After over half a century the Turing Test appears to have gone out of favor, as AI programs become increasingly complex and scientists search for new ways to discern real humans from mimicking machine-learning algorithms.
One such test offers a truly curious set-up worthy of science-fiction. Imagine a world in which AI androids are outlawed and being accused of being an android is punishable by death. Think Philip K. Dick's sci-fi epic Bladerunner. To prove that you are really human, you can speak only one single word. What word would you chose?
This is what two MIT researchers, John P. McCoy and Tomer D. Ullman, call a 'Minimal Turing Test.'
Interestingly (but arguably not unexpectedly), despite freedom of choice, words chosen primarily clustered around a small amount of themes, according to The Verge. The word ‘love' was overwhelmingly popular, chosen 134 times, followed by ‘compassion' (33 times), ‘human' (30 answers), and ‘please' (25 times).
Many answers, while different, clustered around the same theme, such as empathy ('emotion,' 'feelings,' and 'sympathy'), others focused on religion and forgiveness ('mercy,' 'hope,' 'god' and even 'Jesus'). Some answers related to human biology, while others included various kinds of food.
Of particular interest are words that appear to have been chosen specifically because of their complexity, such as 'onomatopoeia,' 'bootylicious,' and 'supercalifragilisticexpialidocious.' Oh, and, of course, there were 'penis,' 'vagina' and 'poop.'
Speaking of poop, this particular word made a spectacular breakthrough during phase two of the test, in which the participants were offered pairs of words and were tasked to choose one which they believed could only have been proposed by human being.
In this test, words like 'love', 'human' and 'please' also scored strongly, but nowhere near as strong as ‘poop.'
"Yes, out of all of the word pairings, 'poop' was selected most frequently to denote the very essence and soul of humanity. Poop," the Verge report reads.
In their interview for the Verge, the researchers admitted that the test was more of a social psychology exercise than pure computer science.
"The practical application it has in the AI computer space is more when you're thinking about user interface design and things like that. In those contexts, it's perhaps useful to know how people think about computers and what they think sets them apart," McCoy said.
The classic Turing Test requires a computer to convince a jury that it is human by speaking. The threshold for the test is relatively low — only 30 percent of the jury must be convinced they are talking to a human being. However, in the world of rapidly evolving computers and especially programs specifically designed to imitate a chat (one such program actually passed a Turing Test in 2014), a one-word test might be enough to expose an artificial mind.
One example provided by the Verge is the asking of the question: "What are the plurals of ‘platch' and ‘snorp?'" The website says that the most likely human response would be 'platches' and 'snorps,' despite these words not being in the dictionary, (although 'platch' actually is in the dictionary, but the point is still taken).
In case of that one-word life-or-death scenario, the nonsense word 'supercalifragilisticexpialidocious' would be a dead giveaway, as some observe that an AI would never come up with anything so creative. Unless it was aware of the 1964 film ‘Mary Poppins.'
Of course 'poop' works just fine, too.