A Taxonomy of Cognitive Stress:

I have been thinking about UI design lately. With some help from my friend Rob Landley, I've come up with a classification schema for the levels at which users are willing to invest effort to build competence.

The base assumption is that for any given user there is a maximum cognitive load any given user is willing to accept to use an interface. I think that there are levels, analogous to Piagetian developmental thresholds and possibly related to them, in the trajectory of learning to use software interfaces.

Level 0: I'll only push one button.

Level 1: I'll push a sequence of buttons, as long as they're all visible and I don't have to remember anything between presses. These people can do checklists.

Level 2: I'm willing to push as sequence of buttons in which later ones may not be visible until earlier ones have been pressed. These people will follow pull-down menus; it's OK for the display to change as long as they can memorize the steps.

Level 3: I'm willing to use folders if they never change while I'm not looking. There can be hidden unchanging state, but nothing must ever happen out of sight. These people can handle an incremental replace with confirmation. They can use macros, but have no capability to cope with surprises other than by yelling for help.

Level 4: I'm willing to use metaphors to describe magic actions. A folder can be described by "These are all my local machines" or "these are all my print jobs" and is allowed to change out of sight in an unsurprising way. These people can handle global replace, but must examine the result to maintain confidence. These people will begin customizing their environment.

Level 5: I'm willing to use categories (generalize about nouns). I'm willing to recognize that all .doc files are alike, or all .jpg files are alike, and I have confidence there are sets of actions I can apply to a file I have never seen that will work because I know its type. (Late in this level knowledge begins to become articulate; these people are willing to give simple instructions over the phone or by email.)

Level 6: I'm willing to unpack metaphors into procedural steps. People at this level begin to be able to cope with surprises when the metaphor breaks, because they have a representation of process. People at this level are ready to cope with the fact that HTML documents are made up of tags, and more generally with simple document markup.

Level 7: I'm willing to move between different representations of a document or piece of data. People at this level know that any one view of the data is not the same as the data, and lossless transformations no longer scare them. Multiple representations become more useful than confusing. At this level the idea of structural rather than presentation markup begins to make sense.

Level 8: I'm willing to package simple procedures I already understand. These people are willing to record a sequence of actions which they understand into a macro, as long as no decisions (conditionals) are involved. They begin to get comfortable with report generators. At advanced level 8 they may start to be willing to deal with simple SQL.

Level 9: I am willing to package procedures that make decisions, as long as I already understand them. At his level, people begin to cope with conditionals and loops, and also to deal with the idea of programming languages.

Level 10: I am willing to problem-solve at the procedural level, writing programs for tasks I don't completely understand before developing them.

I'm thinking this scale might be useful in classifying interfaces and developing guidelines for not exceeding the pain threshold of an audience if we have some model of what their notion of acceptable cognitive load is.

(This is a spinoff from my book-in-progress, "The Art of Unix Programming", but I don't plan to put it in the book.)

Comments, reactions, and refinements welcome.

Mail troubles:

Note to friends: mail to esr@thyrsus.com may be unreliable until mid-Thursday. I had DNS problems. They're now fixed, but there is bad data in a bunch of caches out there.

For the next couple of days, mail to esr@catb.org instead.