Mediated instruction (multimedia, IVD, ICAI, whatever) has tremendous promise. But I just wrote a chapter on Computer Based Instruction for the Handbook of Human Factors and Ergonomics in which I concluded that we can now deliver bad instruction better and in a more interesting way than ever before. It is like loading up Clark's trucks with bullshit. In fact, it is exactly like that.
I have to plead with clients to forego the three camera shoots so they can afford instructional development. My pleas don't always work, either. I love computers, multimedia, CD-ROM, the web, the whole ball of wax--but sometimes it is like being trapped inside a room with 100,000,000 view graphs.
Good instruction has to have a goal (aka, a learning objective) expressed in measurable terms. If you can't measure it, then you don't know if learning happened. (I know, you can ask the student, but he or she is a very unreliable source.)
Then, a good program will be powerful enough to actually engage the student and adapt the instruction to the individual's needs. We are finally getting systems that can actually do this, by the way, but it's hard work.
Read the series of articles in ACM Journal from a couple of months ago (Don Norman was the special editor) on learner centered design. Some of the articles were pure garbage, but others are on to something. There is a description in there of a strategy called, "scaffolding," which resonated with my 25 years of experience in this business. On the other hand, none of the ten or twelve articles presented any (I do mean any) data. Sigh.
Good discussion. By the way, I work with real truck drivers and the good ones do get the lettuce to the stores while it is still fresh.