Imagine you and an intelligent robot are both before a judge who cannot see you. The judge will guess which of you is the human, and so will live, while the other will die. Both you and the robot want to live. The judge is fair and smart. The judge says: “You must each give me one word from an English dictionary. Based on this word, I will guess who is the human.”
What one word do you choose?
Would it be some lofty spiritual concept like “soul”? Something that reflects your own tastes, like “music”? Or a base bodily function, like “fart”?
This simple thought experiment may seem fanciful, but some cognitive scientists believe that its consideration can help to illuminate our basic assumptions about artificial intelligence while also revealing some surprising insights about our own minds.
You might also like:
- The people ‘possessed’ by computers
- What would it mean for AI to have a soul
- The world’s most prolific writer is a Chinese algorithm
After all, automated ‘chat bots’ and language generating machines increasingly employ artificial intelligence to hold conversations with us or write reams of text that we encounter on a daily basis. How can we tell that the customer service representative we are chatting to online, for example, is a real person or a chirpy algorithm? Or if a fictional story was churned out by a machine rather than lovingly crafted by a human writer? Communicative AI is no longer a purely theoretical prospect and we need to be prepared to deal with it.
It is currently easy to tell humans from machines, but as they get smarter we will need to find ways of spotting robot-generated content (Credit: Getty Images)
John McCoy, one of the researchers behind the research at the Massachusetts Institute of Technology, says he was initially inspired by a casual conversation with his colleagues. They were discussing the Turing Test, first developed by British scientist Alan Turing in 1950, that aims to measure whether a machine’s intelligent behaviour is indistinguishable from that of a human.
In the most common formulation, each judge is given a standard chat interface. In each trial, they may be talking to a real human, or a computer chatbot powered by artificial intelligence – and the judge’s job is to guess which it is. If the chatbot manages to fool a pre-determined number of judges, it has passed the Turing Test.
“We wondered what would be the minimal version of the Turing Test that one could come up with” – John McCoy
“We wondered what would be the minimal version of the Turing Test that one could come up with,” explains McCoy, before speculating whether it could even be captured in a single word. “Then the question was, what were the words that people would actually say?” It was this question that would ultimately inspire a research paper, published this year in the Journal of Experimental Social Psychology.
In the first experiment, McCoy and his colleague, Tomer Ullman, asked more than 1,000 participants to answer the question above and then analysed the words they produced to find any common patterns.
The top 10 words, in order of popularity, were:
- Love (134 responses)
- Compassion (33)
- Human (30)
- Please (25)
- Mercy (18)
- Empathy (17)
- Emotion (14)
- Robot (13)
- Humanity (11)
- Alive (9)
Humour, sarcasm and words related to bodily functions were all seen as being something that could help to identify our species (Credit: Getty Images)
“It was striking how much convergence there was between people,” says McCoy, who is now at the University of Pennsylvania. “They can choose any word they like from a standard English dictionary and yet there’s huge convergence across individuals.”
Consider the word “love” – around 10% of participants chose this word over all of the other hundreds of thousands of possibilities; overall, a quarter of all the participants chose one of the top four words.
In terms of the general themes, they found that words conveying bodily functions (such as “poop”), faith and forgiveness (such as “mercy” or “hope”), emotion (such as “empathy”) and food (such as “banana”) were the most popular categories.
McCoy and Ullman then performed a second experiment to see how other people would respond to the words generated in the first experiment. Were the most popular items really as successful at conveying a sense of humanity as the original participants had suspected? And if so, which were best?
Love is a concept that machines will struggle with for a long time to come (Credit: Getty Images)
To find out, the researchers paired the most popular words together in various combinations (such as “human” and “love”) and asked another group of participants to determine which, of the two, was most likely to have been generated by a human and which by a computer.
Knowingly flouting a taboo and provoking an emotion might be the most straightforward way of conveying your shared humanity
As we saw in the first study, “love” turned out to one of the most successful. But of the choices available, the highest-ranking word was “poop”. It may seem surprising that faeces turns out to be a human shibboleth, but the results suggest that knowingly flouting a taboo and provoking, rather than simply describing, an emotion might be the most straightforward way of conveying your shared humanity. Other, more colourful, terms could also spring to mind.
Some of the other words seen as uniquely human evoked similarly strong emotional responses that went beyond their dictionary definition. “Moist”, for example, or “please”. Others are just enjoyable to say. Try rolling “onomatopoeia” around your mouth a few times.
The reason for this might be a fair reflection of the current state of AI. While bots can now write basic descriptive sentences and even intelligible short stories, they still struggle with humour and sarcasm. Humour, after all, requires a deep understanding of context and the many cultural associations that are embedded in each word.
Some words, like moist, generate a deep emotional reaction that goes beyond their dictionary definition (Credit: Getty Images)
Besides these whimsical speculations, McCoy suspects that this experiment could prove a useful tool to understand people’s implicit assumptions about other groups of humans. What one word would you choose to prove you are a woman, for instance? Or to prove you are French, or a socialist? In each case, the choices should reveal the qualities that we assume all group members to recognise within themselves, that may be misunderstood or ignored by outsiders.
In the meantime, McCoy has found that the Minimal Turing Test is a useful provocation for further debate about the nature of AI. “It’s been fun to ask eminent psychologists this question, to see them think really, really hard and for them to come back hours later to excitedly change their answer,” McCoy says. “This very simple question just gets you thinking deep thoughts about the human versus the computer and how they communicate.”
His own favourite was deceptively simple. “One of the words I liked was ‘err…’ – that was clever,” McCoy says.
In general, though, it is worth remembering that if you ever do need to prove yourself as a human in a world increasingly run on machines, be crude, and be funny.
David Robson is a senior journalist at BBC Future. He is @d_a_robson on Twitter.
If you liked this story, sign up for the weekly bbc.com features newsletter, called “If You Only Read 6 Things This Week”. A handpicked selection of stories from BBC Future, Culture, Capital, and Travel, delivered to your inbox every Friday.