could play the game satisfactorily, we need not be
troubled by this objection. He gives sample ques-
tions and answers, and interestingly, these include
a chess question, showing how broad and flexible
his test is.
Finally, Turing considers many possible objections
to the idea that a machine could ever truly be said
to think, and states his own opinion on the matter.
I believe that in about fifty years’ time it will
be possible to programme computers, with a
storage capacity of about 10^9 , to make them
play the imitation game so well that an average
interrogator will not have more than 70 per
cent. chance of making the right identification
after five minutes of questioning. [. . .] at the
end of the century the use of words and general
educated opinion will have altered so much that
one will be able to speak of machines thinking
without expecting to be contradicted.
(1950, p. 442)What a prescient and carefully worded prediction!
Turing was absolutely right about the change in
the use of words. I do not expect to be contradicted
if I say that my laptop is ‘thinking about it’ when it
reacts slowly, or ‘my phone thinks it’s Thursday’
when the date is wrong. We talk to Siri and our ‘OK
Google’ apps in the expectation that they’ll hear,
understand, and respond to us. On the other hand,
even my lowly desktop has a storage capacity far
larger than Turing guessed, and yet it could not pass his test.
When the fifty years were up, many programs could pass limited Turing tests.
The first was ELIZA, which used scripts based on rudimentary pattern-matching –
repeating and lightly transforming sentences in a psychotherapeutic manner –
to give an illusion of understanding, and was genuinely helpful to people with
psychological difficulties (Weizenbaum, 1966). People today are often fooled
in text-based internet chat rooms and on social networking sites. Sometimes
things go wrong, however: Tay, a chatbot developed by Microsoft to interact
with 18–24-year-olds on Twitter, learnt to communicate using anonymised pub-
lic data as well as a group of humans, but rapidly began to tweet like a sexist,
racist troll, as well as failing to show any interest in ordinary human topics like
pop music or TV.
In 1990, the first annual Loebner Prize competition was held, offering an 18-carat
gold medal and a large cash prize for any program that could pass the Turing
test, and an annual bronze medal for the most human-like entry of the year. At
first no computer came close to passing, despite various restrictions imposed to
make the test easier, and Dennett concluded that ‘The Turing test is too difficult
ACtIVItY 12.1
A Turing test for creativityDoes it require a conscious human being to paint,
draw, or write creatively, or could a machine do as
well? If so, it might convince an observer that it
was human – in other words it could pass the Turing
test. In fact, at least one poem has done just that.
As a student, Zachary Scholl used a program which
generated poems so convincing that one was accepted
for publication without anyone doubting it was written
by a human.
For a fun class activity, choose a variety of paintings,
pieces of music, jokes, or poems, some of which are
created by machines, to see how easily people can
tell which is which. A choice of three works well, with
people guessing which is machine-created. You can
either supply these in advance or ask students to bring
in examples without saying where they came from. Get
everyone to vote and find out how well the machine
creations fare. What kinds of features make people
infer human authorship or its absence, and why?
Poetry is a good candidate for this kind of Turing text
because poems can be manageably short. See the
website for suggestions and for another more classic
Turing test activity.