Comparing every pair of items on a 3-item list takes 3 comparisons. On a 20-item list, it takes 190 comparisons: each item must be compared with nineteen others, with each pair counted only once. The list has grown by less than sevenfold, but the comparison work has grown more than sixtyfold.
This explains why deciding what matters most on a long todo list can produce fog, dread, or a blank mind. The task is an O(n²) operation running on working-memory capacity of roughly 7 items. Those feelings are the error signal produced when the operation exceeds available resources.
Algorithmic complexity describes how the work required by an operation grows with its inputs. Willpower accounting gives the price of a task, such as 6 units. Complexity analysis identifies what generates that price, which makes it possible to change the operation itself.
Cost Structure, Not Just Price
Knowing that choosing what to work on costs 6 units of willpower establishes that it is expensive. The cost comes from exhaustive comparison across the available options. Each comparison is small, but the number grows quadratically. Three options require 3 comparisons; twenty require 190.
Different operations grow at different rates. An O(1) operation, such as recalling a fact from working memory, takes constant time and costs almost nothing. An O(n) operation visits each item once. Writing a braindump is linear because you externalize your mental state one item at a time. An O(n²) operation compares each item with every other item and becomes expensive quickly. An O(e^n) operation branches exponentially and can become intractable.
A question such as “What's the meaning of life?” can start the last kind of operation. Each answer creates more questions without a termination condition, and the search continues branching until resources are exhausted.
The distinction between writing and processing a braindump shows where to divide the work. Only you can sequentially traverse your own mental state, so its O(n) externalization cannot be outsourced to AI. Ranking the resulting items, finding priorities, and grouping themes is O(n²) comparison work that AI can perform without fatigue. Asking AI to help produce the braindump while retaining all the ranking reverses the useful division: it outsources the cheap operation that only you can do and keeps the expensive one that AI can absorb.
A closer look
Two ways to choose
The article asks whether a task's procedure creates more work than the decision needs.
Read this diagram
Compare Compare every pair; Use a simpler selection rule.
Questions as Programs with Complexity Classes
A question initiates a search, and its wording determines which search has to run. “What's 2+2?” calls for an O(1) lookup of an existing answer. “What's on my mind right now?” initiates an O(n) traversal. “Which of these 20 todos is most important?” initiates O(n²) comparison among the items and their combinations.
“What should I do with my life?” opens an O(e^n) search. Possible answers branch into questions about meaning, value, time, constraints, and preferences, which then combine with each other. The question has no clear stopping condition.
Question theory uses this cost structure to distinguish a cheap lookup from an expensive combinatorial search. More effort cannot make an intractable question fit the available hardware, any more than willpower can make a computer solve the traveling salesman problem in polynomial time. Changing the question changes the operation being requested.
The Visibility Problem
The algorithm is difficult to identify while it is running because execution and analysis use the same cognitive resources. A task that occupies those resources leaves no spare capacity to inspect itself. This is an observer problem: the CPU needed to observe the CPU is already busy.
People experience “deciding,” not “running exhaustive pairwise comparison.” Even when the operation becomes visible, it can still seem like a fixed feature of thought. Without a concept of replaceable algorithms, someone sees “how I work” where there is one implementation among several. The missing idea is a function pointer: the same task can be assigned to a different procedure.
That is why encountering an alternative can change more than one task. It reveals that thought has procedures that can be replaced. The cybernetic view makes the constraints explicit: an embedded person has finite processing resources and runs particular operations on hardware with particular limits.
Analysis Paralysis Decoded
Analysis paralysis occurs when exhaustive comparison exceeds working-memory capacity, roughly 7±2 items. Comparing 20 items requires holding and manipulating more than the approximately 7 items the hardware can actively retain. The fog or blank mind signals that the operation does not fit.
“Stop overthinking” does not change that mismatch. A diagnosis of O(n²) comparison on bounded memory identifies several possible interventions: reduce the inputs, use another algorithm, move the work to different hardware, accept an approximation, or recognize that the question needs to be reformulated.
The Intervention Toolkit
Reducing n means narrowing the choices before comparing them. Elimination criteria or a time-bounded decision space can reduce twenty options to a manageable top 5.
Changing the algorithm changes how the remaining options are processed. Tournament-style elimination uses O(n log n) work rather than O(n²) for ranking. Satisficing takes O(n): stop at the first option that is good enough. Threshold filtering also takes O(n): evaluate each option against a bar instead of against every other option.
Offloading moves comparison to AI, whose constraints differ from human memory and fatigue. Approximation accepts a good-enough answer when finding the optimum is intractable. Recognizing intractability means declining to run the original question and recasting it in a form that can be answered.
Each intervention changes the relationship between the operation and the available resources. Trying harder or treating indecision as a character defect leaves that relationship unchanged.
The AI Offloading Asymmetry
An operation can be expensive on one architecture and cheap on another. Human working-memory capacity is around 7 items. Comparing item 15 with item 3 requires reloading the earlier item from long-term storage, while AI has a full context window without that reload penalty.
Humans also degrade after roughly 20 comparisons: fatigue reduces comparison quality and evaluation criteria drift during ranking. AI maintains constant quality across 1000 comparisons and applies a rubric uniformly. It can batch and parallelize work that a person performs serially.
Dense comparison tasks therefore offer high leverage: processing a braindump, ranking 50 candidate ideas, classifying 100 journal entries against a taxonomy, or evaluating several design options against several criteria. These are O(n²) or worse, draining for a person and trivial for AI.
The division should follow the sub-problem. The person supplies the O(n) traversal of their private mental state; AI performs the O(n²) ranking after the items have been externalized.
The Cybernetic Pairing
Complexity analysis can itself become an unbounded task. Analyzing the complexity of an operation can lead to analyzing that analysis, then analyzing it again. Cybernetics supplies the stopping condition: resources are finite, analysis consumes them, utility is local, and further analysis should stop when its cost exceeds its value.
Without that constraint, algorithmic analysis can continue indefinitely without producing action. Without complexity analysis, the cybernetic view establishes that resources are limited but does not identify which operation consumes them or how to change it.
Together, the two identify both the source of cost and the point at which diagnosis has done enough. The same analysis that rejects an expensive brute-force comparison must reject further self-analysis when it exceeds the utility threshold.
Early use can involve classifying everything as an intellectual exercise. Mature use includes recognizing when classification costs more than it saves. Cybernetics keeps that judgment tied to available resources and the current purpose.
Meta-Framework Properties
Complexity provides a common measure across cognitive techniques. It applies to a framework as well as to the task the framework addresses, making it possible to compare their operations on the same basis.
The comparison can run in either direction. For a new technique, ask which operation it changes and whether the improvement is O(n²) → O(n) or O(n) → O(1). That indicates whether adoption is worthwhile and by how much. For an established technique, ask which reduction explains its value. The answer either identifies the useful mechanism or reveals apparent productivity without a structural improvement.
The framework quantifies return on investment precisely. A personal lexicon replaces O(n) reconstruction of a concept with O(1) lookup, with the saving repeated every time the term is used. A night protocol replaces morning O(n²) comparison about what to work on with O(1) lookup of the prepared note. Activation energy falls over 30 days as expensive interpreted execution is compiled into a cheaper routine.
The common questions are which operation dominates, how its cost grows, and where the bottleneck lies. They can generate a technique by identifying an expensive operation that needs replacement. They can also evaluate an existing technique by checking whether it reduces complexity or adds overhead. Decision-making, prioritization, journaling, and planning no longer need unrelated explanations for their costs.
Practical Recognition
A felt cost often appears before its operation can be named. A long list produces dread; a question about priority produces fog; twenty todos produce a blank mind while three remain easy to handle. These signals indicate an O(n²) operation exceeding working-memory capacity.
Recognition can develop in stages. After a task, the operation can be identified retrospectively. Repeated encounters then make its signature recognizable before exhaustion. Eventually, asking the task's complexity class before beginning allows the procedure to be chosen in advance.
This still requires room to think. Stepping out of execution frees resources for analysis. The cybernetic view explains why the two cannot both run at full capacity: they share the same resource pool. An embedded observer has to allocate resources to inspecting the operation as well as to performing it.
Algorithmic complexity identifies what makes a cognitive task expensive. When exhaustive comparison exceeds working-memory capacity, reducing the inputs, replacing the algorithm, offloading, or accepting approximation can make the task feasible. Working-memory limits constrain execution; cybernetics also constrains the analysis, which should stop when it costs more than it saves.
Related Concepts
- cybernetics provides resource limits and a stopping condition for complexity analysis.
- question-theory describes how a question's structure selects a search algorithm.
- working-memory supplies the ~7-item limit that comparison quickly exceeds.
- willpower records the price that the algorithm's cost structure generates.
- activation-energy decreases as interpreted execution is compiled over 30 days.
- the-braindump separates O(n) externalization from O(n²) processing.