The goal of the original imitation game, which Turing mentioned in a 1950 paper entitled “Computing Machinery and Intelligence”, was to see if machines could be flexible and display general cognitive abilities well enough to convince human evaluators of their humanity.
The basic form of the test is as follows: in one room there is a person who asks questions, which are then passed to another room, where they are answered by another person as well as a machine. The answers are given in a neutral form (mostly in writing) back to the person who raised the questions. If it cannot distinguish when it is communicating with a human and when with a machine, the given artificial intelligence has passed the Turing test.
Despite more advanced methods of testing “human behavior” in artificial intelligence, this test is still popularly used today. In March of last year, for example, GPT 4.5 – one of the latest language models (LLM) from OpenAI was evaluated in the Turing test. In 73% of cases, he was rated as human, more often than real people. Readers even preferred texts generated by LLM over texts written by human experts, reports Nature.
General intelligence in humans and machines
Large language models (LLM) succeeded, for example, in the International Mathematical Olympiad, they collaborated with leading scientists in the generation of scientific hypotheses, which were then verified in experiments.
They also help programmers write codes, compose poems, and last but not least, they communicate with millions of people around the world every day.
It can be said that they already show a whole range of cognitive competences, which Turing focused on – it is comparable to what we refer to today as general artificial intelligence (AGI).
Even so, many scientists claim that current AI models do not show general intelligence – according to the biggest critics, they will never show it. When looking deeper into the issue, among other things, doubts appear regarding the exact definition of what general intelligence is.
