We’ve all heard of the Turing Test, designed in 1950 by Alan Turing in a test of a machine’s ability to mimic human behavior.
But researchers at MIT and the University of Pennsylvania tried to do something a little different. They tried to create a Minimal Turing Test. Well sort of.
They asked themselves a question: If you could only use one word to convince a human you weren’t a robot, which word would you choose?
One thousand participants suggested 400 words. The most commonly suggested words related to concepts most people think an artificial intelligence might have difficulty understanding — words like ‘love’ or ‘soul’. But plenty of words were suggested more than once, words like ‘compassion’ or ‘mercy’. Even swear words like ‘shit’. One curious choice made by a number of people: ‘please’.
The researchers put all the suggested words into distinct categories, and then selected the most nominated word from each category.
Those words were: ‘love’, ‘please’, ‘mercy’, ‘human’, ‘compassion’, ’empathy’, ‘robot’, ‘banana’, ‘alive’ and ‘poop’.
The next step: Show 2,000 new participants a random combination of two of these 10 words, and have them select which word they thought was chosen by a human, and the word chosen by a robot.
The word most likely to convince another human being you’re not a robot?
Ladies and gentlemen, that word is ‘poop’.
To be clear, robots or AI were not involved in this test — it was a purely psychological experiment, designed to see which word would most likely convince other human beings that someone else was also human.
And when you think about it, poop makes sense. I see that word and I’m thinking ‘robots can’t poop’. Poop would be a remarkably difficult word for an artificial intelligence to parse, let alone use in an attempt to convince human beings it wasn’t a robot.