just so they are perceived). Evidence for: Basic temporal findings for CS and US (e.g., delayed conditioning > trace conditioning); optimal SOA is best, etc. 29 Evidence against: Garcia and Koelling (1966) showed that A/V CS works better with shock US than Li-Cl US; and Rescorla (1968) showed that US-alone trials disrupt conditioning even if number of CS-US pairings is equal. Another problem: Contiguity can’t handle inhibitory conditioning: That is, under the theory, a CS– can’t be neurologically paired with a non-event (nothingness). Contingency theory (Rescorla, 1960’s) Basic explanation: Conditioning occurs when P(US | CS) ≠ P(US | no CS) Note: If P(US | CS) > P(US | no CS), then excitatory; and if P(US | CS) < P(US | no CS), then inhibitory. Evidence for: Basically same as contiguity theory. Plus it accounts for conditioned inhibition (above); and also accounts for the negative effects of US-alone trials on conditioning. Evidence against: Some studies have reported conditioning without contingency (but not conclusive); and possible problems interpreting US-alone trials in Rescorla (1968). 30 Rescorla-Wagner model (1972) An elegant, powerful model Assumptions: 1) Conditioning occurs to the extent that the presence or absence of the US is “surprising”; 2) a CS is excitatory if it precedes US-UR occurrence, and inhibitory if it precedes a US-UR non-occurrence; and 3) if the CS is no longer followed by the US-UR occurrence or non-occurrence, then extinction occurs. Background of model: The essential idea of “surprise” is actually from Kamin’s (1969) explanation of the blocking phenomenon. That is, if a compound CS consists of a tone + light, after conditioning with the compound CS, normally, either the tone or light alone will elicit a CR; however, if either the tone or the light had been paired with the US previously, the other stimulus element of the compound CS will not come to elicit a CR when presented alone after compound conditioning. Kamin (1969) argued that when organisms first experience a US, it comes as a complete surprise. A surprise is not good, so the animal tries to learn to predict the US by “looking back” in memory for a salient event that occurred just prior to US. The tone becomes the new predictor of the US. If the tone later becomes part of a compound CS, only the tone will be noticed and conditioning to other CS elements will not occur (because the US is already predicted by the tone and so US is not a surprise). 31 To support this idea, Kamin then did the following: Pre-training: Tone shock Conditioning: Tone + Light shock Blocking test: Light no CR New experience: Tone + Light shock … shock New test: Light CR present Reason: second shock was surprise Formal (Rescorla-Wagner) model: ∆Va ═ αβ (λ ─ Vax) where: ∆Va is change in associative strength between CS and US on a given trial (subscript a refers to trial #) α is salience of CS β is intensity of the US λ amount of conditioning that the US can potentially undergo (i.e., its associative strength or potential predictability) Vax the amount of conditioning that has already occurred between the US and all other stimuli (i.e., degree to which US is already predicted) So the difference (λ ─ Vax) is the US surprise factor for a given trial. 32 How the model works: Model: ∆Va ═ αβ (λ ─ Vax) Note: subscript a refers to trial # Simplifying assumptions: Salience of CS & intensity of US: let αβ = 0.3 Maximum US associative strength: let λ = 1.0 Prior US conditioning (all stimuli): let V1x = 0.0 Trials: ∆V1 = Δ CS-US strength on T1 = 0.3 (1.0 – 0.00) = 0.30 ∆V2 = Δ CS-US strength on T2 = 0.3 (1.0 – 0.30) = 0.21 ∆V3 = Δ CS-US strength on T3 = 0.3 (1.0 – 0.51) = 0.15