Wait, What? Working Harder Is Not Always More Intelligent

Alicia is solving an examination paper.

Question 1 is routine. She knows the method, the arithmetic is stable and the answer is easy to verify.

Question 7 is unfamiliar. The wording allows two plausible interpretations, the marks are substantial and one wrong assumption could corrupt the entire solution.

If Alicia spends five minutes checking Question 1 and thirty seconds thinking about Question 7, she is not short of effort.

She is allocating effort badly.

This is the central problem of cognitive effort allocation: mental control is useful but costly, and intelligence must decide where another unit of control is likely to change the outcome.

The 60-Second Route

1. Name the task. What decision or operation is active?
2. Read the stakes. What happens if the answer is wrong?
3. Estimate uncertainty. How fragile is the current route?
4. Identify the available control move. More checking, slower reading, another representation, inhibition of a tempting response, retrieval, external evidence?
5. Estimate expected improvement. How likely is that move to improve the result?
6. Price the effort. What time, attention, fatigue and opportunity cost does it consume?
7. Allocate. Increase, maintain or reduce control.
8. Monitor the return. Is the extra effort producing new information or merely repeating the same work?
9. Stop or switch when marginal value falls.

1. Cognitive Control Is Selective by Design

Cognitive control helps people maintain goals, suppress distracting or habitual responses, coordinate several pieces of information, switch strategies and check performance.

These abilities are especially useful when the automatic response is unreliable, the task is novel, competing responses are active or consequences are large.

But deliberate control is not free. It consumes time and limited processing resources. People often experience demanding control as effortful and tend to avoid unnecessary cognitive demand when easier routes are available.

So a capable system cannot simply maximise control everywhere.

2. The Basic Economic Logic of Mental Control

A major research framework proposes that control allocation can be understood as a cost–benefit problem.

In the Expected Value of Control framework, associated with Amitai Shenhav, Matthew Botvinick and Jonathan Cohen, the system evaluates the payoff expected from exerting control, the amount or intensity of control required and the effort cost associated with that control.

The framework is a model, not a license to reduce all human thought to one equation. Its useful insight is:

More control is worthwhile when the expected improvement it produces exceeds the cost of deploying it.

3. Expected Value Is Not Just Reward

A high-value outcome does not automatically justify unlimited effort.

Control is more attractive when:

• the outcome matters;
• control is likely to improve performance;
• the improvement is large enough to matter;
• the task remains solvable;
• the time cost is acceptable;
• there is no better competing use of the same resources.

A difficult question worth one mark and a difficult question worth ten marks may deserve different control budgets even if they feel equally challenging.

4. Efficacy Matters: Effort Must Be Able to Change the Result

Research extending the Expected Value of Control account distinguishes reward from efficacy: whether exerting control is expected to influence the outcome.

If a learner believes that careful reasoning will probably improve performance, more effort becomes rationally attractive.

If the learner lacks a crucial prerequisite and has no method for resolving the gap, simply exerting more control may produce frustration rather than progress.

This creates an important teaching distinction:

Low effort and low expected efficacy can look similar from the outside.

5. Effort Has Several Prices

Cost	What it means
Time	Deliberation delays completion and reduces time available elsewhere.
Attention	Focused processing makes other signals easier to miss.
Working memory	More active rules, quantities and alternatives increase coordination load.
Fatigue	Sustained demanding control can become harder to maintain.
Opportunity cost	Resources spent here cannot be spent on another task.
Motivational friction	Tasks that appear effortful may be delayed or avoided.

6. Automaticity Is Not the Enemy

Some students hear “think carefully” and conclude that every step should be slow and deliberate.

That would be disastrous in an examination.

Well-trained automatic processes are valuable because they release control for the places where control is needed.

Basic arithmetic facts, standard algebraic transformations, familiar vocabulary and routine checking procedures should eventually become fluent enough that they do not monopolise attention.

The goal is not maximum deliberation.

The goal is reliable automaticity plus rapid escalation when the routine route becomes unsafe.

7. Conflict Is One Signal That More Control May Be Needed

When two responses compete, automatic performance becomes less trustworthy.

A familiar but inappropriate method may be activated. Two interpretations of the question may remain plausible. A remembered answer may conflict with the current evidence.

Conflict does not prove that more effort will solve the problem, but it is a useful escalation signal.

Strong learners become sensitive to the feeling: something does not fit.

8. Surprise Is Another Escalation Signal

A calculation produces a negative length.

A probability exceeds 1.

A percentage discount produces a higher final price.

A comprehension answer contradicts the passage’s central event.

Surprise is not an annoyance to suppress. It is often evidence that the current low-control route has crossed a boundary where checking becomes valuable.

9. Uncertainty Should Change the Effort Budget

Two problems can have identical stakes and require different control because one is stable and the other fragile.

If Alicia knows exactly why her method applies, another five minutes may add little value.

If Tricia is choosing between two plausible models, another minute spent checking assumptions may prevent an entire error cascade.

Uncertainty therefore acts as a multiplier on the possible value of control.

10. Stakes Should Change the Effort Budget Too

Not every uncertainty deserves the same response.

A low-stakes spelling choice in a private note may not justify external verification. A high-stakes medical, legal, financial or engineering decision may require much stronger evidence and independent review.

In examinations, marks provide a rough but useful signal of stakes. A ten-mark structured response deserves more control than a one-mark item once the easy marks have been secured.

11. Reversibility Changes How Much Thinking Is Worth Doing Before Acting

If a decision can be cheaply reversed, rapid action followed by feedback may outperform prolonged pre-analysis.

If a decision is difficult or impossible to reverse, more control before commitment can be valuable.

A student can cross out a tentative diagram cheaply. Submitting the final answer ends that flexibility.

Reversibility therefore affects the optimal point of commitment.

12. Mental Effort Is Not the Same as Physical Effort

Both can feel costly and both can be traded against reward, but cognitive effort refers to demands on processes such as attention, working memory, inhibition, rule maintenance and deliberate reasoning.

A difficult mental task may involve almost no physical exertion while still producing a strong subjective sense of effort.

13. Cognitive Effort Allocation Is Not the Same as Motivation

Motivation concerns why an outcome or activity matters enough to pursue.

Effort allocation concerns how much control should be deployed on the active task given its expected benefit and cost.

A student can care intensely about an examination and still allocate effort badly by polishing easy questions, refusing to move on from one blocked item or studying familiar material because it feels productive.

14. Cognitive Effort Allocation Is Not the Same as Prioritisation

Prioritisation asks, Which task comes first?

Effort allocation asks, How much control does the chosen task receive?

One is ranking. The other is dosage.

A high-priority task can still deserve only a small amount of cognition if the correct action is obvious.

15. Cognitive Effort Allocation Is Not the Same as Cognitive Load

Cognitive load describes demands placed on limited processing capacity.

Effort allocation is the control decision about how much of that limited capacity to invest and where.

A task can impose high load while being strategically unimportant. Another can impose modest load while deserving close scrutiny because the stakes are large.

For the broader learning architecture, see How Cognitive Load Budgeting Works.

16. Cognitive Effort Allocation Is Not the Same as Persistence

Persistence is continuing.

Good allocation may require continuing, switching method, obtaining information, asking for help or stopping.

Persisting with a method whose expected value has collapsed is not disciplined effort. It is sunk-cost behaviour.

17. The Marginal Value Question

The most useful operational question is often not:

“Is more thinking good?”

It is:

“What is the expected value of the next unit of thinking?”

The first check may reveal a sign error. The second may confirm a condition. The fifth identical check may add almost nothing.

Effort should usually fall when marginal information gain falls.

18. Study Example: Rereading a Topic You Already Know

Kai Kai has one hour.

She can reread a chapter she already understands or attempt retrieval on a topic she frequently forgets.

The familiar chapter feels easier and produces fluent recognition. But another thirty minutes of rereading may create little new capability.

The weaker topic may feel harder because retrieval exposes uncertainty. Yet the expected learning return can be higher.

Good effort allocation follows expected improvement, not comfort.

19. Study Example: The Missing Prerequisite

Alicia spends forty minutes on a calculus question and makes almost no progress.

The problem is not insufficient persistence. She cannot manipulate the algebraic form required before differentiation begins.

Another forty minutes attacking the same question has low expected value.

Switching to the prerequisite has higher expected value because it changes the control efficacy of future attempts.

20. Study Example: Worked Example or Independent Practice?

For a complete novice, independent problem solving can consume large amounts of control while producing little structure.

A worked example can reduce unproductive search and make the solution architecture visible.

For a more advanced learner, excessive worked-example dependence can underinvest in retrieval and strategy selection.

The allocation should change with expertise.

21. Examination Example: The One-Mark Trap

Tricia spends seven minutes on a one-mark question because the answer “ought to be easy”.

The effort is now being justified by frustration rather than expected return.

A better control rule is:

If the next minute has low probability of producing the mark and high opportunity cost elsewhere, park the question and return later.

22. Examination Example: The High-Mark Representation Decision

A ten-mark problem can be approached with algebra, graphing or geometry.

The first thirty seconds spent comparing representations may save ten minutes of dead-end execution.

This is high-value early control: a small investment before a large downstream branch.

23. Examination Example: Checking Should Be Risk-Weighted

Checking every line equally is inefficient.

High-value checkpoints include:

• after translating words into an equation;
• after changing units;
• after a sign-sensitive algebraic transformation;
• before accepting a root that may violate a domain condition;
• after a calculator output that is surprising in magnitude;
• before submitting a long response whose conclusion depends on one key inference.

These are places where a small control investment can prevent a large error cascade.

24. The Speed–Accuracy Trade-Off

Many decisions can be made faster by lowering the amount of evidence required before responding, and more accurately by waiting for more evidence.

There is no universally correct point. The useful setting depends on stakes, time pressure, uncertainty and the cost of an error.

Examinations make this trade-off visible because the candidate must balance accuracy on the current question against coverage of the paper.

25. Why Easy Work Can Consume Too Much Control

Easy tasks often provide immediate closure and positive feedback.

This can attract excessive polishing: rewriting notes neatly, rechecking arithmetic already confirmed, perfecting a low-value introduction, reorganising folders instead of studying the unstable concept.

The learner feels busy because control is active.

But active control is not the same as high-value control.

26. Why Hard Work Can Consume Too Much Control Too

Difficulty can also become a trap.

A blocked learner may interpret continuing effort as evidence of virtue and stopping as failure.

But if no new information is entering the system, repeating the same failed move can have near-zero marginal value.

High-quality effort includes changing strategy when the return curve collapses.

27. Sunk Effort Distorts Allocation

“I have already spent twenty minutes on this” is not evidence that another twenty minutes is worthwhile.

Past effort is already spent.

The allocation question is forward-looking:

From this point, what action has the highest expected return?

28. Fatigue Changes the Cost Function

The same task can cost more at different times.

A difficult proof at the beginning of a study session may be manageable. The same proof after several hours of demanding work may require more time, more checking and stronger inhibition of distraction.

Good allocation therefore depends on state, not only task.

This is why planning high-control work into high-quality attention windows can outperform equal scheduling by clock time alone.

29. Stress Can Reallocate Control Poorly

Under stress, people can become hyperfocused on one threat, rush, repeatedly check low-value details or revert to habitual responses.

The examination problem is not simply “stress reduces intelligence”.

Stress can alter which signals receive control and how long the system persists before switching.

Training under realistic time pressure helps learners practise allocation, not merely content retrieval.

30. Confidence Should Influence—but Not Dominate—Control

Very low confidence can justify more checking or a different strategy.

Very high confidence can justify moving on.

But confidence itself can be miscalibrated. Familiarity, recent success and fluent retrieval can create overconfidence.

That is why external checks, answer constraints and alternative representations are useful when the consequence of a confident mistake is high.

31. Metareasoning: Thinking About Whether More Thinking Is Worth It

Modern work on metareasoning examines how people monitor ongoing reasoning and control the resources invested in it.

Monitoring asks: How well is the current reasoning process working?

Control asks: Should I allocate more attention, change strategy, continue or terminate?

A 2025 review of theoretical and methodological developments in metareasoning describes this monitoring-and-control distinction explicitly, including allocation of attention and working memory and termination of failing strategies.

32. The First Weak Link Can Change the Best Effort Allocation

If a learner’s first weak link is conceptual, more execution practice may have low return.

If the concept is secure but arithmetic fluency is weak, targeted execution practice may have high return.

If the learner understands and can execute but fails under time pressure, examination pacing and control switching become the correct target.

Diagnosis therefore precedes efficient effort allocation.

33. More Practice Is Not Automatically More Valuable

Practice changes capability when it exercises the mechanism that needs improvement.

One hundred routine questions may add little if the learner’s problem is selecting the correct method under mixed conditions.

Twenty contrastive questions may produce more value because they train the decision boundary.

Effort allocation therefore applies not only inside a question but across the whole training programme.

34. Retrieval Changes the Cost of Future Control

Knowledge that is well learned becomes faster and cheaper to retrieve.

This matters because fluent retrieval reduces the control burden of later complex work.

A student who must consciously reconstruct every multiplication fact has fewer resources available for algebraic structure. A reader who decodes every word laboriously has fewer resources available for inference.

Training can therefore be viewed partly as changing the future economics of control.

35. Tools Can Reduce or Redirect Cognitive Effort

Calculators, notes, diagrams, search systems and software can offload some cognitive operations.

That can be beneficial when the tool removes low-value processing and frees control for interpretation, verification or decision-making.

But offloading can become harmful when the learner outsources the very mechanism that must remain available under test conditions or for independent judgement.

For the dedicated treatment, see How Studying Works | Cognitive Offloading.

36. Search Has an Effort Budget Too

More searching can improve an answer by adding evidence.

It can also waste time, increase noise and create false confidence from quantity.

The value of another search depends on whether it is likely to change the decision.

Once several strong, independent sources agree and the remaining uncertainty is immaterial, continued search may have low marginal value.

37. Teams Allocate Cognitive Effort Too

Organisations decide which issues receive:

• senior review;
• independent checking;
• specialist analysis;
• simulation;
• formal audit;
• additional testing;
• extended meetings.

These are collective control-allocation decisions.

A weak organisation can overprocess trivial matters and underprocess consequential assumptions.

Good governance puts expensive cognition where uncertainty and consequence intersect.

38. Review Gates Are Institutional Effort Allocation

Not every decision needs a committee.

Not every decision should be made casually.

Materiality thresholds, technical review gates, triage protocols and escalation rules are ways of precommitting additional cognitive resources when risk crosses a threshold.

39. Artificial Intelligence Faces the Same Allocation Problem

An AI system can answer immediately, retrieve sources, call a calculator, inspect a file, compare alternatives, run a stronger reasoning process or escalate to a human.

Each additional operation has cost in latency, compute, complexity and sometimes external-resource usage.

Too little effort produces brittle answers. Unlimited effort wastes resources and can increase opportunities for drift.

The general design principle is the same:

Escalate reasoning when another step has meaningful expected value for accuracy, safety or decision quality.

40. Effort Allocation Failure Atlas

Failure mode	What happens	Repair
Maximum-control default	Routine work becomes slow and exhausting	Allow trained automaticity where reliability is high
Automaticity everywhere	Habit dominates novel or high-stakes cases	Escalate on conflict, surprise and uncertainty
Easy-task polishing	Control is spent where confidence is already high	Move resources to unstable decisions
Difficulty worship	Hard-feeling work is assumed valuable	Measure expected learning or decision return
Sunk-effort persistence	Past effort justifies more effort	Evaluate only the next unit of effort
Fatigue blindness	The cost of control rises but strategy stays fixed	Reallocate, rest or switch task
Stakes blindness	Low- and high-consequence decisions receive equal checking	Use marks, risk and reversibility
Efficacy blindness	More effort is spent where control cannot solve the missing prerequisite	Change the problem state first
Premature stopping	Search ends before key uncertainty is resolved	Ask what remaining uncertainty could change the decision
Endless checking	Repeated verification adds no new information	Use stopping rules

41. A Practical Exam Control Budget

• Low uncertainty + low marks: execute fluently, quick check, move.
• Low uncertainty + high marks: execute fluently, protect key transitions, verify final conditions.
• High uncertainty + low marks: attempt briefly, park if marginal return collapses.
• High uncertainty + high marks: invest early in interpretation and representation before long execution.
• Unexpected answer: escalate control immediately.
• Blocked with no new idea: switch representation, retrieve prerequisite, or move and return.
• Final minutes: prioritise checks with the highest probability of recovering marks.

42. A Practical Study Control Budget

1. Identify which learning outcome is unstable.
2. Choose the activity most likely to change that instability.
3. Use focused effort until new learning or diagnostic information stops appearing.
4. Switch from explanation to retrieval when the concept becomes understandable.
5. Switch from blocked practice to prerequisite repair when efficacy falls.
6. Use spacing rather than continuous repetition once immediate fluency is achieved.
7. Return later to test whether capability survived.

43. The Control Escalation Triggers

More deliberate control is especially worth considering when one or more of these appears:

• two plausible interpretations;
• high consequences;
• unexpected output;
• novel structure;
• conflict with prior knowledge;
• weak source confidence;
• large downstream dependence on one assumption;
• irreversible commitment;
• low confidence near a decision boundary;
• evidence that a habitual response is unreliable.

44. The Control De-Escalation Triggers

Control can often be reduced when:

• the task is routine and well trained;
• independent checks agree;
• remaining uncertainty cannot change the decision;
• the next check repeats the same information;
• opportunity cost elsewhere is rising;
• the problem requires a missing prerequisite rather than more effort;
• the action is cheap to reverse and feedback will arrive quickly.

45. A Counterfactual Test for Effort

Before investing another block of effort, ask:

If I spend the next five minutes here, what specific state might improve?

If the answer is concrete—resolve an ambiguity, check a sign-sensitive transformation, recover a definition, compare two models—the effort has a plausible mechanism.

If the answer is only “I will keep looking at it,” expected value may be low.

46. The Receiver Test

Effort allocation succeeds only if the learner can still perform the next job.

A study method that consumes so much attention that the student cannot later retrieve, choose and execute under examination conditions has not allocated resources well.

The target is usable capability, not maximal exhaustion.

47. The World-Return Test

After the examination, project or decision, compare the effort allocation with what actually mattered.

• Where did extra checking recover value?
• Where did thinking continue long after the answer was stable?
• Which routine failed unexpectedly?
• Which uncertainty should have triggered escalation earlier?
• Which prerequisite would have made control more effective?

Future allocation improves when past returns are observed rather than guessed.

48. How Do We Know?

Cognitive effort allocation sits at the intersection of cognitive control, motivation, decision-making and metacognition.

The Expected Value of Control framework proposed by Shenhav, Botvinick and Cohen in 2013 formalised the idea that control allocation can be understood by integrating expected payoff, control intensity and effort cost. Later work tested how expected reward and expected efficacy influence the amount of control people deploy.

Research on mental effort more broadly treats effort as costly and examines why people discount rewards that require greater cognitive demand.

Recent work on metareasoning adds a complementary perspective: people monitor the effectiveness of ongoing cognition and use control processes to allocate attention and working memory, initiate strategies and terminate failing ones.

No single framework settles every question about why effort feels costly or how control is implemented in the brain. The practical synthesis used here is narrower: control is limited, it has costs, its benefits vary by state and task, and intelligent performance depends partly on matching control to expected return.

Evidence and Further Reading

• Shenhav, Botvinick & Cohen (2013), The Expected Value of Control — foundational cost–benefit framework for allocating cognitive control.
• Shenhav, Cohen & Botvinick (2016), Dorsal Anterior Cingulate Cortex and the Value of Control — development of the control-allocation account.
• Grahek et al. (2020), Expectations of Reward and Efficacy Guide Cognitive Control Allocation — evidence that expected reward and efficacy jointly shape control investment.
• Shenhav et al. (2017), Toward a Rational and Mechanistic Account of Mental Effort — review of effort, cognitive control and cost.
• Grahek, Musslick & Shenhav (2021), Decomposing the Motivation to Exert Mental Effort — review of cost–benefit determinants of effort.
• Metareasoning: Theoretical and Methodological Developments (2025) — monitoring, control, strategy selection and termination in reasoning.

49. What This Article Does Not Claim

• It does not claim people consciously calculate a numerical expected value before every act of thought.
• It does not claim one brain region single-handedly controls all effort allocation.
• It does not claim high effort is bad.
• It does not claim low effort is efficient by definition.
• It does not claim automatic processing is always inferior to controlled processing.
• It does not claim every difficult task should be abandoned when progress slows.
• It does not claim motivation, cognitive load, prioritisation, persistence and effort allocation are the same construct.
• It does not claim examination marks are the only measure of value.

50. Teaching Guide

Begin by giving students three questions: one routine and low-mark, one unfamiliar and low-mark, and one unfamiliar and high-mark. Ask them not to solve yet. Instead, allocate an effort budget to each.

Then ask what evidence justified the budget. Students should identify stakes, uncertainty, novelty, expected usefulness of checking and opportunity cost.

Next, give one worked solution containing a deliberate conflict point. Ask students to mark where automatic execution should stop and deliberate checking should begin.

Finally, run a timed mixed set. Students must record one moment when they escalated control, one moment when they deliberately moved on and one moment when a later return changed the answer. Review whether those allocations recovered marks or wasted time.

The aim is not to teach students to think less. It is to teach them to recognise where thinking harder changes the result.

51. Key Ideas to Carry Forward

• Cognitive control is valuable because it is selective.
• Effort should track expected improvement, not discomfort or pride.
• Reward and efficacy both matter.
• Conflict, surprise, uncertainty and high stakes can justify escalation.
• Routine fluency is useful when it frees control for harder decisions.
• Past effort does not justify future effort.
• The next unit of thinking should be judged by its marginal value.
• Fatigue and stress change the cost of control.
• Checking should be concentrated at high-risk transitions.
• Missing prerequisites often require a state change rather than more persistence.
• Good learners know when to persist, switch, verify, retrieve, park or stop.

Conclusion: The Best Mind Is Not the One at Maximum Effort

Intelligence does not become stronger simply by turning every control process to maximum.

A strong mind protects automaticity where it is trustworthy and spends deliberate control where another unit of attention, checking or reasoning can materially improve the result.

It notices conflict. It respects stakes. It monitors uncertainty. It prices time. It recognises sunk effort. It changes strategy when efficacy falls. It stops when additional thought stops changing the answer.

Thinking harder is a resource decision. Intelligence is learning where that resource pays.

Continue the intelligence system: How Intelligence Works · Prioritisation · Working Memory · Value of Information · Stopping Rules · Abstention