In the above sections you met some chatterbots, and (hopefully!) have drawn the conclusion that they aren’t entirely convincing in terms of sounding like a human (although some are better than others!). But maybe soon, there will be new chatterbots that don’t have the same limitations. Should we consider them to be intelligent? How could we tell? Is there a formal way we can determine whether or not a chatterbot is of the level of human intelligence?
A very famous computer scientist, Alan Turing, answered this question back in 1950, before the first chatterbots even existed! Alan Turing had an extraordinary vision of the future, and knew that coming up with computers that were intelligent would become a big thing, and that we would need a way to know when we have succeeded in creating a truly intelligent computer.
He thought about how intelligence could be defined (defining intelligence is surprisingly difficult!), and decided that one way would be to say that a human was intelligent, and that if a computer was able to communicate convincingly like a human, then it must be intelligent also. This definition doesn’t cover all of intelligence, as it only considers what a person or a computer says and ignores other components of intelligence such as determining the best way to walk through a building (or maze) or deciding how to act in a specific situation (such as at a social event, when deciding what to do next at work, or when lost). However, communication is still a very significant component of human intelligence.
In order to test whether or not a computer program can communicate like a human, Turing proposed a test. In addition to the computer program, two humans are required to carry out the test. One of the humans act as an interrogator, and the other as a "human" to compare the computer program to. The interrogator is put in a separate room from the computer running the computer program and the "human". The interrogator has conversations with both the human and the computer program, but isn’t told which one they are having the conversation with at each time. The conversations are both carried out over something like an instant messaging program so that actual speech isn’t required from the computer program. During the conversations, the human has to convince the interrogator that they are indeed the human, and the computer program has to convince the interrogator that IT is actually the human. At the end of the conversations, the interrogator has to say which was the computer and which was the human. If they can’t reliably tell, then the computer is said to have passed the test.
This test proposed by Turing eventually became very famous and got the name "The Turing Test". One of the motivations for writing chatterbots is to try and make one that passes the Turing Test. Unfortunately, making a chatterbot that successfully passes the Turing Test hasn't yet been achieved, and whether or not it is even possible is still an open question in computer science, along with many other questions in artificial intelligence that you will encounter later in this chapter.
Other forms of the Turing Test exist as well. Action games sometimes have computer controlled characters that fight your own character, in place of a second human controlled character. A variation of the Turing Test can be used to determine whether or not the computer controlled player seems to have human intelligence by getting an interrogator to play against both the computer character and the human character, and to see whether or not they can tell them apart.
In fact, many parts of human intelligence could be tested using a variation of the Turing Test. If you wanted a computer chess player that seemed like a human as opposed to a computer (as some people might prefer to be playing against a human rather than a computer), you could use a Turing Test for this as well! What other possible Turing Tests can you think of?