How Do We Learn What Predicts What?

Contingency, surprise, attention, and the explanations we give for learning

Level: Intermediate
Reading load: 10 core readings in 5 units
Prerequisites: The Learning & Behavior foundation is useful preparation, especially its distinction between learning and performance. Comfort comparing experimental conditions and reading a simple graph. Model terms are explained; programming and a formal learning-theory course are not required.
Last reviewed: September 2026

Introduction

A cue can precede an outcome often and still tell us little. To judge whether a sound predicts an event, we may need to know how often the event occurs without it, what other signals are present, and whether the relationship has changed. Learning to expect something therefore raises a question beyond how many times two events appeared together: what information does the learner use?

This syllabus begins with that distinction between repeated pairing and an informative relationship, called a contingency. It then examines a model in which learning depends on the difference between what was expected and what occurred. This difference is called prediction error. The model gives a precise proposal about how experience changes an expectation, which experiments can then test.

The next readings ask which cues receive attention and why learning in one situation carries into another. Replication also matters: before using an experimental pattern to choose among explanations, we must examine how dependable that pattern is. Across these comparisons, keep track of what the learner is assumed to notice and remember.

The closing unit compares accounts in which learning builds associations among events with accounts in which learners represent causal relationships or claims about how events are connected. Rather than treating the names as answers, ask what each account predicts in a case where the others would expect something different.

By the end, you should be able to explain why pairing alone may not make a cue informative, describe how cues can compete, and propose a comparison that distinguishes explanations of learning. Several theoretical papers use equations; the guides help translate them into claims about experience and behavior.

Unit 1 of 5

What makes a cue informative?

Start by comparing what happens with a cue to what happens without it. Pairings describe part of the experience; contingency also depends on the background. The experiment gives us a contrast to explain, while leaving open how the learner represents that contrast.

Reading
1 of 10
Probability of Shock in the Presence and Absence of CS in Fear Conditioning
Rescorla, R. A. (1968). Probability of shock in the presence and absence of CS in fear conditioning. Journal of Comparative and Physiological Psychology, 66(1), 1–5. https://doi.org/10.1037/h0025984
Landmark empirical study · Methodological contribution · Moderate · Full work: 5 pages
Trace the cue-present and cue-absent conditions, the response measure, and the comparisons that make pairings alone an incomplete account.
Why you’re reading it. Establish the empirical problem that a predictive-learning theory must explain. If you met this article in the foundation, now connect the experimental contrasts to an explicit representation of cue and background information.
A question to consider. A signal precedes an outcome on 80% of its appearances, but the outcome also occurs during 80% of comparable periods without the signal. What would a comparison with a lower-background condition reveal, and what would it leave unresolved about the learning mechanism?

Unit 2 of 5

How does surprise change learning?

A formal rule can make one cue’s learning depend on what other cues already predict. Rescorla and Wagner show how this generates cue competition. Maes and colleagues then make a crucial complication visible: a well-known effect does not appear under every procedure expected to produce it. Judge the rule, its representational assumptions, and the experimental test separately.

Reading
2 of 10
A Theory of Pavlovian Conditioning: Variations in the Effectiveness of Reinforcement and Nonreinforcement
Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical conditioning II: Current research and theory (Chapter 3, pp. 64–99). Appleton-Century-Crofts.
Foundational argument · Historical turning point · Challenging · Full work: 36 pages
This selection is Chapter 3 of Classical Conditioning II, pp. 64–99. Translate the equations into changes in prediction and follow how the same rule handles different compound-training arrangements.
Why you’re reading it. Move from the contingency finding to a specified process that makes testable predictions. Understanding the argument does not depend on reproducing every numerical calculation.
A question to consider. After A already predicts an outcome, X is added and AX receives the same outcome. Why does the summed prediction matter for learning about X, and what would changing the outcome let you test?
Reading
3 of 10
The Elusive Nature of the Blocking Effect: 15 Failures to Replicate
Maes, E., Boddez, Y., Alfei, J. M., Krypotos, A.-M., D'Hooge, R., De Houwer, J., & Beckers, T. (2016). The elusive nature of the blocking effect: 15 failures to replicate. Journal of Experimental Psychology: General, 145(9), e49–e71. https://doi.org/10.1037/xge0000200
Replication · Critical reappraisal · Challenging · Full work: 23 pages
Compare control histories, responding at test, and the authors’ explanations for the missing blocking effects. Keep procedural failures and evidence about a model’s boundaries distinct.
Why you’re reading it. Evaluate the empirical reach of the effect used to motivate the model. The optional commentary and reply offer a focused dispute about whether the failures could have been predicted.
A question to consider. Choose one reported procedural feature that could affect blocking. What result in a new comparison would support it as a boundary condition rather than a convenient explanation after the fact?

Unit 3 of 5

Which cues receive attention?

Surprise can affect the amount learned, but experience may also change which cues are processed. A synthesis distinguishes reliable prediction, uncertainty, and value; an eye-tracking study makes the measurement problem concrete. Ask both what attention follows and whether it causes the later learning difference.

Reading
4 of 10
Attention and Associative Learning in Humans: An Integrative Review
Le Pelley, M. E., Mitchell, C. J., Beesley, T., George, D. N., & Wills, A. J. (2016). Attention and associative learning in humans: An integrative review. Psychological Bulletin, 142(10), 1111–1140. https://doi.org/10.1037/bul0000064
Major synthesis · Critical reappraisal · Challenging · Full work: 30 pages
Separate the predictiveness, uncertainty, and value principles, and compare evidence from different attention measures. Follow the distinction between goal-directed selection and a learned bias that persists despite current goals.
Why you’re reading it. Broaden the fixed learning-rate picture into an account of how experience changes processing of cues themselves.
A question to consider. When a reliable predictor suddenly becomes unreliable, what would predictiveness-based and uncertainty-based accounts lead you to look for, and at what point in training?
Reading
5 of 10
Overt Attention and Predictiveness in Human Contingency Learning
Le Pelley, M. E., Beesley, T., & Griffiths, O. (2011). Overt attention and predictiveness in human contingency learning. Journal of Experimental Psychology: Animal Behavior Processes, 37(2), 220–229. https://doi.org/10.1037/a0021384
Methodological contribution · Moderate · Full work: 10 pages
Use the two-stage design to track how previous predictive history relates to gaze and new learning when current predictiveness is equal. Notice the authors’ explicit causal limitation.
Why you’re reading it. Apply the review’s distinctions to a concrete study and separate an attention measure from an established attention-mediated explanation.
A question to consider. If the previously predictive cues attract more gaze and are learned about faster, which part of the review’s proposed attention–learning relation is demonstrated most directly, and which needs another experiment?

Unit 4 of 5

What makes a new situation similar?

Learning rules operate on representations, not on the experimenter’s labels alone. Pearce asks whether a compound behaves like separate elements or a configuration. Shepard asks how to describe psychological similarity across stimuli. Together they show why a theory needs to specify what counts as the same, different, or similar at test.

Reading
6 of 10
Evaluation and Development of a Connectionist Theory of Configural Learning
Pearce, J. M. (2002). Evaluation and development of a connectionist theory of configural learning. Animal Learning & Behavior, 30(2), 73–95. https://doi.org/10.3758/BF03192911
Foundational argument · Major synthesis · Challenging · Full work: 23 pages
Compare elemental and configural predictions for compounds, including the problems and extensions Pearce discusses. Relate generalization assumptions to the earlier blocking debate.
Why you’re reading it. Identify how changing a representation can change predictions even within the broad family of associative explanations.
A question to consider. In a blocking design, how could incomplete generalization from A to AX, and from AX to X, change the expected test result? What measurement could constrain these assumptions before observing that result?
Reading
7 of 10
Toward a Universal Law of Generalization for Psychological Science
Shepard, R. N. (1987). Toward a universal law of generalization for psychological science. Science, 237(4820), 1317–1323. https://doi.org/10.1126/science.3629243
Foundational argument · Methodological contribution · Challenging · Full work: 7 pages
Follow the argument from observed gradients to psychological distance and the proposed exponential relationship. Keep a mathematical derivation distinct from evidence for its assumptions.
Why you’re reading it. Extend the question of representation beyond compounds to the measurement and prediction of similarity-based transfer.
A question to consider. How could you use an independent similarity task to make a generalization prediction more informative than drawing a curve through the observed test responses?

Unit 5 of 5

Do learners form associations, causal models, or propositions?

The final unit asks what a learner’s interpretation contributes. Causal-model experiments challenge a symmetry assumed by particular associative accounts. A propositional argument extends the challenge, and a short commentary questions whether it reaches more complex networks. Compare specific explanations rather than deciding the debate by labels.

Reading
8 of 10
Predictive and Diagnostic Learning Within Causal Models: Asymmetries in Cue Competition
Waldmann, M. R., & Holyoak, K. J. (1992). Predictive and diagnostic learning within causal models: Asymmetries in cue competition. Journal of Experimental Psychology: General, 121(2), 222–236. https://doi.org/10.1037/0096-3445.121.2.222
Landmark empirical study · Foundational argument · Moderate · Full work: 15 pages
Compare predictive and diagnostic instructions while keeping presentation order separate from causal direction. Identify the cue-competition pattern that each account expects.
Why you’re reading it. Test whether a model built around cue–outcome statistics needs additional assumptions about the learner’s causal interpretation.
A question to consider. Keep a training sequence constant but describe its cues as causes in one condition and effects in another. What judgment difference would challenge an account that represents only that sequence?
Reading
9 of 10
The Propositional Nature of Human Associative Learning
Mitchell, C. J., De Houwer, J., & Lovibond, P. F. (2009). The propositional nature of human associative learning. Behavioral and Brain Sciences, 32(2), 183–198. https://doi.org/10.1017/S0140525X09000855
Foundational argument · Critical reappraisal · Challenging · Full work: 16 pages
The selected work is the target article, pp. 183–198; the accompanying commentaries are separate publications. Identify the proposed mechanism and examine how the authors use awareness, instructions, and cognitive resources to distinguish it from alternatives.
Why you’re reading it. Consider an argument that moves beyond modifying associative models to questioning the need for a separate link-forming process.
A question to consider. Which finding from the earlier units could fit both an associative model and a propositional explanation? Specify an additional manipulation that the two accounts would need to predict differently.
Reading
10 of 10
The Associative Nature of Human Associative Learning
Shanks, D. R. (2009). The associative nature of human associative learning. Behavioral and Brain Sciences, 32(2), 225–226. https://doi.org/10.1017/S0140525X09001149
Critical reappraisal · Moderate · Full work: 2 pages
Track which claims Shanks accepts and where he thinks the propositional argument overreaches. This two-page commentary is a separate core work, not a requirement to read the entire commentary package.
Why you’re reading it. Finish with a substantive objection and assess whether evidence against simple links necessarily counts against a more complex associative system.
A question to consider. Choose one prediction from your comparison of accounts. Does Shanks’s objection change it, or mainly change what mechanism could produce the result? State what additional specification would make the disagreement testable.

Scope and omissions

The central subject is learning relationships among events, with animal conditioning and human contingency-learning tasks serving different evidential purposes. The sequence includes theoretical arguments, formal models, experiments, an integrative review, replication work, and a published objection. It is a selective route through explanatory problems, not a chronological history or a catalogue of every model.

Learning through the consequences of actions, practice and skill, and learning from other people require their own paths. Extinction and persistence are touched on where they constrain an account, rather than developed as treatment applications. Detailed neural mechanisms, temporal-difference algorithms, Bayesian model fitting, language learning, and category-learning architectures lie beyond this core. Findings from restricted laboratory tasks do not establish the same process in every species or everyday setting.

Synthesis

Predictive learning concerns the information events carry about one another, and the processes by which experience changes its use. Pairing, contingency, prediction error, attention, and representation make different contributions to the account. Associative, causal-model, and propositional explanations must be compared through their commitments and predictions. A pattern of responses becomes more informative when we can show which alternatives it rules out.

Imagine two lights on a machine that sometimes precede a breakdown. Begin by specifying what happens with and without each light. Then describe what prior training, the two lights appearing together, and a changed outcome would let you investigate. Separate the learning rule from assumptions about attention, representation, and the mapping from learning to an observed response.

Now change only the instructions: the lights are described either as possible causes of a fault or as separate consequences of it. Choose two accounts from the syllabus and state a prediction for which they might disagree. Explain what would make a null result informative, what independent measure would constrain a flexible explanation, and why success in this task would still leave some everyday learning unexplained.

Return to How Does Experience Change Behavior? to reconnect predictive learning with consequences, observation, practice, and persistence. For further work on evaluating causal claims and scientific uncertainty, How to Read Psychological Science provides a broader methodological route.

Continue with How Do Consequences Shape Action? to examine what changes when an outcome depends on an action. The path connects reinforcement, schedules and choice, punishment, and the evidence distinguishing goal-directed actions from habits.