Trained Interruptibility

It is twenty to four and you could not tell anyone what you did today.

You were at the desk by eight. You did not slack off, take a long lunch, or spend the morning on anything you'd be embarrassed to describe. You were busy in a way that felt continuous — there was always something in front of you, you were always responding to something, you never once sat idle.

And the thing you actually needed to do today has not moved.

The word waiting for you at twenty to four is laziness, and you will use it on yourself tonight. It is the wrong word, and it is wrong in a way that matters, because laziness is the absence of effort and you spent every minute of that day expending effort. Nothing was withheld. It was all put in, and none of it accumulated.

What happened to you is more specific than laziness, it has been measured with a stopwatch, and the people who measured it think you were trained.

CHAPTER I: THE MEN WITH THE STOPWATCHES

Between the late nineties and the mid-2000s, Gloria Mark and Victor González did something that sounds tedious and turned out to be devastating. They followed information workers around their offices and wrote down what they did, moment by moment, all day.

Not surveys. Not self-report. A researcher standing behind you with a timing sheet, recording every activity you engaged in, for three full days each.

In the 2005 study, twenty-four informants were observed for an average of twenty-five hours and forty-two minutes apiece — over seven hundred formal hours of observation. The researchers grouped activity into what they called working spheres: coherent bundles of work with their own purpose, people, and documents. Preparing the budget is a working sphere. Handling a client's complaint is another.

Their informants were running an average of 11.7 working spheres.

Not tasks. Spheres. Twelve separate contexts, each with its own state to hold in mind.

Time spent inside one before switching to another: about eleven minutes. And 57 percent of working spheres were interrupted before completion. An earlier study by the same researchers had clocked the switching between individual work events at roughly every three minutes.

There is one more finding in that paper that reads like a curse. The longer someone worked inside a working sphere — the deeper in they were, the more loaded up with context — the more likely it was to be interrupted, and the longer the interrupting event lasted. Mark and her colleagues called it the irony of the workday. The reward for getting properly into something is a bigger interruption.

CHAPTER II: NOBODY PROMPTED THE EIGHTEEN PERCENT

Now the study that names the mechanism.

In 2011, Laura Dabbish, Gloria Mark and Victor González analysed 889 hours of observed task-switching from 36 individuals across three technology organisations. They took every task switch in the dataset — 5,089 of them — and sorted each one by what caused it.

Three thousand and fifty-nine switches, 60 percent, came from finishing something. The event ended and the person moved on. That is work behaving properly.

One thousand one hundred and forty-one, 22 percent, came from external interruption. Someone walked over. The phone rang. A notification landed.

And 889 switches, 18 percent, were self-interruptions.

Their definition is worth quoting exactly, because it forecloses the excuse: a switch was coded as a self-interruption if, for no apparent reason, the person stopped what they were doing before the event was complete and switched to something else, without prompting by any external event or entity.

Nothing happened. Nobody arrived. No sound was made. Roughly one in five times you abandon what you are doing, the call is coming from inside the house.

Then the researchers asked whether those self-interruptions were random, and found they were not. External interruptions experienced in the previous hour significantly increased the incidence of self-interruption in the subsequent hour.

Get interrupted a lot at eleven. Interrupt yourself more at twelve, unprompted, with nobody bothering you.

The authors' own interpretation, in their own words: people may be conditioned to self-interrupt. By experiencing external interruptions, they may become habituated to self-interrupt.

That is the whole thesis of your workday, stated by researchers in a conference paper in 2011. You were interrupted, repeatedly, for years. The pattern took. Now you supply it yourself, at no cost to anyone, in an empty room, on a Saturday.

Be precise about the size of it, though, because the paper is more careful than the headline would be. External interruption accounted for only about 4 percent of the variance in self-interruption. Individual differences between people accounted for around 14 percent — and interestingly, job role explained nothing at all. So the conditioning effect is real, directional and statistically significant, and it is a modest slice of the picture. Who you are, and where you sit, matter more.

Which is precisely why the fix is not a resolution. We'll come back to where you sit.

CHAPTER III: THE ONE YOU GAVE YOURSELF COSTS MORE

Here is the finding that should genuinely disturb you.

In the same 2011 data, self-interruptions were compared against external interruptions on how long it took people to get back. Self-interrupted tasks took longer to resume than externally interrupted ones.

The interruption you chose is more expensive than the interruption inflicted on you.

That inverts the entire folk model. We think of the boss appearing at the desk as the real damage and the quick glance at your own phone as the cheap version — a small, self-controlled diversion, easily reversed because you're the one driving. In the data it is the other way around. When someone else interrupts you, the interruption has an obvious external cause and an obvious end, and you come back. When you interrupt yourself, you were the mechanism, and there is no external event whose conclusion returns you.

And how long does coming back take? Here we should kill a number you have certainly seen.

You have read that an interruption costs 23 minutes and 15 seconds. That figure has been repeated for two decades and it does not appear in any of the papers it is attributed to; it appears to trace to an interview, not a result.

The real number in Mark's 2005 study, for interrupted work resumed the same day, is an average of 25 minutes and 26 seconds. But read the dispersion, because it is the honest part: the standard deviation is 54 minutes and 48 seconds — more than double the mean. There is no reliable per-interruption tax. Some come back in a minute; some do not come back for hours.

The more useful figure sits next to it. Before returning to interrupted work, informants passed through an average of 2.26 other working spheres. You do not get interrupted and return. You get interrupted, and then attend to two entirely different topics, and then return to a context you now have to rebuild from scratch.

Then there is the result that explains why no organisation has ever fixed this.

In 2008, Mark, Daniela Gudith and Ulrich Klocke ran a controlled experiment on interrupted work. People completed interrupted tasks in less time than uninterrupted ones, with no difference in quality.

Faster. Same quality. Which, if you are looking at output, means nothing is wrong.

The subjects paid for that speed somewhere else. Interrupted work produced significantly more stress, more frustration, more time pressure, and more effort. People compensate for interruption by working harder and faster, and the compensation works — on the deliverable.

So the cost of a fragmented day does not appear in what you produced. It appears in you. There is no line on any report where this shows up, no dashboard that flags it, no manager who sees it, which is exactly why the arrangement persists and why the only person who ever notices is the man sitting at his desk at twenty to four wondering what is wrong with him.

Nothing was wrong with him. He paid the bill in the only currency the system doesn't measure.

CHAPTER IV: WHERE YOU SIT

If the conditioning effect were the whole story, the fix would be to stop being interrupted, and you would have tried that and failed.

The larger lever in the data is duller and far more actionable.

Dabbish and her colleagues examined one feature of the physical environment: office design. Sitting in an open-plan cubicle rather than an enclosed office was associated with a 64 percent increase in the rate of self-interruption.

Not a 64 percent increase in being interrupted by others. A 64 percent increase in interrupting yourself.

That is the finding to hold onto, because it locates the problem outside your character entirely. The same man, with the same discipline and the same job, generates dramatically more self-interruption depending on the room he is in. The environment is not a backdrop to your willpower. It is a large fraction of the measured variance.

And now the compounding, which is where this becomes a life rather than a bad afternoon.

The training does not stay at work. A man conditioned across a decade of open-plan days does not switch it off at six. The pattern is in the hands: the phone at the red light, the second screen during the film, the reach for something during the four seconds it takes a page to load. You are running the schedule in rooms where nobody has interrupted you in years.

Then the capability itself erodes, and the erosion is invisible because you never get feedback on what you can no longer do. You do not receive a notice saying you have lost the ability to hold a hard problem for ninety minutes. You simply find that the hard problems stop getting solved, and you explain that with the word from the beginning of this piece.

Worst of all, the training makes uninterrupted work feel wrong. Sit down with one thing and nothing else, and within a few minutes something rises that feels like urgency — a thought that you should check, and it arrives with excellent reasons attached. That sensation is not information about your work. It is the conditioned response arriving on schedule, and you have been obeying it for so long that you have never once let it crest and pass.

CONCLUSION: STOP TRAINING, THEN RETRAIN

Take the range of this seriously. These are observational studies of information workers in a handful of technology organisations, with modest samples and effects of modest size — the conditioning result explains a small share of the variance, and it is one finding, not a law. What the work establishes solidly is the shape: work is heavily fragmented, a meaningful portion of the fragmentation is self-inflicted, and the cost lands on the person rather than the output.

Work with that.

Fix the room before you fix your resolve. The largest environmental effect measured here is architectural — 64 percent, for a wall. (Individual differences explained more of the variance than the environment did, so a wall is not destiny; it is just the one variable you can actually change this week.) If you have any say over where you do your hardest work, spend it on enclosure and spend it before you spend anything on discipline. A door outperforms a system.

Cut the incoming rate on purpose, because it is the training signal. Not for your own comfort — to stop the conditioning. Every external interruption in this hour raises your self-interruption rate in the next one. Notifications off, doors shut, hours blocked. You are not protecting the ninety minutes. You are protecting the following ninety minutes, which is where the damage actually lands.

Then count your self-interruptions, for one day. Every time you leave a task with nobody prompting it, make a mark. Do not try to stop, just count. You will be shocked, and the shock is doing the work — an unmeasured reflex is invisible, and a counted one is not.

Learn to let the urge crest. When the pull to check arrives mid-task, do nothing for sixty seconds. Do not resist heroically and do not comply; simply do not act, and observe what happens to it. In most cases it collapses, and it collapses without leaving anything behind, which tells you what it was.

And finish things, or park them properly. Sixty percent of healthy task switches in the data came from actually completing something. When you must stop mid-stream, spend thirty seconds writing down exactly where you were and what comes next — you will pass through two other contexts before you return, and the note is the only thing that survives that trip.

One caution about what you will find on the other side. Get good at this and you will start encountering stretches with nothing in them — and that turns out to be its own problem, with its own literature, and a room where people gave themselves electric shocks rather than sit in it.

But that is the next thing.

First, notice that you were never lazy.

You were taught to leave, by a decade of people leaving you no choice, and you learned it so thoroughly that you now do it to yourself in an empty room with the door closed.

SOURCES

Primary: Laura Dabbish, Gloria Mark and Victor M. González, "Why do I keep interrupting myself? Environment, habit and self-interruption," Proceedings of CHI 2011 — 889 hours of observed task-switching behaviour from 36 individuals across three high-technology organisations; of 5,089 observed task switches, 3,059 (60%) from natural event completion, 1,141 (22%) from external interruption and 889 (18%) from self-interruption; the coding definition of self-interruption; the finding that external interruptions experienced in the previous hour significantly increase the incidence of self-interruption in the subsequent hour, and the authors' interpretation that people may be conditioned or habituated to self-interrupt; external interruption accounting for approximately 4 percent of the variance in self-interruption against roughly 14 percent for individual differences, with job role showing no association (F(4,80) = 0.91, p = 0.46); the 64 percent higher self-interruption rate in open cubicles versus enclosed offices (M = 1.13, SD = 1.35 versus M = 0.69, SD = 1.13); and longer resumption times for self-interrupted than externally interrupted tasks.

Primary: Gloria Mark, Victor M. González and Justin Harris, "No Task Left Behind? Examining the Nature of Fragmented Work," Proceedings of CHI 2005 — 24 informants observed for an average of 25 hours 42 minutes each, over 700 formal hours total; an average of 11.7 (SD = 2.4) working spheres per informant; approximately 11 minutes in a working sphere before switching; 57 percent of working spheres interrupted; the correlation between longer engagement and longer interrupting events (r = .132, N = 1281, p < .001); same-day resumption taking an average of 25 minutes 26 seconds (SD = 54 minutes 48 seconds); and an average of 2.26 (SD = 2.79) intervening working spheres before resumption.

Primary: Gloria Mark, Daniela Gudith and Ulrich Klocke, "The Cost of Interrupted Work: More Speed and Stress," Proceedings of CHI 2008 — interrupted tasks completed in less time with no difference in quality, at the cost of significantly higher stress, frustration, time pressure and effort.

Correction of a common claim: the widely repeated figure that an interruption costs "23 minutes and 15 seconds" does not appear in any of the papers to which it is routinely attributed and appears to originate in a 2006 interview rather than a published result. The sourced figure is 25 minutes 26 seconds for same-day resumption, with a standard deviation larger than the mean — which is why the text presents it as a distribution rather than a per-interruption tax.

These are observational field studies with modest samples in technology-sector information work, plus one controlled experiment. Effect sizes for the conditioning result are small. The text states both limits.

RELATED

• Attention Residue
• Just Think — forthcoming, week 4
• Enterprise AI Finally Arrived