COGNITION & BEHAVIOR
Adaptivebehavior.
We design experiments to test how actions adjust when their consequences change, separating learning from changes in motivation or movement.
Begin with the question
Context · flexibility · transitions
- Research lens
- Learning when the world changes
- Working material
- Published evidence & dataset scoping
- Current stage
- Experimental design
The question
02 / Arzola ResearchWhen an outcome changes, what does the action depend on?
01Why this question
A changed choice has more than one explanation.
An unexpected outcome creates an ambiguity: has the environment changed, or was this one outcome noise? The response creates another: did the subject update an expectation, change how much an outcome matters, or simply become less able or willing to act? This workstream develops comparisons that separate those possibilities. The task and experimental system remain under selection.
Behrens and colleagues studied people making choices in reward environments with changing volatility. Participants adjusted the influence of recent outcomes as the environment changed. This provides a concrete computational question: when should experience be retained, and when should new evidence outweigh it? A learning-rate estimate is one model of that adjustment, with assumptions that must be tested.[3]
In rat experiments by Yin and colleagues, disrupting the posterior dorsomedial striatum reduced sensitivity to outcome devaluation and to weakened action–outcome contingency. Their reversible-inactivation experiment also checked discrimination of actions and outcome identities. These comparisons help separate a deficit in flexible control from a general inability to discriminate or respond.[1]
Related dorsolateral striatum experiments verified that devaluation reduced sucrose consumption before testing lever pressing without sucrose delivery. That manipulation check matters: persistent responding is interpretable only if the outcome’s value actually changed. The neural findings belong to those rat experiments; applying their logic to another system requires validation of the task and measures.[2]
02What to examine
Find the comparison that separates the accounts.
The immediate design choice is between questions about a changed outcome value, a changed action–outcome relationship, and changing environmental uncertainty. They require different manipulations. One should anchor the first study, with the others treated as alternative explanations or later extensions.
- 01
Did the intended variable change?
Outcome devaluation changes how desirable an outcome is; contingency degradation changes how strongly the outcome depends on an action. Choose one manipulation and verify its effect independently of the response being used to infer adjustment.[1][2]
Proposed contrastA verified change in value or contingency, against an unchanged reference.
- 02
Is the behavioral change selective?
Compare the action associated with the changed outcome with an unaffected action or control condition. A selective reduction and a uniform reduction across actions carry different information. Track response opportunities, latency, omissions, movement, and engagement so that learning can be distinguished from a change in performance. Preserve the trial sequence and each participant or animal’s identity.
Proposed contrastSelective adjustment to the affected outcome, separated from overall suppression of responding.
- 03
What did the subject update?
Compare candidate accounts of learning, choice consistency, and a tendency to repeat the previous action. A high fitted learning rate can absorb behavior produced by a misspecified choice rule. First simulate each model to check whether the planned task can recover its own parameters, then test predictions on held-out sessions or individuals against a simpler baseline.
Proposed contrastDistinct trial-by-trial predictions, recoverable model parameters, and performance on independent observations.
03Reading a result
An observation.
More than one explanation.
The possible outcomes below guide study design. Each interpretation depends on a successful manipulation and a task sensitive enough to reveal the predicted difference.
If responding decreases after a change
Alternative explanation
Reduced responding could be selective to the changed outcome, or it could reflect lower movement, appetite, engagement, or opportunity to respond.
How to distinguish it
Estimate the affected-versus-reference contrast alongside overall performance. A uniform decrease across responses does not isolate outcome-specific adjustment.
If behavior persists despite a changed outcome
Alternative explanation
The manipulation may have failed, the subject may not have learned the action–outcome relationship, or the test may lack sensitivity. Persistence alone does not establish a habit mechanism.
How to distinguish it
Verify changed outcome value or contingency and intact task performance. Report the size and precision of the sensitivity estimate, and which meaningful effects remain compatible with the data.
If an analysis identifies a new behavioral state
Alternative explanation
The partition could reflect model assumptions, session boundaries, or noise. A more flexible model can fit the same observations better without explaining more.
How to distinguish it
Separate training and test data by session or animal when the intended prediction requires it, rather than randomly mixing adjacent trials. Compare simpler models and test robustness across fits.
04Where the work stands
Experimental design
Designing the test of adaptation.
Study design is in development, including task selection, manipulation checks, and comparisons between competing models. New laboratory experiments are planned work, contingent on the collaborators, resources, and approvals required to carry them out.
- Working material
- Published experiments and dataset scoping. A suitable task and experimental system have not yet been selected.
- Current effort
- Defining the selective behavioral contrast, its manipulation check, and the performance controls needed for an interpretable result.
- Next useful output
- A study concept with a task timeline, competing behavioral predictions, a primary contrast, manipulation and performance checks, and a simulation showing what the planned analysis can distinguish before data collection or confirmatory reuse.
05The next useful step
Make the competing predictions diverge.
Advance the design only when the task can distinguish a selective adjustment from a performance change. Greater model complexity cannot repair a missing comparison.
- 01
Choose one manipulation.
Specify value or contingency, the reference condition, the observation window, and a separate check that the intended change occurred.
- 02
Audit the task and records.
Require individual identifiers, trial timing, outcomes, and relevant controls. If a dataset lacks the decisive comparison, use it for exploration rather than a confirmatory test.
- 03
Test the analysis on known cases.
Generate illustrative sequences from competing models and examine whether the proposed analysis distinguishes them. If learning, lapse, and repetition parameters trade off too strongly, revise the task or simplify the claim before testing the biological question.
06Background reading
The field around the question.
Primary studies supporting the scientific context. The proposed comparisons above are Arzola Research’s study logic; these external papers are not results from our projects.
- [1]
Yin HH, Ostlund SB, Knowlton BJ & Balleine BW · 2005
The role of the dorsomedial striatum in instrumental conditioningEuropean Journal of Neuroscience · 22:513–523
Rat experiments combining devaluation, contingency degradation, and neural intervention with discrimination controls.
- [2]
Yin HH, Knowlton BJ & Balleine BW · 2004
Lesions of dorsolateral striatum preserve outcome expectancy but disrupt habit formation in instrumental learningEuropean Journal of Neuroscience · 19:181–189
Devaluation verified by consumption, followed by an extinction test of instrumental responding in lesioned and control rats.
- [3]
Behrens TEJ, Woolrich MW, Walton ME & Rushworth MFS · 2007
Learning the value of information in an uncertain worldNature Neuroscience · 10:1214–1221
Human learning under changing reward volatility, providing a task-based example of adapting the weight of recent experience.
An open research question
Build the next behavioral experiment.
Collaborators in learning, behavioral neuroscience, and quantitative behavior can help select a validated task or an appropriate dataset. The useful contribution is a comparison that separates explanations, with access to the measurements needed to evaluate it.
- Learning & decision-making
- Behavioral analysis
- Experimental design