Turing test

The Turing test, originally called the imitation game by Alan Turing in 1949, is a way to see if a machine can act as intelligently as a human. In this test, a person talks to both a human and a machine using text messages. The person tries to guess which one they are talking to. If they cannot tell the difference, the machine has passed the Turing test.

Turing shared this idea in his famous 1950 paper, "Computing Machinery and Intelligence," while he worked at the University of Manchester. He asked, "Can machines think?" Because it is hard to define what "thinking" means, he suggested a simpler question. He used a party game where someone guesses if they are talking to a man or a woman. He then asked if a computer could do well in a game like this.

Since Turing introduced his test, it has had a big impact on the philosophy of artificial intelligence. Many people have talked about and debated the idea. Some philosophers, like John Searle, have argued that the test cannot really show if a machine is conscious. Even with these debates, the Turing test remains an important idea in understanding how we can tell if a machine is acting like a thinking being.

History

The idea of whether machines can think has been discussed for a long time. Early thinkers like René Descartes wondered if machines could think like humans. Later, Denis Diderot suggested that if a machine could answer any question, it might be intelligent.

Alan Turing introduced what we now call the Turing test in 1950. He suggested a game where a person tries to tell if they are talking to a machine or a human by sending text messages. If the person cannot tell the difference, the machine has passed the test. This idea started many discussions about whether computers could ever think like humans. People have debated what the test means, with some saying it doesn’t prove a machine is thinking, only that it can talk well. The test is still a famous way to explore what artificial intelligence can and cannot do.

Attempts

Some early computer programs were said to pass the Turing test by using simple tricks, like pretending not to understand English very well.

In 1966, a program named ELIZA was created to mimic a therapist. It worked by repeating keywords from the user's messages, making some people believe they were talking to a real person. In 2001, a program called Eugene Goostman pretended to be a 13-year-old boy learning English, and some judges thought it was human.

More recently, advanced programs like ChatGPT have come close to passing the Turing test. In March 2024, researchers reported that ChatGPT passed the test by being more cooperative than average human behavior. In a study in March 2025, another program was identified as human most of the time.

Large language models

Main article: Large language model

Google LaMDA

Main article: LaMDA

ChatGPT

Main article: ChatGPT

Versions

Alan Turing described a game with three players to see if machines could think like humans. In the game, one player tries to trick another into guessing wrong about who is a man and who is a woman. They only send written notes. Turing wondered what would happen if a machine took the place of the person trying to trick the other player.

Later, Turing suggested a new version where a computer tries to trick someone, and a human helps the person being tricked. Some people think the goal of the Turing test is to see if a computer can act like a human. This idea is called the "standard interpretation." Not everyone thinks this was Turing's original idea. In this version, the person tries to figure out if they are talking to a human or a machine.

Interpretations

People have talked about what Alan Turing meant with his test. Some think he had two ideas. One is like a game where a judge looks at conversations to guess who is a person and who is a computer. The other is where a judge talks to both a person and a computer to see which one is which.

Others believe Turing wanted to know if a machine could think like a person. They say his test shows how well a machine can seem human. Some also say the test is more about showing smart skills and solving problems.

Strengths

The Turing test is popular because it is simple. Unlike fields such as the philosophy of mind, psychology, and neuroscience, the Turing test gives a clear way to measure intelligence. It lets the person testing ask many kinds of questions, which is useful for many areas of study.

To pass the Turing test, a machine must be able to use natural language, reason, have knowledge, and learn. The test can also include tasks that need vision or robotics skills. This makes it a good way to explore challenges in artificial intelligence research. The test also shows that understanding feelings and beauty may be important for creating safe and friendly AI systems.

Weaknesses

Alan Turing created the Turing test. He did not say it could measure intelligence. Some people have suggested using it that way. This idea has faced criticism. Some worry that the test only shows how easy it is to trick humans, not if a machine is truly intelligent.

The Turing test has several challenges. It tests if a machine can behave like a human, not if it is truly intelligent. For example, it might need to copy human mistakes to pass. Also, the test does not show if a machine could solve hard problems better than humans. Because of these issues, many experts think the Turing test is not the best way to study artificial intelligence. They prefer testing specific skills, like recognizing objects, instead of trying to copy humans exactly.

Variations

Many different versions of the Turing test have been suggested over the years.

One version is called the reverse Turing test. In this test, the computer tries to tell if it is talking to a human or another computer. CAPTCHA tests, where you must read distorted words to prove you are not a robot, are an example of this reverse test.

Another version is called the subject-matter expert Turing test. This test asks if a machine can talk like an expert in a certain area, such as medicine or science. There are also tests that try to see if a machine can understand language deeply, not just copy words. There are tests that include seeing and moving objects, like a robot might need to do.

In 2023, a company called AI21 Labs created a big online game called "Human or Not?" that was played millions of times. The results showed that some people could not tell if they were talking to a human or a computer.

Alternative tests for machine intelligence

The Lovelace test is named after Ada Lovelace. She thought computers should only be trusted with intelligence if they can create original ideas.

In 2023, David Eagleman said that truly intelligent systems should be able to make scientific discoveries. He described two levels. Level 1 is when the AI connects facts we already know. Level 2 is when the AI creates new ideas and tests them. Other tests for AI intelligence include the Winograd Schema Challenge. This tests understanding of language. There is also the Allen AI Science Challenge for answering science questions. Finally, there is the Artificial General Intelligence (AGI) Test. This sees if a machine can do any intellectual task a human can.

Conferences

In 1990, the Turing Colloquium was held at the University of Sussex to celebrate Alan Turing's work. Experts talked about the Turing test and its future.

The same year, the Loebner Prize competition began.

In 2008, a special meeting was held at the University of Reading with the Loebner Prize. Famous thinkers talked about the Turing test, but they did not all agree on what the test should be.