In 1950 Alan Turing, the British mathematical genius who with
the help of crude early computers broke Hitler's codes, proposed a
test to determine if a machine was intelligent, in the event that
machines ever became intelligent. This is the Turing Test: if you
are communicating through a computer and believe you are
communicating with a human being when in fact you are communicating
with a computer, then the machine can be deemed to be
intelligent—it passes the Turing test.
There have been many serious and some not-so-serious attempts to
create computer programs that would pass the Turing Test. Among
these was a program called ELIZA (1966) which attempted to imitate
a nondirective therapist. Programs like ELIZA try to identify key
words and phrases in things you type and try to compose what
appears to be relevant replies based on those keywords. In the role
of nondirective therapist, ELIZA could cut many corners because it
could always fall back on nondirective prompts such as "Go on,"
"How did you feel about that?" and "Tell me more about ______,"
where the blank could be filled in with a key word from whatever
you had had typed. A transcript of a conversation with ELIZA could
have long stretches that looked like a real communication between
two people, but it was not too difficult to trip ELIZA up and get
ELIZA to emit a response that was nonsense or was so ungrammatical
that no human being could have composed it (such as "How do you
feel about you do not how you feel about it?").
If the Turing Test is a test of Artificial Intelligence, I want
to propose an Anti-Turing Test. It is a test of Artificial
Stupidity. Here is the test: if a customer cannot tell when
interacting with your call center or your web-based feedback form
that he is communicating with a human being and not with a
computer, then your feedback system has Artificial Stupidity.
One of the surest signs of Artificial Stupidity is failure to
answer simple direct questions. True enough, sometimes that answer
is not one that will be pleasing to the customer. Sometimes the
answer is "No." Sometimes the answer is "I don't know," or "I don't
have that information," or "I cannot release that information."
Those are not pleasing answers, but they do acknowledge the
customer's question.
If you have a problem-tracking system, a sure sign of Artificial
Stupidity is failing to note the history of the problem in
responding to the current communication. If the tracking system is
not used to produced a better answer this time, what do you have a
tracking system for? When the chain has got two or three
question-response cycles deep communication is not taking place.
The thread needs to be directed to a specialist or supervisor who
has the skills, knowledge, and policy latitude to acknowledge the
issues, to answer novel questions and issues, and to refer the
thread appropriately if necessary. Not every customer-care employee
needs this level of skill. Not even very many of them do. After all
most calls by far do fall into recurrent patterns that can be
responded to by script. But when the script doesn't work, someone
who can go off-script should be available.
A red-line example of Artificial Stupidity is the thought that
customer-care performance can be measured by how often the customer
goes away. Yet, when performance is measured in call numbers, the
incentives favor employees who make customers go away in the most
efficient manner. Sometimes the customer goes away permanently. An
issue cannot be presumed resolved just because the customer stops
bothering the customer-care representatives. The metric for
customer-care performance is how often customers who have contacted
customer care come back to make a purchase. If that isn't the
outcome you are aiming at, why have any customer care at all? Maybe
you are a philanthropist who can afford, out of the goodness of
your heart, to employ people to wear headsets and sit in front of
computer screens drinking coffee regardless of the effect on your
revenues.
