Tutorial on Artificial Intelligence To best illustrate how Adaptiqs implements
artificial intelligence in its surveys, imagine
a job related-game of "20 Questions" in which you have to guess the job
of a new acquaintance named Joe using information he provides about the
his job. To help you guess, you may ask questions from a list of over
450 questions that assess every aspect of Joe's job including typical
working conditions or knowledge and skills required. From Joe's responses
to these questions, you will try to guess his job in the fewest number
of questions possible. So, what would be the best strategy?
One strategy would be to ask each question in the list in order until
you have asked every question in the list. After collecting all of this
data, you will analyze his responses to each question in order make your
guess. Obviously, this is the least efficient of all strategies. During
this process, Joe's patience will wear thin and he will likely lose interest
and stop responding. Surprisingly, this is how most surveys are administered
today.
A more efficient strategy would be to eliminate redundant and irrelevant
questions from the list based on Joe's responses to questions. For example,
if Joe reports that he does not work outdoors, chances are questions about
exposure to radiation or heights might be irrelevant so you can decide
not to ask those questions.
In addition, if Joe reports that he uses mathematics frequently on his
job, chances are good that he also uses a computer frequently (to perform
calculations). Therefore, asking both questions seems a bit redundant.
In short, you could create a set of rules (or "decision tree") to help
you select which questions to ask. Survey tools that claim to be "adaptive"
use this approach as a way to reduce administration time. However, decision-tree
approaches like this have two major weaknesses.
First, the process of determining the set of rules to guide when to skip
questions is very time consuming, especially with long lists of questions
that are interrelated. The time it takes to create these decision rules
can reduce the benefit gained from using them to reduce the number of
questions.
Second, the decision-tree approach can lead to incorrect assumptions and
poor decision-making. For example, if you assume that because Joe does
not work outside, it is safe to rule out jobs also involving exposure
to radiation and heights, you would be wrong if Joe were a Nuclear Engineer
or Commercial Airline Pilot.
Similarly, you would be wrong to assume that since math is important in
Joe's job, the use of computers are also important, especially if he worked
as a blackjack dealer or a tollbooth attendant. When using rules or decision-trees
to eliminate questions, you are assuming that because someone responds
in a particular way to one question, you know with 100% certainty the
answers to other questions. This is rarely the case in most surveys, so
using this strategy can often lead to inaccurate results.
The best approach is to use a probability-based approach and use each
piece of information in making decisions and determining the most informative
questions to ask next. This will provide the most accurate decisions with
the fewest number of questions. Although this may sound complex, people
use this strategy daily to make decisions without consciously thinking
about it.
For example, assume you ask Joe the following questions:
1. How important is written communication to the performance of your job?
To this question Joe responds "Very Important." Using this information,
there is a fairly good chance he could be a creative writer, but not necessarily.
He could also be an editor, a professor, even a secretary. Frankly, many
jobs require written communication. With efficiency in mind, it would
be nice to ask another question that best differentiates among jobs that
require written communication. After scrolling through the list of jobs,
you decide on the following question:
2. How important is teaching and instructing
to the performance of your job?
To this question, Joe answers "Extremely Important." Based on this information,
chances are fairly good that he might be a teacher of some sort, but there
are many other professions that require written communication and teaching
and instruction (i.e., manager). In addition, chances are good that he
is probably not an editor, creative writer, or secretary. Still, we are
not sure by any means, as many individuals in those jobs are required
to mentor, coach, or supervise others. We definitely need more information.
After looking through the list of questions, you might decide to follow
up with a question related to what a typical college professor would do:
3. How frequently do you speak in front of groups
of people?
To this question, Joe answers "Daily." Now you are more confident that
he is some kind of teacher, but what kind of teacher is unknown. As you
keep asking questions, you can use the information you gained to help
be increasingly more certain about his job title. After continuing this
process for several questions, you will feel that have enough information
to make an educated guess.
At this point, it's not productive to continue asking questions because
additional questions will not add much information to your decision. This
is the underlying concept behind artificial intelligence. It uses patterns
it learns from previous data to make predictions on future events. Adaptive
surveys are capable of performing over 3.3 million calculations between
questions to determine the ideal question to ask next.
