What Instructional Designers can learn from IBM’s Watson

By now you have heard about IBM’s super computer, Watson is a modern day wizard answering questions that seemed impossible for a computer to answer only a short while ago.

Why? Well until the Waston, Jeopardy problem, one of the ways that it was thought computers could answer questions was to follow rules. So the goal was to create a branching tree with enough rules and if-then-else statements for the computer to look up and answer any question that it encountered. Well, this seems to work but the concept quickly breaks down. A good example is “i” before “e” except after “c”. Sure that rule works but not all the time.

For example the word “neighbor” or the word “deity” break the “i before e rule. However, with enough logic, those exceptions can be defined.

• The rule only applies to digraphs, so words like “deity” and “science” don’t count.
  
• The rule “i before e except after c” should be extended to include “except when said ‘ay’ as in ‘neighbor’ and ‘weigh'”.
  
• The rule only applies to digraphs that have the /i:/ (‘ee’) pronunciation, as in ‘piece’. (Note the conflict between this and the previous item.)
  
• The rule doesn’t apply to words that are recent imports from foreign languages, such as “gneiss”, “dreidel”, and “enceinte”.
  
• The rule doesn’t apply to the large number of plurals of words ending in “cy” (“fallacies”, “frequencies”, “vacancies”, … ) because in the UK – in traditional RP – “cies” is pronounced with the “i” of “pin”, even though it is pronounced with the “ee” of “feed” by most World-English speakers and by younger UK speakers.

Source for block quote.

So researchers thought for many years that all they had to do was create enough rules and the computers would “get it.” Turns out that doesn’t work.

For example, take the letter A. What rules would you define for a letter “A” so the computer could recognize it? Well maybe two lines at an angle with the third in the middle. That might be a good rule, however, what about an images like the ones below where the A in NOVA has no middle line or the old print doesn’t have the line.

And we haven’t even discussed lowercase A. So the rule method doesn’t work.

It also turns out that teaching humans a bunch of rules and expecting them to apply them perfectly every time doesn’t work either.

What is the alternative? Examples.

For the letter A example, the IBM research team fed the computer thousands of examples of the letter A so the computer could “recognize” the pattern, general shape and construct of a letter A. The examples are what gives Watson the ability to recognize an A. The IBM scientists followed a similar pattern to “teach” Watson clues as they are used on the game show. In fact, Watson was given thousands of old clues from the game to help it work on answers for new clues. It “learned” for the pattern and examples of previous questions.

Enter instructional design. What happens in a typically designed program? The designer creates abstracted bulleted lists or items the learner must know and apply to a situation and then, once or twice in the course of the class/e-learning event, etc. The learner is given an example. One or two examples at the most. Not enough to develop pattern recognition or to create an internal construct of how to deal with a particular situation.

That’s all wrong. Instead of giving learners abstractions of concepts or lists of rules, we need to give them examples, not one or two examples but dozens and dozens of examples.

We know expertise comes from experience with situations that build a generalization by the expert who then compares a current situation with past situations to decide how to problem-solve. Designers of instruction can create “learning experiences” using case studies, simulations, etc. to immerse the learner in dozens of similar (but not exactly the same) situations so the learner can recognize situations, not-by-rules, but by experience.

We can’t teach every rule in compliance training, or every answer to a customer’s objection in sales training or every combination of troubleshooting customer problems but we can provide example after example after example that can help learners develop the ability to recognize and address situations and the right response.

So, next time you develop instruction, provide examples, not one or two but dozens.