Agre@OZ.AI.MIT.EDU.UUCP (12/29/86)
I need volunteers for an experiment. I've spent the last few years
studying small details of everyday routine activity, hoping to use my
observations to constrain theories of cognitive architecture. In doing
so, I've found it useful to write down anecdotes of small episodes from
ordinary activities like making breakfast and driving to work. This
method has some amazing properties. Suppose you've been worrying over,
say, deciding to perform two steps of an activity in a different order
than you have in the past. Then over the next few days, on several
occasions when such a thing occurs you will notice it. No need to
deliberately look out for them (and presumably much better if you
don't). Whenever this happens to me I write it down. Habitually
writing these things down then makes you notice them a LOT more. This
has to be experienced to be believed.
I want to get a lot of people to do this and see what happens. To this
end, I am going to describe two topics I've been interested in and
present some example stories about them. Read these descriptions. Then
when you spontaneously notice an example of them in your own activity,
write a paragraph or two about it. Collect these stories and send them
to Survey@MIT-AI (that's AI.AI.MIT.EDU in fancy notation).
First topic: Small mistakes.
Several psychologists have collected lists of what are often called
"action slips", mistakes one makes in the course of ordinary activity.
In reading these lists, I am always concerned at how remarkable they
are: how interesting or funny or odd. So I'd like to collect examples
of absolutely trivial mistakes of all types, ones that you quickly
recovered from without swearing or pondering or breaking stride. (Doing
so tends to make you think about whether there's a clear difference
between a mistake and something you tried that simply didn't work out.)
Example: I'd tipped my chair backward to lean against a shoulder-high
shelf. I was drinking a cup of tea and reading a book. It was kind of
a pain keeping the cup steady, so I went to put it down. Glancing
about, I found noplace convenient to put it except the shelf. Since my
shoulder was against the shelf, I saw the only way to put the cup on it
was to extend my arm fully. So I did this. I didn't bother watching
where the cup was going, instead I looked back at the book. I extended,
raised, moved back, and lowered my arm, expecting to feel the cup
landing on the shelf. After lowering my arm quite a lot this didn't
happen, so I looked and saw the cup wasn't over the shelf. Watching
this time, I did it again right.
Example: I often take the subway to work. Normally, given a choice, I
get on the train around the middle because the nearest exit from my
station is near the middle of the platform. Except now they're
rebuilding the station and they've closed that exit, as I discovered
yesterday morning. Nonetheless, this morning I got on near the middle
as usual. In fact, I got on more toward the front because there were
free seats in the next car along. I left the station through the main
exit.
Second topic: Anticipatory actions in a cyclic activity.
This happens an awful lot but for some reason there's no word for it.
When you start an activity, you do it in the obvious straightforward
order, but then you start rearranging and parallelizing the steps,
seemingly automatically.
Example: I had a stack of records propped up against a box and I was
alphabetizing it according to the artist's name, forming another,
sorted, stack propped up next to it. I would take a record from the top
of the first stack with my left hand, find and hold open the right place
for it in the second stack with my right hand, place the record in its
space, let the stack close over it, and repeat the cycle. After a while
I found I was doing something different: whereas before my eyes stayed
on the new record until I had picked it up, now I would read the
artist's name as soon as I was done with the last record. Then as I
picked it up with my left hand, my eyes were already helping my right
hand find the right place in the second stack.
Example: I was trying to get a long C program to compile. I was working
on a Sun and had divided the screen between two Unix shell windows so I
wouldn't have to exit and reenter the editor to run the compiler. I'd
run the compiler and it'd get errors, e.g., "syntax error near { on line
173", so I'd go back to the editor window. The only way I knew to get
to line 173 was to go to the top of the buffer and go down 172 lines.
This got to be a cycle, fixing errors and recompiling. After a while, I
found that I would move the editor to the top line before the compiler
had even starting generating error messages. (Finally one time the
compiler completed without errors and half of me had to skid to a
confused halt, but this detail is too amusing to be legal.)
[Try vi command 173G to skip to line 173. And for examples of
trivial little errors, you can't beat switching between vi and
emacs. Don Norman and others have done extensive studies of such
little errors, including the little errors that kill pilots. -- KIL]
Rules.
1. The episodes you write about must happen to you, in the course of
some solitary activity. They must happen after you read this note.
You cannot be aware of having this note or any other AI-ish topic on
your mind when they happen. You must have no memory of having
remarked on that same thing before.
2. They must be utterly mundane. They cannot be markedly stereotypical,
funny, disastrous, or otherwise interesting. They cannot have
occasioned any confusion, amazement, or careful reasoning-through.
3. You must write them down on the same day they happen. Write them
down accurately, being careful not to make them more clear-cut or
to-the-point than they actually were, in plain unscientific English,
the way you'd retell them as a story.
4. Though your descriptions naturally have to include any information
necessary to understand what happened, they cannot include any
speculations about "what was going on in your head" that weren't
definitely part of your experience of the episode at the time. If
you're unsure about some detail, say so.
If this experiment works out well, we can keep doing it periodically.
I suppose I'll write a paper about the experiments someday. Send any
notes requesting a copy to Survey-Request@MIT-AI.
Phil Agre