The Role of Cognition in Classical and Operant Conditioning

https://pubmed.ncbi.nlm.nih.gov/15022268/

While cognition is generally accepted a component of operant and classical conditioning, many psychologists view cognitive and classical/operant conditioning as opposing camps. The paper I’m summarizing here shows how the two views could work together.

A Review

SR conditioning, a stimulus and response. EG - sawdust is flung towards your eye when woodworking, and you engage your safety squints automatically.
Instrumental/operant conditioning - where behavior is modified through consequences, through positive or negative reinforcement. EG - cracking a lewd joke at the dinner table will get you points with friends and excused from the table with family. Being excused from the table would be an expectancy.

There’s no question whether or not classical conditioning is effective - but modern theorists view it as a procedure, rather than a direct cause, rendering it theoretically neutral. Classical conditioning is seen as S-O and S-S, where operant/instrumental conditioning shows R-O associations. The question isn’t whether or not these processes work, but the mechanisms behind them.

It’s generally accepted that conditioning involves some creation of an expectancy.

From a cognitive perspective (e.g., Kirsch, 1985), instrumental learning situations produce expectancies that particular behaviors will produce particular outcomes (e.g., that food can be found in a particular location), and classical conditioning produces expectancies that certain stimuli will be followed by other stimuli (e.g., that food will be presented soon after the bell is sounded).

Even though some expectancies resemble S-O, not all S-O associations have expectancies. EG - hearing No Scrubs come on the radio might unexpectedly create nostalgia, but being forced to listen to Nickelback without consent and a pre-talk would be pretty irritating.

Data indicating higher-order cognition

Tolman

The earliest challenges to the 1930s-1940s mechanistic properties of behaviorism came from Tolman. Instead of showing simple stimulus response, they highlighted a wide array of behaviors displayed by rats that appeared to be cognition. The rats engaged in “VTE” (vicarious trial and error) when rewarded - looking back as if to make sure they figured out the right stimulus. They also observed them trying out different hypotheses, like trying all the doors on the right getting out of a maze and changing their behavior if their first strategy didn’t work.

Another nod towards rats displaying cognition in learning was comparing ‘place learners’ vs ‘response learners.’ Place learners would be positioned in a random location on a maze and were expected to find their way back. Response learners were encouraged to make the same repeated patterns (EG - left right left right,) and rewarded for doing so. More of the place learners than response learners were successful, however, even some of the response learners were able to adjust their strategies to find food.

Rats were also shown to, without reward, memorize the layout of a maze. When rats were just dumped into a maze without any reward and allowed to roam around, or even given a ‘rat ride’ in a cart to be shown around, they were able to take in the location and memorize it. When they added food as a reward, rats that learned the layout earlier, even without stimulus, had an easier time finding dinner.

Brewer’s 1974 Review

Brewer, in a review of 200+ studies, found strong support for a “cognitive interpretation” of learning in humans, as compared to conditioning. The paper provides a list of their findings and categories, but here’s a few of my takeaways.

By simply verbally informing people that the conditioned stimulus (CS) would no longer produce the conditioned outcome (CO,) studies were able to stop the expectation of the CO (extinction) without reinforcing trials. In addition, when asked not to produce a CR, participants were able to do so.
People are cognitively aware of contingencies. EG - removing the shock electrodes removed the expectation of shock.

Brewer, in conclusion, wrote:

…all the results of the traditional conditioning literature are due to the operation of higher mental processes, as assumed in cognitive theory, and that there is not and never has been any convincing evidence for unconscious, automatic mechanisms in the conditioning of adult human beings…

Rescorla’s Reviews

Rescorla also concluded in his reviews that conditioning is not a simple, mechanistic process. They found that to get a behavioral response, the relationship between the CS and US was more important than just the CS and US pairing. EG - if the CS happens only occasionally predicting a US, the CS is less valuable as information.

Simple SO Associations

Multiple S’s can be conditioned together without SO, then used later for the SO. This suggests cognition.
Temporal CS-US pairings are enough to establish associations, but continued conditioning creates more chances for learning the contingencies of the CS-US relationship.
Expectancy theory doesn’t have much to offer as far as explaining responses without a goal. (EG - rats will jerk their heads when hearing a tone, but there’s no goal behind this, even if they do expect to get a snack afterwards.)
CS-US conditioning is contextual. EG - subjects go in to trial 1 for un-fun zappy-zappy time with a CS, then later go in to a new context of trial 2 where zaps are not present and the same CS is used. Go figure, when subjects went back to the trial 1 context, they expected to be zapped again. The mechanistic view of conditioning would suggest the context wouldn’t matter, but cognition mediated this lack of extinction.

R-O, S-R-O in Operant Conditioning

Given that rats are already trained with an R-O association to get food, if you put half of them into an experimental group where in the second trial, they’re given unappealing food, in the third trial they’re less apt than the un-abused rats to emit the response required to get more food. A mechanistic SR would suggest that their behavior would be unchanged, and the cognitive view predicted the results correctly.
Rescola theorized S-(R-O) associations existed - whereby a stimulus (situation) suggests certain responses (R) would lead to contextual expected outcomes (O.) S-(R-O) associations are a part of a modern understanding of conditioning theory. Mechanistic views expect responses because they were conditioned, and cognitive views expect goal directed behavior.

Cognitive Mediation in Clinical Research

Systematic desensitization is a treatment or phobic anxiety based on Clark Hull’s theory of conditioning. The strategy is to associate the CS (cues causing anxiety) with a CR incompatible with the anxiety (relaxation.) Also in line with this strategy is repeated exposure to the CS will eventually lead to extinction of the CR (anxiety.)
😅 Kirsch and Henry created a study to specifically rule out a conditioning hypothesis. They emitted shocks during the CS, explaining the shocks would ‘punish’ the anxiety, and it worked. This suggests expectancy, not desensitization was the cause of change.
Agoraphobic subjects were split into two groups - one told they were being measured for response, the other that they were receiving treatment. They were both asked to go take a walk outside until they started to feel uncomfortable, stop, then return. The one told they were receiving treatment improved more.
Placebos (pills, syringes) can be seen as a CS, generally lowing pain reports. They did find that it was possible through conditioning that pain reduction could occur, but they had unexpected results for conditioning theories after explaining they were given placebos.

…participants in the informed-pairing condition were given trials in which reduced pain was paired with the application of a placebo. However, they also were given accurate information about how the reduction in pain was being produced. This verbal information completely reversed the effect of conditioning trials on the placebo response, which is not anticipated by even modern conditioning theories.

Another argument against a simple CR to a CS was found when conditioning to use morphine or a tranquilizer. When switching to placebo, pain response increased. They use this as an argument against conditioning theories. In contrast to the idea that you could be surprised that you were given a dud, the administration could create a voluntary compensatory response. EG - being given placebo alcohol could make you automatically try to compensate for the drink if you were forced to do something consequential.

Data Indicating Automatic Conditioning

Ah - here’s a line that puts this section of the paper into perspective, as well as the whole document, to a degree…

…the data reviewed above provide clear evidence of cognitive mediation… Other data, however, reveal conditioning phenomena that do not appear to be mediated cognitively.

They make the argument that conditioning models make more sense than cognitive models in specific cases. Let’s see what they’ve got…

Evaluative Conditioning

When you pair a conditional stimulus (CS) with an an affectively valenced (liked or disliked) unconditional US, the affect will imprint on the CS. These can be conditioned without awareness, as well as are prone to be resistant to extinction. There’s a specific prevalence of this in smell and taste.

S-R Persistence Following Devaluation

Ratio reinforcement of operant behavior (modifying the percentage of stimulus given for a behavior) can cause devaluation, but switching to interval reinforcement (on a timed schedule) does not reduce the behavior. Intervals may not establish R-O since the relationship is much weaker than a ratio schedule. One theory suggests that ratio schedules drive behavior through R-O, and interval schedules rely on S-R.

Continuing with ratio schedules, that type of reinforcement over time can be resilient to devaluation.

There’s evidence that cognitive and mechanistic processes can support the same behavior. An S-R association can maintain the initial response after long term conditioning, and cognitive behavior can take care of the goal driven component.

Resistance to Change and Excessive Behavior

Nevin was able to observe Behavioral Momentum (resistance to change.) Given that rats are given stimuli on two different interval schedules (1 and 3 minutes for this example,) with short breaks in between, the 3 minute schedule period is less resistant to a change in behavior after throwing in a ‘break’ with a randomized food stimulus thrown in. Changing the value of the stimulus also showed similar changes in behavior.

Nevin has shown that the overall rate of responding is determined by the response–outcome contingency, but resistance to change is determined by the overall rate of outcomes obtained in the stimulus situation (Nevin, Tota, Torquato, & Shull, 1990).

Put simply, the response-outcome affects the behavior rate, but resilience is controlled by the regularity of the stimulus.

Second-Order Conditioning

When a CS is paired to a new CS in absence of the US (snacks,) the second US still produces a response. When the first CS is extinguished, the second CS remains effective. The cognitive account would guess that the CS would also be extinguished, but that’s not the behavior they observe, pointing towards mechanistic SR associations.

Conditioned Taste Aversions and Flavor Preferences

Welp this one explains Taco Bell. While administering electric shocks, giving someone upper-intestinal discomfort, or giving them lower-intestinal discomfort all made people avoid the food food, only the upper-intestinal discomfort shifted the affective component from like to dislike of the taste.

This sentence might explain why energy drinks are loaded up with nutrients…

For example, flavor preferences conditioned with intragastric nutrient infusion as a US are very resistant to extinction in both deprived and nondeprived animals (Drucker et al., 1988), enhance sham-feeding responses (Myers & Sclafani, 2001a), and shift taste reactivity toward responses typically seen with higher concentrations of sweet solutions (Myers & Sclafani, 2001b).

Conditioning with Subliminally Presented CSs

Welp, TIL subliminal conditioning works. I just assumed it was horseshit. Brief visual stimuli (<300msec) immediately before a ‘masking’ image are enough to produce an affective (emotional) response. Several studies have been able to use this as a CS to shift a US, suggesting we can be conditioned outside of cognition.

Conditioned Immunosuppression

Administering a sweet CS snack right before giving cyclophosphamide injections was able to condition the same response as the cyclophosphamide, reducing the number of T-lymphocytes as compared to control. WTF.

Conditioning in Simple Organisms

Uh - you can condition snails to avoid a type of touch when pairing it with an electric shock. These bad boys unfortunately don’t have much horsepower at around a few hundred neurons, so this leads itself to the conditioning model over cognition.

Conclusions

There are two wolves inside of me.

The first conclusions we could reach is that cognition is a side effect of classical and operant conditioning.
The second is that cognition mediates expectations, and expectations produces the response.

Given the body of data presented, both wolves are missing the mark. Kirsch proposes that the more complex the organism, the more it’ll rely on cognition over conditioning.

Another way to look at is S-R relationships can be looked at response sets. S-O and R-O associations are just perceptual sets for interpretation. Some expectancies are cognitively available, and others may be outside of our awareness (implicit expectancies.) Classical and operant conditioning could just be two ways we create response sets, and are observable.