1
00:00:00,366 --> 00:00:10,259
TURING TEST A test of whether a computer can carry on a conversation well enough for a human interlocutor to be unable to distinguish it from another human.

2
00:00:10,303 --> 00:00:25,628
Named after the mathe– matician and father of computing, Alan Turing (19121954), who proposed a thought experiment he called the ‘imitation game’ in an article in 1950.

3
00:00:25,644 --> 00:00:36,285
Turing was addressing the question, ‘can a machine be intelligent?,’ but dismissed it as ‘too meaningless to deserve discussion.’ After all, what is meant by ‘intelli– gent’?

4
00:00:36,297 --> 00:00:46,516
Instead Turing proposed that if a machine could appear to be as intelligent as a human, then for all intents and purposes it could be considered as intelligent as a human.

5
00:00:46,524 --> 00:00:59,559
He referred to an imitation game in which a man tries to answer questions as he thinks a woman would, and a questioner tries to tell the imitator apart from a real woman by means of written questions and answers.

6
00:00:59,613 --> 00:01:19,355
Substituting a computer for the imitating man, Turing suggested that if an interlocutor is allowed to ask wide– ranging and penetrating questions via text, and cannot on the basis of the answers given distinguish a computer respondent from a human one, then the computer could be said to be, in some senses, intelligent.

7
00:01:19,371 --> 00:01:26,171
Turing thought that by 2000 machine intelligence would be able to pass this test seventy per cent of the time.

8
00:01:26,185 --> 00:01:35,673
In practice no machine has ever come close, highlighting the difficulty of the task facing those attempting to engineer artificial intelligence (AI).

9
00:01:35,673 --> 00:01:45,065
The Turing test has been more fruitful in stimulating debate about the nature of intelligence, and what it means to talk about artificial or machine intelligence.

10
00:01:45,069 --> 00:01:56,310
The Turing test suggests a behaviourist–style approach to intelli– gence, with an intelligence seen as a sort of ‘black box’ into which you feed questions and out of which come answers.

11
00:01:56,325 --> 00:02:05,632
This sort of input/output or I/O model says that what’s inside the black box isn’t important, only the inputs and outputs matter.

12
00:02:05,632 --> 00:02:15,747
It’s an approach that has been criticized for missing essential qualities of intelligence, most notably by the philosopher John Searle in his Chinese Room thought experiment.

13
00:02:15,815 --> 00:02:22,663
Searle imagined a man in a room who speaks no Chinese, but is passed Chinese messages through a slot in the wall.

14
00:02:22,680 --> 00:02:31,741
Using a big book of syntactical rules written in English he is able to process the Chinese symbols into answers, which he posts back out of the slot.

15
00:02:31,754 --> 00:02:40,086
To a Chinese interlocutor outside the room he seems to speak Chinese, but in fact he has no semantic understanding of Chinese.

16
00:02:40,143 --> 00:02:50,232
Similarly, Searle argues, a machine intelligence passing the Turing test is processing language to produce answers, but has no real semantic understanding.