The Turing test, named after British mathematician Alan Turing was a concept proposed to test if a machine could deceive a person into thinking it was human. Turing called it the ‘imitation game’ and first discussed it in his 1950 paper, ‘Computing Machinery and Intelligence’ while working at the University of Manchester. Turing was addressing a philosophical problem that other mathematicians and scientists were wrangling with, given that the first computers — based on vacuum tubes and large electrical systems — were already around. The question was: Can machines think? Given that this was a subjective question, Turing broke it down into a testable format.
The hypothesis
Turing proposed that an objective way to test for intelligence in machines is to have a computer perform a task in the same way a real person would. Computers in their early days were envisaged as calculation machines and, though nowhere near the speed at which they compute today, they could perform calculations beyond human capability. Therefore, having a computer and a person perform complex math calculations wouldn’t be an appropriate test of a machine’s humanity. Therefore, Turing proposed the equivalent of a ‘party game’ wherein a man and woman would go into separate rooms and answer, in writing, questions posed by guests. The aim of the game was for the man and the woman to convince the guests that they were the other.
Turing is considered the father of modern computer science, having in 1936, conceived of a ‘Universal Machine’ that in theory could compute anything and lies at the heart of the modern digital computer. His conception of computers as thinking machines gestated the development of Artificial Intelligence.
The results
Passing the Turing test would mean that computers were able to convince people they were human. Turing stated that it wouldn’t be inconceivable for computers, 50 years from the writing of his paper, to pass the ‘imitation game.’ (Mathematician Alonzo Church was the first to refer to this problem as the ‘Turing test.’). Since then, at various points, machines have superficially passed the test.
The earliest was a program called ELIZA, created at the Massachusetts Institute of Technology by Joseph Weizenbaum in the 1960s. ELIZA, was a chatbot modelled on a psychotherapist. In response to typed questions, Eliza was programmed to pick on certain words, reframe that as a question and pose it back. For eg: Eliza would begin: “Is something troubling you?” A response: “Men are all alike.” Eliza: “What is the connection, do you suppose?”
Because these conversations could go on indefinitely, many users believed that they were talking to a sentient machine who understood them and sometimes, offered insight. Weizenbaum developed ELIZA to show that it was possible to simulate conversation with a machine, without the latter grasping the context. ELIZA was never part of a formal evaluation of a Turing Test but is nevertheless considered important in the history of machine intelligence because it suggested that machine-human interactions could be conducted in a way that gave an illusion of intelligence.
The Loebner Prize is a competition that has provided a platform for practical Turing tests. A panel of judges awarded prizes for the program that came closest to imitating human like conversation. The interactions between man and machine are conducted via text-chats. The chief criticism of programs that have won is that they aren’t really designed to determine intelligence but rather simply to fool humans. The assumption is that intelligence is the ability to deceive. There have been significant advances in machine’s ability to understand human language, as the increasing use of Alexa and Siri suggest, but it's quite apparent that they are far from being considered to have reached a stage where they can easily trick most people into believing them to be human.
