Systems aren't adaptable
Next: Rule-based systems and some themes in AI
Up: What's wrong with expert systems?
Previous: No learning
Back: to main list of student notes
Novices and experts don't need the same type of explanation. Much
research has gone into ``adaptable systems'' which adapt their
explanation to the user's skill. This incidentally can apply to programs
other than expert systems, and I'll assume it does in the rest of this
section. For example, operating systems or databases whose error
messages adapt to the user's knowledge.
Building such systems is tremendously difficult. Firstly, you need to
find out how capable the user is. That's not easy when the only channel
is his keyboard responses. Compare with with all the information a human
expert can acquire about his client.
Secondly, you need to know something about the user's intentions and
goals. For example, suppose you have a computer filing system, and the
user asks to delete a file that doesn't exist. Possible reasons for his
action:
- He has made a genuine typing error: e.g. typed 1 for I.
- He has forgotten what the file was called.
- He knows what the file is called; to save typing he tried to
abbreviate its name, but did so wrongly (e.g. assuming he could leave
the .DOC bit off a file with that extension).
- He's got the file name itself right, but the directory wrong.
- He thought he'd created the file with an earlier command, but he
used that command incorrectly and it failed to create the file. E.g. a
botched edit.
- He thought he'd created the file with an earlier command, but he
used that command incorrectly and it gave the file a name different from
what he expected.
When reporting the error, should the system
- Say 'Error'?
- Give a ten-line essay on the exact syntax of filenames?
- Display a list of all the files he owns?
- Keep a log all his previous commands, and search for any that went
wrong, then try to work out if any of these failed to create a file he
intended to create?
Again, there's been a lot of work on inferring users' goals, plans, and
intentions, but it hasn't resulted in much of any use.
Next: Rule-based systems and some themes in AI
Up: What's wrong with expert systems?
Previous: No learning
Back: to main list of student notes
Jocelyn Ireson-Paine
Wed Feb 14 23:39:25 GMT 1996