The Turing test gets a lot of attention in the world of artificial intelligence (AI) programming. It’s a test devised by Alan Turing back in 1950, when the first computers were being developed, as a way to measure whether the machine exhibits intelligent behavior (by human standards).
Conceptually, the test is very simple; in it, a human being has a text-based conversation with a machine and with another human, and if the test subject is unable to distinguish between the two, the machine can be said to have “passed” the Turing test. In other words, the Turing test measures whether a machine can pass as a human in conversation.
In recent years, there’s been a surge of interest in the Turing test, as sophisticated chatbots and other programs appear to pass the test in various conditions. But is the Turing test really as significant as it sounds? If our machines are currently passing it, by technical standards, does that mean we’ve reached the bold, advanced future that Turing predicted?
Why the Turing Test Is Irrelevant
The Turing test isn’t the ultimate sign of machine complexity, nor does it carry the significance it once did. Here’s why:
1. We aren’t using it the way it was designed.
When the Turing test was first introduced, by Turing, it was conceived as a way to evaluate whether or not machines could think in a manner similar to humans. It was meant to be a guide on how the term “artificial intelligence” could be applied. These days, it’s portrayed in media franchises and popular culture as a test to gauge whether a machine appears human in some way. What’s the difference? Consider a chatbot; its conversation may appear very similar to those you’ve had with humans in the past, but is the programming behind that chatbot “thinking” the way conscious beings like us do? Your answer will depend on the questions you ask, but we aren’t even using the test this way.
2. We’re measuring generously.
Our standards for “passing” the Turing test are arguably quite generous. As originally designed, a machine passes the test if it manages to convince at least 30 percent of people that it is thinking or operating like a human (depending on how you’re using the test). Assuming the only options to choose are “human” or “machine,” it could feasibly pass with flying colors if participants were guessing using pure chance. If the test’s threshold were higher, perhaps at 60 or even 70 percent, it may carry more weight.
3. Turing test variations have already been passed.
One of the most important reasons why the Turing test should no longer be held as a standard is the mere fact that it’s been passed so many times already. Though we’re still years away from the development of general AI, specific applications have shown that machines can surpass or mimic human capabilities in music composition, strategic games like Go and chess, and even subjective artistic endeavors like poetry. If it’s this easy to pass the test, it either means the test is flawed, or we’ve moved so far beyond the need for it that it’s no longer a relevant measure.
4. The importance of distinction.
The Turing test was devised as a measure, but also as a challenge. Its question—whether machines can think like and act like human beings—encourages developers to create machines that are indistinguishable from humans. This calls to mind a number of ethical and philosophical concerns, including the importance of preserving and distinguishing human identities from machine identities—no matter what form they take. Prioritizing a lack of distinction could result in some tricky moral questions down the line.
5. Conversational ambiguity.
Some of the most publicized “passes” of the Turing test have been robots that mimic human conversational patterns. However, they aren’t mimicking the conversations of great linguists and orators; they’re passing for average human beings, and average human conversational patterns are messy, ambiguous, and rife with errors. That gives machines a disproportionate amount of wiggle room when responding to conversational prompts; it’s the equivalent of going up against an amateur chess player, rather than a grandmaster.
6. Humans love patterns (and other humans).
Human beings are around today because we’re highly skilled at pattern recognition, and we’re hard-wired to see human qualities in even the inanimate objects around us; it’s the most plausible explanation for why we’re so inclined to “see” human faces in inanimate objects like rocks or clouds. When we read lines of dialogue or see the results of actions taken, it’s easy for us to spot human-like patterns; it’s an innate bias in our way of thinking, and the Turing test doesn’t necessarily account for it.
7. It doesn’t help us solve problems.
Last but not least, the Turing test was meant to spark a discussion about a revolutionary new idea in computing; today, we’ve advanced so far, our top philosophers are invoking new schools of thought and pushing new boundaries that completely ignore or overwrite the concerns of the past. The main problems we have with AI are in the logistics of development, and in the ethical quandaries of large-scale production. The Turing test doesn’t help us solve those problems.
The Turing test was a radical, breakthrough idea that completely revolutionized how the world thought about (and designed) AI. The test’s original publication in “Computing Machinery and Intelligence” was an important step forward in computing and philosophy.
However, it can no longer be considered the standard gauge of machine intelligence, nor should we put it on a pedestal. It’s a standard that’s 67 years old, and it’s time we found a replacement.
This post is part of our contributor series. It is written and published independently of TNW.
This post is part of our contributor series. The views expressed are the author's own and not necessarily shared by TNW.