Artificial Intelligence

Turing Test

For a given machine, how can we test for its intelligence or how can we say that the machine is intelligent or not? Turing test is the first test proposed to determine whether a machine is intelligent or not. In 1950 an English mathematician called Alan Turing proposed a test called the imitation game, to test how to extend did a machine can imitate like a human or to calculate the machine’s ability to exhibit human-like intelligent behavior. This testing method is called the Turing test.

What is the Turing test?

Turing test is a method used to test whether a machine can act or think like a human or not. the test agrees a computer to be intelligent only if it can mimic like a human under specific conditions. The result of this test is the answer to the question “CAN MACHINE THINK?”. The English mathematician, computer scientist, cryptanalyst, and theoretical biologist Mr. Alan Turing was the one who was behind this proposal, and hence the test was named as Turing test.

The test requires three players or terminals. Each of them is completely hidden or separated from the others. One of them is a computer or operated by a computer. The other two are humans or operated by humans.

During the test one among the humans acts as a questioner and the other one and the computer act as the respondents. The questioner will ask questions to the respondents about any specific area in a specified format. After a fixed time or a fixed number of questions, the questioner must differentiate the computer from the human.

The test will repeat for many times. If the questioner cannot differentiate the machine from the human in half the time, the machine is said to be won the test and can be said that the computer is considered to have artificial intelligence. The result of the test depends on how extended the machine can answer as a human would give.

HISTORY

Alan Turing, the great English computer scientist who is also considered as one of the fathers of AI and also a pioneer of machine learning is the man behind this concept. He introduced the test in his paper called “Computing Machinery and Intelligence” in 1950

Turing Test

Through this test, Turing tried to answer the question “can machines think?” in an indirect way. He first rearranges the question itself to “can a machine imitate a person?”

Turing proposed a game called imitation game with no involvement of any AI. He conducted the game with three human participants in three different rooms and they are connected through the keyboard and screen. There are 2 males and 1 female in this game as participants. Player B (female) tries to convince player C (judge, male) that she is a male. C tries to discover who is male and who is female by asking questions. 

Turing Test

An illustration of the imitation game is shown above. A – Male, B – female, C –male (judge) are placed in three separate rooms. Only C can interact with A and B through a keyboard.

How did the test do?

After his imitation game Turing put forward a question “can machines think? Or are there any digital computers which will pass the imitation game?” The imitation game is played by replacing one human player with a computer.  

Turing Test

Consider the above figure. Here there are three players. Player A, Player B, and Player C. Player A is a computer, player B is a human responder and C is a human questioner or interrogator. Here three of them are isolated from each other. The interrogator knows that one among the players is a computer and the other is a human. But the interrogator needs to determine which one is the machine and which one is the human based on the questions asked and the answers replied. Here the interaction between players is through keyboards and screens. The computer or player A is allowed to do anything possible to cheat player C (interrogator). The final result of the tests depends on how closely player A responds like a human. 
The interaction between the players would be like this:

Player C: Are you a computer?
Player A: No.
Player B: No
Player C: Multiply 18765439*8749049
Player A: After a while, an incorrect answer 
Player B: incorrect answer after a long time.
Player C: Add 524310, 34521
Player A: pause about x seconds and then give an answer 558831.
Player B: pause about Y seconds and then answer 558831
Player A would be considered intelligent only if its conversation couldn't be easily distinguished from player B by player C.

Importance of Turing Test in AI

Turing tests can’t be considered as a relevant method for evaluating an AI system. But it must be noted that the Turing test was proposed in the year 1950, 6 years before the notion of AI does not exist. Yet Turing was already thinking about the question can machine think? He is the one who described the framework to determine the machine’s intelligence. Till the Turing test is remaining elusive for AI systems. No AI has passed the Turing test. ELIZA and PARRY had come close to passing. Eugene Goostman, a computer programmer, is the first AI to pass the test. Turing test can be used to judge the conversational skills of a bot. The Turing test gives plenty to think about in terms of how to define intelligent behavior and what would we want from an intelligent robot.

Limitations of Turing Test

  1. Mistakes are encouraged.- According to the Turing test, mistakes are necessary for machines. That is the aim of the machine is to keep its identity hidden from the interrogator so it must make mistakes. For example, if the interrogator asks to multiply 2 large numbers, the machine is forced to give a wrong answer even if it knows the correct answer.
  2. A ramp of intelligence is not provided - the Turing Test provides only two conditions for the machine, whether the machine is intelligent or not. It doesn't encourage defining the level of intelligence
  3. Machine intelligence is not tested - During this test, the machine intelligence is not tested. It only tests whether the machine can behave like a human. Human behavior and intelligent behavior are not the same in all senses.   Turing test requires the machine to execute some unintelligent human behavior, for example, the temptation to lie. If the machine failed to imitate such type of unintelligent behavior then it fails the test. And also the test doesn't consider the ability to handle difficult problems as an intelligent measurement. If the machine could do more intelligent work than a human can then it fails the test and is not considered as intelligent. This in turn results in an unintelligent machine passing the test and an intelligent machine failing the test.

Turing test has both proven to be both influential and also widely criticized with serious flaws as discussed above. Anyways it has become a very important concept in the philosophy of AI.
The argument presented by the philosopher John Searle in 1980 called the Chinese room argument and is a thought experiment is one of the main criticisms against the Turing test. It holds that a digital computer cannot have a mind regardless of how intelligently the programs make the computer react or a system can be intelligent without being intelligent actually.  Similar arguments were presented by other experts also.

Variations of the Turing test

There are numerous versions of the Turing test that have been raised. Some of them are:

Reverse Turing test 

As like the name it is a reverse procedure. A modified Turing test where one or more of the roles have been reversed between humans and machines. This modified method
can overcome most of the objections of the standard version. CAPTCHA is an example of a reverse Turing test. The user is presented with some alphanumeric characters in a distorted graphic image and asked the user to type those characters before allowing the user to perform some action on a website. This method is used to prevent abuse of sites by automated systems. Any system that can do it correctly will be human because such software that can read and reproduce such distorted images does not exist.

Subject matter expert Turing test.

The subject matter Turing test also called the “Feigenbaum test” is another variation of the Turing Test proposed by Edward Feigenbaum. Here the computer attempts to replicate an expert in a specific field like accounting or marketing.

“Low-level” cognition test

The low-level or the unconscious process of human cognition is revealed through questions. The interrogator can easily unmask a computer through these types of questions unless the computer experiences the world as humans do. This was introduced by Robert French.

Total Turing test

The cognitive scientist Stevan Harnad adds two more requirements to the traditional Turing test. In addition to linguistic interrogation, it involves visual abilities and physical interaction. This modified version is called the Total Turing test.

Minimum intelligent signal test

As the maximum abstraction of the traditional Turing test, Chris McKinstry proposed a version with only binary responses permitted. The focus is only on the capacity for thought.