Cognitive Load: Sweller's Theory, Miller's Number, and How Operators Cut the Load That Is Not the Work

Your working memory is small, it was always small, and most of what fills it on a working day is not the work.

Operators do not usually fail on the hard problem. They fail on the forty small things sitting in the mind at the same time as the hard problem: the unanswered message, the half-decided hire, the invoice that might be wrong, the client who went quiet. Each one is cheap on its own. Together they crowd out the thinking the business is paying you for.

Cognitive psychology has a name for that crowding and a body of research on how much a mind can hold before it degrades. The numbers are smaller than most people assume, and the fixes are structural rather than heroic.

This article covers what cognitive load is and the three kinds the theory names, what working-memory research actually found about capacity, where an operator's load comes from, the Stoic version of the reduction, five moves that cut the load that is not the work, and a seven-day audit.

What cognitive load is

The reference entry defines it in one line: "In cognitive psychology, cognitive load is the effort being used in the working memory." Working memory is the small workspace where you hold and manipulate information right now, as opposed to the long-term store you retrieve from.

The theory behind the term is cognitive load theory, which the same entry says "was developed in the late 1980s out of a study of problem solving by John Sweller," and that "Sweller argued that instructional design can be used to reduce cognitive load in learners." The founding paper, Sweller's 1988 article in Cognitive Science, made a specific argument: that "conventional problem solving in the form of means-ends analysis requires a relatively large amount of cognitive processing capacity which is consequently unavailable for schema acquisition."

Translate that out of the classroom. When you solve a problem by juggling the goal, the current state and every possible step at once, the juggling itself consumes the capacity you needed to actually learn the pattern. That is an operator's Tuesday.

The three kinds of load

The theory names three types, and the reference entry defines them briefly:

• Intrinsic. "Intrinsic cognitive load is the effort associated with a specific topic." The difficulty that belongs to the problem itself. Pricing a new offer is intrinsically hard. You do not want to remove this load; it is the work.
• Extraneous. "Extraneous cognitive load refers to the way information or tasks are presented to a learner." The difficulty that comes from how the task arrives: scattered inputs, unclear ownership, six tabs. This is the load to cut.
• Germane. "Germane cognitive load refers to the work put into creating a permanent store of knowledge (a schema)." The effort that builds a reusable pattern, so the next instance is cheaper. This is the load to protect.

The entry adds a caution worth keeping: the additivity of the three types "has been investigated and questioned," and "Now it is believed that they circularly influence each other." Treat the three as a lens, not a formula. The lens still does the job: most of an operator's day is extraneous load wearing the costume of work.

What working memory can actually hold

The most famous number in psychology is seven. George Miller's 1956 paper, "The Magical Number Seven, Plus or Minus Two", opens: "My problem is that I have been persecuted by an integer." He proposed a term for the limit, "the span of absolute judgment," and maintained "that for unidimensional judgments this span is usually somewhere in the neighborhood of seven."

Two of Miller's other points matter more to an operator than the seven. First, the unit is not an item but a chunk: "The span of immediate memory seems to be almost independent of the number of bits per chunk, at least over the range that has been examined to date." Second, the way around the limit is to recode: "recoding is an extremely powerful weapon for increasing the amount of information that we can deal with." A system, a checklist or an SOP is recoding. It turns twelve things into one chunk.

The seven did not survive as a hard figure. Nelson Cowan's 2001 review in Behavioral and Brain Sciences, "The magical number 4 in short-term memory", states that Miller's number "was meant more as a rough estimate and a rhetorical device than as a real capacity limit." Bringing together a wide range of data, Cowan concludes that "A single, central capacity limit averaging about four chunks is implicated along with other, noncapacity-limited sources."

About four. That is the workspace. Four chunks, held at once, before something falls out. Every open loop you carry into a decision is occupying one of them.

What switching costs

Load is not only about how much is held. It is about how often you change what is held. A 2001 study in the Journal of Experimental Psychology, "Executive control of cognitive processes in task switching," ran four experiments in which participants alternated between tasks or repeated one. The finding: "Task alternation yielded switching-time costs that increased with rule complexity but decreased with task cuing." The authors also found that "rule activation takes more time for switching from familiar to unfamiliar tasks than for switching in the opposite direction."

Complex rules, no cue, unfamiliar task: that describes switching from a client call to the tax question to the proposal draft. Each switch is paid for in time, and the research says the price rises with complexity.

Interruptions carry a second cost. A 2008 study presented at CHI, "The Cost of Interrupted Work: More Speed and Stress," found something counter-intuitive: "people completed interrupted tasks in less time with no difference in quality." The price was elsewhere. "Our data suggests that people compensate for interruptions by working faster, but this comes at a price: experiencing more stress, higher frustration, time pressure and effort." You can absorb an interrupted day and still ship. You pay for it in the state you end the day in, which is the state you start tomorrow with.

Where an operator's extraneous load comes from

None of the research above was run on founders. The mechanisms transfer anyway, because a working memory does not know whether it is in a laboratory or a coaching business. Four sources account for most of the load that is not the work.

• Open loops. Every undecided thing you are carrying in your head rather than on paper occupies a chunk. The reply you owe, the price you have not set, the hire you have not finished thinking about. They do not wait quietly; they resurface during the thing you are trying to do.
• Switching. The inbox, the phone and the calendar are switching machines. Each context change pays the cost the task-switching study measured, and the complex, unfamiliar switches cost most.
• Unclear decisions. A decision with no rule attached must be made from scratch every time it appears. Whether to take the call, whether to discount, whether to answer tonight. Each one is an intrinsic problem that should have been extraneous, meaning solved once and stored.
• Unstructured inputs. A client brief spread across a voice note, three messages and a shared folder is Sweller's point in miniature. The format of the information, not the information, is doing the damage.

Notice that none of these is the hard problem. The hard problem is the intrinsic load you want all four chunks for. The four sources above are what you are spending them on instead.

The Stoic reduction

Marcus Aurelius had no working-memory research and arrived at the same conclusion by observation. In Book IV of the Meditations, in the Long translation, he quotes and then sharpens an older saying: "Occupy thyself with few things, says the philosopher, if thou wouldst be tranquil." His own version is more precise about why:

For the greatest part of what we say and do being unnecessary, if a man takes this away, he will have more leisure and less uneasiness. Accordingly, on every occasion a man should ask himself, Is this one of the unnecessary things?

Is this one of the unnecessary things? That is an extraneous-load filter, written down in the second century. He goes further in the same passage: take away "not only unnecessary acts, but also unnecessary thoughts." The open loop is the unnecessary thought.

In Book II he describes the alternative state: "Every moment think steadily as a Roman and a man to do what thou hast in hand with perfect and simple dignity, and feeling of affection, and freedom, and justice, and to give thyself relief from all other thoughts." One thing in hand. Relief from all other thoughts. That is the full working memory on the intrinsic load, and nothing else.

The dichotomy of control is the Stoics' other load filter. Anything outside your control that you are nonetheless holding in mind, the client's decision, the market, the competitor's launch, is a chunk spent on something no amount of attention will move. Sort it out and the workspace clears. The business application is in how to stop overthinking.

Five moves that cut the load that is not the work

Each move below takes something out of working memory and puts it somewhere that does not degrade: paper, a rule, a document, a system or a calendar. This is Miller's recoding applied to a business.

1. The written open-loop list. Every undecided or unfinished thing goes on one list, on paper or in one file, the moment it appears. Not a task manager with forty categories; one list. The act of writing it moves the chunk out of the workspace. Review it at the end of each block and at the end of the day, so the mind learns the list is trustworthy and stops rehearsing.
2. One document per block. A work block has one output and one open document. Everything else is closed, including the inbox. This is the direct answer to the task-switching research: no alternation, no switching cost. The structure for it is time blocking, and the protected version is deep work.
3. Decision rules. Any decision that has appeared three times gets a rule: no discounts below a set floor, no calls before noon, every proposal gets a follow-up date when sent. A rule converts an intrinsic problem into a stored answer. It also removes the fatigue that builds through a day of unmade decisions, which is the subject of decision fatigue.
4. SOPs for anything repeated. A procedure written once is a chunk you never have to reconstruct. Client onboarding, the weekly report, the content publishing sequence. Each SOP is recoding in Miller's sense: twelve steps become one named thing. The method is in how to write SOPs.
5. A reviewed calendar. Decide the shape of tomorrow the evening before, in writing, so the morning does not begin with the decision of what to do. Then review the week on Friday: which blocks held, which were interrupted, what the interruptions cost. The review is where germane load lives, because it is where the pattern gets built.

Training belongs on this list too, though it is not a cognitive technique. A session with one lift and one number to beat is the lowest extraneous load an operator will experience all day, and it demonstrates, in the body, what a single focused block feels like. Operators who train before the business day carry that state into the first block.

The seven-day load audit

You cannot cut what you have not measured. One week, pen and paper, no new tools.

1. Day 1: count the loops. Write down every open loop in your head before you open the laptop. Count them. Most operators find between twenty and fifty on the first attempt. That number is your extraneous load baseline.
2. Day 2: count the switches. Keep a tally mark for every time you change task before the current one is done. Do not change behaviour yet; just count.
3. Day 3: sort the loops. Take the day-one list and split it three ways: decide now, needs a rule, outside your control. The third column is deleted, with the Stoic question attached: is this one of the unnecessary things?
4. Day 4: write the rules. For everything in the "needs a rule" column that has appeared more than twice, write the rule in one sentence. Ten rules is a good first set.
5. Day 5: one block, one document. Run a single ninety-minute block with one output and nothing else open. Note how many times you reached for a switch. That reaching is the habit the research measured.
6. Day 6: write the first SOP. Pick the task you did most often this week and write its steps. One page. It is now a chunk.
7. Day 7: review. Recount the open loops. Compare to day one. Count the rules written and the switches you declined. Decide which two moves from the five above get installed as permanent appointments next week.

The audit does not make you smarter. It gives the four chunks back to the work.

Frequently asked questions

What is cognitive load?

Cognitive load is the effort being used in working memory, the small mental workspace where information is held and manipulated in the moment. Cognitive load theory, developed by John Sweller in the late 1980s, distinguishes intrinsic load (the difficulty of the task itself), extraneous load (difficulty created by how the task is presented) and germane load (effort that builds reusable knowledge).

How many things can working memory hold?

Miller's 1956 paper proposed a span "somewhere in the neighborhood of seven" for simple judgements, but Cowan's 2001 review concluded that number was "a rough estimate and a rhetorical device" and that the evidence points to "a single, central capacity limit averaging about four chunks." The practical figure for an operator is about four, and the unit is a chunk, not an item.

What are the three types of cognitive load?

Intrinsic load is the effort tied to the topic or problem itself. Extraneous load comes from the way information or tasks are presented, and is the type to reduce. Germane load is the work of building a permanent pattern or schema, and is the type to protect. Researchers now believe the three influence each other rather than simply adding up.

Does multitasking increase cognitive load?

Switching between tasks carries measurable time costs. A 2001 study found switching costs "increased with rule complexity but decreased with task cuing," and a 2008 interruption study found people finished interrupted tasks faster but with "more stress, higher frustration, time pressure and effort." Each switch spends working memory on reorienting rather than on the task.

How do you reduce cognitive load at work?

Move what you are holding in your head into stores that do not degrade: a single written open-loop list, one document per work block, written decision rules for anything that recurs, SOPs for repeated procedures, and a calendar decided the evening before and reviewed weekly. Each is a form of what Miller called recoding, turning many items into one chunk.

Give the four chunks back to the work

The research is consistent on the shape of the problem. Working memory holds about four chunks. Switching costs time and ends the day in a worse state. And most of what fills the workspace is extraneous: the format of the day, not the substance of the problem.

Marcus Aurelius got there without the studies. Ask of everything, is this one of the unnecessary things, and take away the unnecessary thoughts along with the unnecessary acts. The five moves above are the modern version: write it down, close the tabs, make the rule, write the procedure, decide the day before. Training does the same for the body, one lift, one number, nothing else in the room.

The free 5-Day Stoic Operator Challenge installs the daily review and the single-block structure that make the reduction permanent. Five days, one page a night, and the workspace starts to belong to the work again.