Two Monitors, One Mind

For about fifteen years I believed that adding a monitor added a channel. The logic felt tidy. One screen holds one thing, two screens hold two things, I have two eyes and a wide field of view, so surely I can watch both. I built desks around it. At one point I had three monitors and a television, and I felt extremely sophisticated about it.

There is a television running on my left monitor as I write this. I know roughly what it costs me. I haven't turned it off.

The question turned out not to be how many screens a person can watch. It's which pairs of things fight each other, how badly, and whether that matters for whatever I happen to be doing that evening.

The rule I work from, before any of the evidence that shaped it: two activities interfere roughly in proportion to the machinery they both want. Two streams of words fight hard. Words and a diagram barely fight at all. What follows is me testing that against what the research does and doesn't support, and marking the places where I've gone further than it does.

Start with the narrowest well-replicated piece. When two simple tasks each require choosing a response, and one arrives shortly after the other, the later choice is delayed, and the delay grows the closer together they come. Pashler's review gathers a great deal of evidence for a bottleneck at the response-selection stage in these paradigms (Pashler, 1994). The single door people take from that is a metaphor for a proposed bottleneck under laboratory conditions, and I've watched it carried a long way past anything the experiments cover.

It also doesn't explain why I can hold a conversation while walking and not while reading. One influential model treats concurrent tasks as threads contending for shared resources, and predicts more interference the more two of them demand the same ones (Salvucci & Taatgen, 2008). It's a computational theory, and its resources are constructs rather than anything measured in a head. As a way of deciding what belongs on a second screen it's still the most useful thing I've found, which is why the question I ask now is:

How much do these two things want the same machinery?

The piece that reorganised my desk is fifty years old. Working memory looks less like a single store than a set of partially separable components, one dealing in speech-based information and another in the visual and spatial kind (Baddeley & Hitch, 1974). The separation isn't clean, the model has been revised many times since, and the component now called the phonological loop went by a different name in the original paper.

Speech you aren't listening to still costs you on verbal tasks. In the classic demonstration, unattended speech disrupted short-term recall of visually presented digits, including when it was meaningless and participants had been told to ignore it (Salamé & Baddeley, 1982). That's serial recall in a lab, not reading. The extension is mine: when I work with a show playing I lose something, and this is the best explanation I have for it.

I can write code with instrumental music going and lose nothing I'm able to detect. Put dialogue on the second monitor and my output drops off, even though I'm not watching it, and I couldn't tell you afterward what happened in the episode.

Thirty participants once read comprehension passages under four sound conditions: quiet, instrumental music, lyrics they liked, and thrash metal they hated. Both vocal conditions hurt comprehension, and did so about equally (Perham & Currie, 2014). I fought that for a month, ran it on myself, and lost. The half I keep forgetting to mention is the encouraging one. Instrumental music came out indistinguishable from silence.

There's a refinement. Changing-state sound, the kind whose character keeps shifting, impaired serial recall about as much as speech did, which points at acoustic variation rather than at meaning (Jones & Macken, 1993). Serial recall again. What I've built on top of it is a playlist policy rather than a finding: a repetitive loop bothers me less than dense, surprising instrumental music, and shuffle is worse than an album, which I resent.

So what actually belongs on the second screen. The combinations that work for me hold different kinds of material. Code on the left and documentation on the right. Prose on one screen and the diagram it describes alongside. A video call in front of me and my notes beside it. In each case the second display holds reference rather than competing for the same words.

What fails is two streams of language. Code beside a chat window. Writing beside an open inbox. Anything beside dialogue. Prose set next to more of it is one channel cut in half, and a second display doesn't repair that, because the hardware was never the constraint. The show running on my left monitor right now is dialogue. That is the exact case I have just called hopeless. I know. We will come back to that.

Logs are the interesting case, and it depends on what I'm asking of them. Letting output scroll past while I work costs me very little. Reading it, and deciding from it, is a second job wearing the same clothes. A dashboard splits the same way: glanceable is one thing, and a panel I'm actively making calls from is another.

There's a wrinkle running the other way, and it's why I distrust tidy rules here. Everybody knows the traffic version. You get lost, and you turn the radio down to read the street signs. Nobody taught anybody that, and the intuition arrived long before the theory did.

This is load theory's territory. High perceptual load can reduce distraction, because the senses have little left to spare, while pressure on cognitive control can increase it (Lavie, 2005). The theory is influential and still argued over, and the effects depend a good deal on the task.

Two results sit underneath it. In a driving simulator, hands-free conversation impaired performance as much as holding the phone did (Strayer & Johnston, 2001). And under high visual load, seventy-nine percent of participants failed to report a plainly audible tone (Macdonald & Lavie, 2011). That second one is a failure to notice, not a loss of hearing, and the difference is worth keeping.

Neither study tested the radio manoeuvre, and reaching for a stereo at the moment you're most loaded isn't something I'd recommend. Cut the distraction before it's urgent, or stop the car.

The change that did the most for me was a schedule. My phone makes no sound and doesn't vibrate for anything except an actual call, and I look at it about twice an hour. That's my routine, and my work permits it, which isn't true of on-call rotations, caregiving, or anyone whose colleagues need to reach them.

On a sustained-attention task in the lab, notifications participants never touched impaired performance by an amount comparable to what was seen when they took the call or the text (Stothart, Mitchum & Yehnert, 2015). Not answering bought them nothing measurable. That's a lab task rather than a working day, and it's the result that stopped me trusting my own restraint.

In a randomised field experiment with two hundred and thirty-seven people, batching notifications to three fixed times a day left participants more attentive, in a better mood and less stressed than ordinary push, while the arm that received none at all gained little and reported higher anxiety and more fear of missing out (Fitz et al., 2019). One study, one population. Three checks a day is also not the same intervention as my twice an hour, and I should stop implying that it is. What I take from it is narrower than batching wins: going completely dark carries a cost of its own, and predictability seems to be doing some of the work.

While I'm correcting things. The twenty-three-minute figure attached to interruptions doesn't come from where people think it does. Observing twenty-four information workers, the researchers found they spent about eleven minutes in a working sphere before switching, and that fifty-seven percent of those segments got interrupted. Going back to one took about twenty-five and a half minutes, with more than two other activities occurring in between (Mark, González & Harris, 2005). That interval isn't twenty-five minutes of recovery. It's elapsed time with other work inside it. Where the twenty-three came from I don't know, and I'm not going to invent a story about it.

Attention residue is worst when a task is left unfinished under time pressure (Leroy, 2009). So I write a one-line note about where I was heading before I switch away. That's my technique, and Leroy's work doesn't test it.

One loosening. Practised tasks come to lean on automatic processing and demand less controlled attention than they did when they were new. Less is not none. Walking and talking coexist until the pavement turns to ice, and the same holds for anything I've done ten thousand times, because the cost returns the moment conditions get strange.

What I do, then. These are operating rules I've arrived at rather than results anyone demonstrated.

One verbal channel at a time. If the work is made of words, the background can't be, and liking the song doesn't appear to help.

Steady beats shifting, for me at least. Instrumental cost nothing in the one study that measured it against silence.

A second display earns its place when it holds reference rather than another argument. If both screens are making a case, I only have one screen.

Alerts can cost you even when you ignore them, so batch if your responsibilities allow it. Plenty of people's don't.

Leave a note at the seam before switching away. Ten seconds of it, naming where I was heading. Usually six words in a scratch file I never reopen, which turns out not to matter, because the writing is what does the work.

Notice whether it's your eyes or your judgement that's full. The two don't behave the same way, and I get it wrong often enough that I no longer trust the feeling on its own. This is the rule resting on the most contested theory here, and I use it anyway.

None of that tells me what to optimise for. Maximum output, sustainable comfort, staying reachable, keeping half an eye on things, or simply liking my evening. Those pull against each other, and no amount of reading settles which one I ought to want.

Which brings me back to the television. When the work is real, the second screen goes dark, the phone moves to another room, and the sound becomes something with no words in it. When the work is maintenance, I let the show run. I'm worse at both, and the evening is better, and I spent years feeling guilty about that instead of choosing it.