Another solution would be simply to separate the two functions completely: Alt+Tab for switching in a spatial way, and a separate hotkey, say F4, for switching between the two most recently-used applications.

I wonder if HRMRU would work in a tabbed-browsing environment. Tabs are always visible, and HRMRU would require tabs to move around, and that can feel unpredictable. Also, in my own browsing habits, moving to the next tab on the right is far more common than switching back and forth between two tabs.

I also wonder if people’s productivity would increase if Alt+Tab weren’t as unpredictable (and annoying), but actually preserved the visual order in the taskbar or dock. If it did, then my other concerns about tabbed-browsing might also apply.

In the end, it’s possible that HRMRU would work as a unified solution, but it also may turn out that the best solution would be to separate the two functions, as I described. Either way, F4 for switching between the two most recently-used tabs in Firefox would be quite nice.

Indeed, the alt-tab behavior is unpredictable. When I got a Mac about 2 years ago, one of the things I really liked, and replaced my alt-tab habit/struggle almost completely, was Exposé: it is quite easy to switch to another window. The minor thing about it is that it does require me to grab my mouse; it does have support for keyboard-only operation, but I don’t find it very practical.

For the last half year I am again using Windows 7 a lot, and missing the expose dearly. However, I’ve recently discovered that in Windows 7 the programs in the task bar can be accessed by pressing -1, , etc.

Reading your post made me realize that this feature can perhaps give me predictable sort-of alt-tab behavior. If you pin your most-used programs to the taskbar, you can use this key combination to quickly and predictable open the required program. It’ll just require some muscle memory training…

Hi Aza,

KDE is trying to solve a very similar problem. They want to group windows (or rather any interface element), that belong together into so called activities. The idea is, to eventually use “semantic technologies” to do the grouping automatically. When the same document is open in your editor and your bowser, for example, your editor and browser windows would be combined into one activity.

As with most KDE ideas however, they sound great on paper, but the team is notoriously bad at living up to their potential.

The related search terms would be “plasma activities” and “nepomuk”.

michael

One more thing (as Steve would say ;) )

I wonder whether that graphical interface for switching apps in Windows was more because of unpredictability or because of the poor performance that computers had at that time.

It doesn’t make a lot of sense to me – in most cases, I would be much more happier if there was no such interface at all. it just slows down interaction in most cases…

Why don’t you just use spaces? Shesh. Spaces + keyboard shortcuts = exact context switching.

When I press ctrl-3, that switches to firefox. When I press ctrl-1, that switches to terminal. ctrl-2 is vim. ctrl-4 is ichat.

ALT+TAB+MOUSE ON OS X WORKS PRETTY WELL

I think the keyboard-based task (tab) switching is almost unsolvable in an ideal way, because there’s such a gulf between the ways that different people have learned to use it. Perfect “magic” for one user may be anathema to the next.

However, if I could have things all my own way, this is how I would implement keyboard-based switching.

Alt-Tab brings up a vertical menu of running tasks (or tabs) that doesn’t go away when you release the alt key. Each task has a number or letter next to it from 1-9 – tasks after that have no number. The MRU or your magic guessing heuristic orders them, but I can type letters in a filter box (to which my cursor is automatically moved) to reduce the number of visible tasks. So if I haven’t used firefox in a while (it’s off my list of 1-9), my keystrokes are alt-tab+fir+1. It seems like the new gnome-shell is pretty close to this type of interface with the Activities menu (it would be Windows key instead of alt-tab) – I’m looking into writing an extension for it. I don’t know why they don’t have it in there by default.

Of course, if you didn’t want to use filtering or number-switching in that menu, you could continue pressing tab until you got to the task you wanted and then press enter. Or you could use your mouse to click on the task you want to go to.

A completely separate idea for mouse-based task switching is a menu that groups task icons with some heuristic around the mouse in the center of the screen. Maybe it’s horizontal, maybe vertical (my personal preference so you can include text), or maybe it is circular. Let’s say it’s vertical – the current task is in the center of the screen and the mouse is located over it. If I press alt-tab and click immediately I’m right back in whatever application I was just using. Alt-tab again. Above and below the current task are the two MRU apps, and above and below them are the 3rd and 4th MRU apps. So my time to mouse to the 2nd,4th, etc MRU app is half of what it would be if they were ordered linearly.

You can extend the idea to a circular presentation pretty easily. Alt-tab and the mouse is in the center of the screen on nothing, and three icons are positioned in a circle around the mouse – the three MRU apps. The next five MRU are in the next larger circle, etc.

For tab switching in firefox, anything that is icon-based for switching seems problematic, because we don’t associated icons with websites very often. I would prefer my “list with filter” idea so I could type “alt-tab+aza+1″ or maybe “alt-tab+aza+[enter]” to get to this tab in a hurry.

+1 for Nate’s type-to-filter-a-list-of-windows idea. Specific applications can go into muscle memory, and it’s scalable and constant.

My friend used the same idea to read timestamps from mozilla browser history and predict the next page based on hidden variables like time of day current page previous page etc…

Cool article but I think there is a bigger elephant in the room.

How does a user discover the alt-tab mechanism?

Alt-tab has to be the single least discoverable keystroke in personal computing.

Christ-come-to-Earth would need his mate with a BSc. in Computer Science to tell him about it.

Alternative/Multiple inputs

Something not to ignore is the fact that the design can include more advanced means of selecting. For example in OS X you can use the mouse to select an application after holding down the command key. Automatically assigning keys to applications, searching, and other means could all be implemented to provide a rich interface to switch.

A simple implementation could be to let the user decide which applications are “paired”. I could for example select to pair Firefox with TextMate and as soon as I’d do this I would take the habit that Firefox TextMate switching = 1 alt-tab. Firefox or TextMate -> Tweetie = 2 alt-tab and finally Tweetie -> Firefox or TextMate = 1 alt-tab.

Even better I could select which one of the pair is the first one to appear. So for example, I set Firefox as the “primary” of that pair. Then, anytime I alt-tab to an application other than those 2, I’d take the habit that switching back to Firefox is 1 alt-tab and to TextMate is 2 alt-tab because never would that pair be in another order when another application is open.

I’ve had a similar thought wrt markov modeling app switching. I hypothesize that the same reasoning could be applied to the osx Dock and windows task bars, so as to resize the representation on an app based on the predicted likelyhood of wanting to switch to it.

Nice article – thx

You can stick with the idea of a stack of papers and just try to order them slightly more intelligently, or you can go with a different idea.

Personally, I like a cube. Open your browser on one face, switch to an adjacent face and open your text editor. Switch back and forth all you want. When you want to do something else, open it on a third face.

The nice thing about this approach is it leverages the human mind’s ability to think in 3D. As you spin the cube you can intuitively know where your applications are instead of losing them somewhere below the top two pages. Find just 4 sides too restrictive? Add more. Oh, and you can also put an Alt-Tab stack on each side of the cube, so you could alternatively group your browser and text editor onto one side of your cube.

Okay okay… I’ll come clean. This is not an original idea. Actually, it’s already implemented and something I use everyday. Just flip on the cube in desktop settings of KDE4. Done. No hacking or custom software needed. Next problem?

Great post… but I’m with Karl G, above, on preferring something very predictable over something that tries to second-guess me. For example, on Windows, I always make sure that the first window in my taskbar is Firefox, the second window is Thunderbird (similarly, I keep my tray icons always in the same order) — using Taskbar Shuffle. I do the same thing with Firefox tabs: first one is GMail, second GCalendar, third one work intranet, fourth one work Trac. I disabled the Firefox pref that tries to insert tabs somewhere other than at the end, because I want it to be predictable. When I have multiple terminal windows open, I’ll order their taskbar buttons to match their location at the screen.

All that is to say, let me build muscle memory by having things in a fixed order/location, and don’t try to fuzz with that.

“As people browse the web, we know that an average user will switch tabs more times in a day than they click on a link”

We know? I don’t remember someone who does this is in the first few seconds I think about it, people I know browse the web (and use tabs less than I do).

To both completely avoid the problem and enjoy a great user interface, use virtual desktops with a direct shortcut for each desktop, and organize your work with one task per desktop.

This greatly reduces the number of windows on each desktop, typically to just 1 or 2 (and sometimes even 3!)

Now if alt-tab is only used to switch between 2 or 3 windows (and on some desktop they will already all be visible at the same time, and if you are using a mouse on those desktop you won’t even want to alt-tab), you don’t really have to worry about the order in which they are presented.

The problem doesn’t just stop at switching between apps. In a document centric environment, I find it really annoying that I have to decide whether switching between two docs involves switching between two different apps or not!

For example, I’m looking at a web page and writing about it. Text Editor and Firefox are my two apps – is use cmd+tab to switch between them. Now I open a text document and try to refer to that while writing my own stuff on it – cmd+tab _no longer_ works becuase my ‘reference document’ is in the same application. Why do I have to pay attention to this detail!?

Oddly, this has never been an issue for me. All of my opened apps, docs, browsed pages, etc., are all right there on my task bar. At the height of my productive day I may have 15-20 open apps down there. However, still, not a problem. Using the mouse, I swiftly move my hand, positioning the cursor on the correct tab, and click. BOOM. Done. Problem solved. WTH is wrong with you people? rofl. This is not rocket science, nor does it need to be made into it.

To me alt tabing will go away completely in a short few years. Windows 7 provides the method / logic which will be used moving forward. Where your system bar provides tab drilling on a visual level. For example I mouse over the firefox tab and poof a set of visual based mini views opens up showing my tabs and their current content I click the one I want and am instantly there. With alt+tab the best I can do is get to an application. With the drill down effects you remove that added level. Same for Instant messengers. I have 1 IM tab on my system, but 10 windows, when I get a new IM I can mouse directly to it faster than I can alt tab, cause my mouse is usually in my hand. Since I have a navagation tree I can follow to drill down into the application I get more benefit than alt tabbing.

That is the other key part the system bar can / is based on the programmers of the application. Application aware, meaning if your downloading it can show the progress in system bar. Without the need to alt tab to begin with.

so really even dwelling on this subject starts to become a mute point. Not that it doesn’t have merrit thought is always worth the effort. Trying to derive a solution though should require more thought than an algorithm cause humans don’t work the same, we are always the X factor to any equation. You can only look to statisfy X to a level of Y percentage.

I’m impressed that you noticed this but more that you put it in the category of Habits. I never use Alt+Tab but go to the task bar to select. It does help, of course, that I use two (19″ LCD) monitors — that really cuts down on Alt-Tab and task bar clicking.

My beef is with the number of “click” it takes to do anything, and worst of all having to select paths for open/save operations. Few applications can record their last path.

Nice article, though.

To keep the user focused, it is best to scramble the icons each time you cmd-tabbed. This way, the user will never form a habit and thus there is no problem anymore. Also, this is much easier than trying to detect habits :)

Good idea, it would make sense to test the adaptive app switching in reality.

Personally, and considering the “spatial memory” you mentioned, I like to have fixed keys for most used applications. I’m used to press ALT-F2 to open the “run command/krunner” (in KDE) or Launchy (Windows) and would like to have the same for other applications.

Just configure ALT-1 for the browser, ALT-2 for E-Mail client, ALT-3 for Twitter, ALT-4 for text processor… Very easy to use and remember after few minutes. It’s limited to ~5-9 applications, but this is probably than enough.

For browser tabs, it shouldn’t “just be tabs”, but your browser should be aware of the application running in the tab. It should always open CNN.com on tab1, google reader in tab 2, my webmail client in tab 3…

Maybe the concept shouldn’t be limited to “applications”, but group several applications. KDE has a similar concept with activities [1], not sure how good it will work with 4.5, but I hope to be able to easily switch between activities with my keyboard.

[1] http://chani.wordpress.com/2009/11/28/the-future-of-activities/

Ok, but where is the app?

This is interesting and a problem I have all the time and never really gave much thought to until now. What’s weird for me is that most of the time I don’t do a CMD+Shift+Tab to go backwards through the list and keep right on CMD+Tabing until I get the one I want.

i REALLY like the idea of my text editor and browser being grouped together. i switch to photoshop a lot when building websites though so i think those 3 could end up creating the same problem in the group.

I made CMD+/ a shortcut for tweetie. hitting it pulls it to the front of the circulation where i can skim through the tweets, hit CMD+K to mark all as read and then hit CMD+/ again to send it to the back of the line leaving my two desired applications still at the front.

this solution helped me though it’s just one application. might help some other people. it actually helps use tweetie too by making the process of checking it much faster. i dont have to find it in my stack. its always at the back.

There’s a simple recasting of Asa’s suggestion that makes things a lot clearer. (There other issues that I’ll discuss after the simple description.)

Rule: Make the order of the list reflect the historical frequency of transition from the current app, not anything about immediate history.

Using Asa’s example, when Firefox is current, the most frequent transition is to TextMate so it is second. When TextMate is current the most frequent transition is to Firefox so that is second. When Tweetie is current, we don’t know, but given that Asa is usually in Firefox or Textmate, probably they are second and third, or at least near the front of the list.

Specifically the pair-wise back and forth is solved, for any number of disjoint pairs. For frequent transitions among more than two apps, at least the apps will cluster near the front of the list once you go to one of them, even if you don’t enter that cluster very often.

Furthermore if you have a habitual sequence between various apps, that sequence will be captured in the list you see from each one. So the system will “learn” your common work flow habits.

For any app, the sequence you see when you hit cmd-Tab is likely to be fairly stable, since it reflects historical frequency of transition from that app, not recent events. So for example someone briefly using your machine wouldn’t nuke your habits. However if a machine is shared between multiple people they probably need separate accounts / profiles.

The transition history should decay over time — maybe using the frecency algorithm from Firefox.

Note that the rule is very simple: Keep track of where you go when you leave an app, and order the list by the frequency of going there from that app.

Technically this implements a Markov net, with the transition probabilities derived from switching history — just as Asa suggested. But it is important to realize that is a very simple, comprehensible rule.

I think this addresses a lot of the comments — it solves Asa’s problem (as he conjectured), it leads to fairly stable lists, it is simple enough to be learnable (even without any explanation). We have seen with the various history driven completion algorithms that people are comfortable with this kind of learning.

I would guess that the list order being specific to the app would be very compatible with habits — habits are always context specific.

Issues that people brought up that aren’t addressed: windows (and browser tabs), and other ways of switching.

The switching history that determines list order should probably include all ways of switching to another app, not just tabbing to it.

I’d very much like some mechanism for switching between windows and tabs. However these are bigger collections — in my case, my browser tabs and sometimes my open mail are much bigger collections. Also because windows and tabs aren’t nearly as persistent as apps, and my habits of accessing them aren’t nearly as stable, the long term frequency wouldn’t work.

I can imagine some mechanism that would provide access to an even larger collection, such as my frequently accessed web addresses, recent mail, recently accessed files, etc. But obviously navigation is a huge problem. Note however that this is a problem I already have to solve, using multiple un-coordinated modes (Finder, Mail, browser (multiple sub-modes), etc. Surely there’s a way to do better at the system level…

Filtering by typing is interesting for these big collections. It only works when the name or properties of the thing you’re trying to find have some substring you know, but maybe we’d find we know “names” for the things we want to access. We know this sort of filtering works for large collections (e.g. browser history, Google queries). It is powerful because we exponentially narrow the search with every character we type, and we can get immediate feedback.

I’d like to experiment with window switching where I get to type the name and see a list of windows I can arrow through. Maybe there are task bars that do this but I don’t currently use them.

Dude, drop your floating window metaphor and move to tiling windows… like xmonad for example. Like seriously, you guys will never learn, but the solution has been around for years. You could even make a tiling windowing system “user friendly” (as in: not only for power users, cause my tiling window linux is 10x as user friendly as OSX or win7… alt tab being one of the reasons).

Aza, if you really want to simplify and really want to push interfaces forward you already lost when you consider traditional floating window managers…

First off, I disagree somewhat that the thinking in alt-tabbing could be applied to Firefox; the problem domains are very different. In Firefox, your tabs are constantly visible, so there’s never a risk of switching to a tab you don’t expect. User reordering makes perfect sense when the tabs are visible, and it makes alternating between tabs very intuitive–you drag the tabs you want next to each other, and you alternate between the Next and Previous Tab keyboard shortcuts to alternate between two tabs. I’m not saying that Firefox’s tab system couldn’t be improved, but Firefox doesn’t HAVE the Alt-Tab problem.

That said, here’s the most elegant solution I could come up with without introducing too much complexity.

Right now, the order of apps or windows is MRU. When you first call up the switcher, the current app is on the left, and the second app from the left is selected, which means that you can simply release the keyboard shortcut to switch to it.

In my scheme, apps would be displayed in some fixed order. Launch order makes a lot of sense, but any fixed order is just as spatially beneficial. Alphabetical order could be neat in that the spatial order would be the same between app launches and sessions.

When you call up the switcher, the selected app wouldn’t be the second app from the left, or the app immediately to the right of the frontmost app; it would be the most recently used app. This would allow you to switch between two apps really easily.

This still seems conducive to the three-apps problem, but there’s a key distinction: it’s more spatially intuitive which app you’re selecting. You know where each app is, so when you do the Alt-Tab dance, you don’t need to look at the app icons or the window preview to know which app you’re going to; you can look at the positions of the apps, which is much faster. If you switch to the wrong app, you can tell while you’re still in the switcher, not right after you dismiss the switcher and the wrong app pops up.

This solution introduces more problems than it abolishes, though, and the current scheme used by OS X and Windows is better. I like Tijn’s solution. :)

Here is my green + user friendly solution: http://erikvold.com/blog/index.cfm/2010/5/8/my-solution-to-the-alttab-problem

Worth a try! You need an undo action to get back the previous window. And why not a redo action.

Still, this may be less efficient (although more consistent) than MRU_n + OO: MRU for the first n windows, followed by order of opening. I believe this is what Vi$ta uses, but I remember being annoyed that it started iterating in order of opening after only the second window. I use MRU bwtween 4 windows effectively.

In this post Aza is looking for a better way to do most-recently-used (MRU) window switching. He says the way it currently works can be confusing.

I don’t disagree. But I think there are more important problems with Alt-Tab. For example, the increased use of tabbed windows has made the shortcut much less useful than it used to be. And I find it frustrating that the master shortcut key (Alt/Cmd-Tab) uses MRU order, but the supporting ones (Ctrl-Tab and Cmd-~) don’t.

I’ve written a blog post describing these problems in more detail, and proposing how to solve them:
http://blog.isotoma.com/2010/05/solving-the-real-alt-tab-problem/

Good problem-picking! With the adoption of tab-based interfaces across many applications, it is in dire need of good interaction design to get the screen that you want.

Btw, here are some wild ideas that I was thinking of as I was reading your article:
* A gaze based approach – I generally think of gaze interface work as somewhat unintuitive as there is no explicit action in the world from the user that causes the system to react. But a large number of users use multiple monitors with huge screens. Based on the gaze of the user, the system could move up the applications in that window up the hierarchy, ensuring that those windows are hit first. (As an aside, certain interface elements need to be replicated when on a multiple monitor usage. For eg., the cmd-tab bar itself. Design for multiple monitors need much improvement)
* An update based approach – The probability of navigating to an application increases when there is an event that has occurred that needs user attention. This could be anything from a new IM message to a modal window. On such events, the priority of those windows could be increased significantly if it is learnt that the user usually heeds to those updates.
* Grouping applications – Textmate, Console, Browser, Sequel Pro… if the user is cycling within a group of applications constantly, a model can be developed to push these apps together. When another application comes to the foreground, its own group gets proiritized higher..

This idea of yours is what I’ve been dreaming about for all my life. Sure, it needs some testing… And there’s a big chance users won’t be able to figure out the algorhythm… Now at least they know what to do to ‘teach’ the system how to switch between the two windows and know where the switcher will go if they press alt+tab several times to reach some 5th window.

P.S. You should link this to your ‘You Can’t Multitask’ post.

Thanks for the info, Keep up the good work

I didn’t read all the posts, so this might have been discussed already, but Linux solved my alt tab problem LONG AGO… the multiple workspace solution rocks… i mean i keep like 16 workspaces in a 4 by 4 grid… and keep at most 3 applications on each… and habit makes you generally keep certain applications on certain workspaces… so for example the browser is in workspace (1,1) (row, column), music player in (1,3), messengers and such in (1,2), stuff i work on in (2,1) and (2,2) (i rarely work on more)… and I didn’t spend time to sort my apps like this… they just felt natural / organized themselves… + the workspace switcher shows previews of the apps when i change workspaces.

Alt tab really fails compared to this… to bad windows doesn’t implement this concept.

First of all, great post – it phrases a frustration I hold, and was never able to point out.
Recently I began using a dual-monitor setup, and the frustration just grew bigger:
With a single monitor there’s a visual queue to the fact that alt-tab was performed: the other window comes forth. It’s so obvious, that I can even look off-screen and still my eyes will pick up the change.
But with two monitors, two windows are allowed to be on top – one in each screen. If you’re switching between them, then the visual queue – the title of the active window has a different color – is negligible.
Since I group related windows on each screen, the suggested solution is exactly what I’d expect the OS to do.
Again, great job on both analyzing the problem and finding how to solve it.

Again, great

Thanks for this great idea!
There is one thing that bothers me even more than the problem you describe above (even though you are right it is very annoying): I dont want to have to think about whether the window I want to go to is a program or a window (or tab) in a program. I started using http://manytricks.com/witch/ to solve this which is great. But it has other issues (performance mainly).
How could this be solved?

Hi,

I agree that it was a pain to use the Alt-Tab but eventually I now get it to do what I want it to do.

I currently use Win 7 and think the same applies to Win XP, The problem of selecting Alt-Tab then changing your mind once you have proceeded with the tab option can be overcome by transferring from Alt-Tab To Alt-Esc but you must not leave go of the Alt key.

Give it a whirl and I think you may change your mind and start using Alt-Tab.

Cheers,

John_Betong