Turing Test in the Real World
The Turing Test uses three different terminals. Two of these terminals are occupied by humans and the third one by a computer.
One of the humans acts as a questioner while the other human and the computer respond to the questions. These questions are on specific subject areas with an agreed-upon format.
As the test progresses and after a set period of time, the human questioner is asked to determine which of the two respondents was a human and which one was a machine. This sequence of events is conducted several times, and if the machine is mistakenly judged as a human greater than 50% of the time, it is judged to have passed the Turing Test.
Turing Test in the World of AI
Historically, there have been some limitations with the Turing Test. Primarily, the fact that questions needed to be formulated so that the answers could be provided in a Yes or No format. Open-ended questions were ones where the machine was often caught out.
The Turing Test was first defeated in 1966 when Joseph Weizenbaum created a program known as ELIZA. While ELIZA was not 100% successful, it was able to fool some judges. In some cases, additional persuasion was needed to prove that ELIZA was not only a machine; it was not human.
More recently, the Turing Test was defeated in 1990 when Hugh Loebner created a chatbot called Eugene. Eugene was programmed to emulate a 13-year old Ukrainian boy, which helped ensure that the conversation did not get too broad. In the test that Eugene was a part of, he was able to convince 33% of the judges that he was human.
While the Turing Test no longer holds as great weight in the world of AI, its impact has definitely been felt. In fact, today, reverse Turing Tests use CAPTCHA to require humans to prove they are not, in fact, machines.
A CAPTCHA for those not in the know stands for Completely Automated Public Turing test to tell Computers and Humans Apart, which demonstrates the value of the Turing Test more than anything else.
