What Is the Turing Test in Artificial Intelligence?

If you’ve spent much time reading about AI, you’ve probably come across the phrase “Turing Test.” As much as people reference it, there seems to be a lack of understanding about what exactly it is. So, what is the Turing Test, and what does it mean for artificial intelligence?

The test comes from Alan Turing, a British mathematician in the early 20th century. He proposed the idea in 1950, calling it “the Imitation Game.” At its core, the “game” tests whether a machine is intelligent, but that may be an oversimplification.

Turing emphasized that determining if a machine can think is mostly meaningless. Instead, the Imitation Game looks at whether a machine can convince someone that it’s human.

How the Turing Test Evaluates Artificial Intelligence

The original version of the Turing Test involves a judge and two subjects, one human and one machine. Without seeing the subjects, the judge has to decide which one is the person and which is the machine. If the computer can convince the judge it’s human through its conversation, it passes the test.

According to Turing, if a computer can win this game, then it has demonstrated human intelligence. A machine doesn’t have to be self-aware or have human levels of thought processing to win. The point is that if a machine can pass as a person, it might as well be intelligent.

Since Turing first proposed it, the Imitation Game has evolved. Different researchers and organizations use various versions of the test, but most use more than one judge. If the machine can fool more than 30% of the judges, it passes the test.

Has Artificial Intelligence Passed the Turing Test?

Turing predicted that computers would be able to win the Imitation Game consistently by 2000. While AI hasn’t quite lived up to that expectation, a few machines have reportedly passed the test. The most widely acknowledged example happened in 2014 at a Turing Test contest.

At the competition, a chatbot called Eugene Gootsman convinced 33% of the judges that he was human. As you’d expect, a lot of people claim that Eugene didn’t really pass the test. Critics say that too many judges didn’t believe Eugene or that they didn’t have enough time. To add to this, even modern chatbots can’t pass the Turing Test with any consistency.

In 2018, Google revealed a conversation its AI, Duplex, had with a hair salon. Some claim that Duplex passed the Turing Test there since the salon employee never questioned that Duplex wasn’t human. While this wasn’t a controlled scenario, many people say that it’s a real-life example of AI passing the test.

Possible Issues With the Test

Not all researchers agree that the Turing Test is an accurate method of measuring artificial intelligence. One of the most widespread criticisms is that the test is too narrow. People argue that the test only measures intelligence in a specific scenario when intelligence can take many forms.

The Imitation Game relies solely on conversations to see if a machine can pass as a human. Some researchers argue that whether something can hold a conversation doesn’t make it more or less intelligent. AI might be intelligent, but not in conversation, or it could be good at conversation, but unintelligent overall.

Other critics say that as technology advances, the criteria of what constitutes AI should rise. By that argument, a test from the 1950s is insufficient in measuring modern machines.

Measuring AI Isn’t Easy

If there’s one thing to learn from the history of the Turing Test, it’s that measuring intelligence isn’t straightforward. Artificial intelligence is too broad a concept to fit into any specific set of qualifications. While the Turing Test may not be foolproof, you can’t deny its significance.

The Turing Test, even if it isn’t perfect, helps us think about how we classify AI. If nothing else, it’s given AI developers a goal for longer than AI’s been around. Alan Turing proposed this test 70 years ago, and it’s still relevant today, perhaps more than ever.