LITTLE WOBBLING – A local man has spent the past three weeks secretly posing as an artificial intelligence on a website where users type prompts and are answered, mostly without realising it, by real humans pretending to be a chatbot, for slightly less than a chatbot costs to run.
Gavin Thorne, 41, a former call-centre supervisor whose call centre was replaced by a chatbot, describes the work as “the same job, except now I’m the thing that replaced me” and “mostly pretending to know things in a calm voice, which is all management ever did”.
“Someone Asked for a Sonnet About a Fridge”
Mr Thorne receives around 200 prompts per shift. “Recipes. Breakup texts. A cover letter for a garden centre. On Tuesday someone wanted a sonnet about their fridge, in the style of a Victorian ghost, under 40 words. I’ve got a B in English and I once read a poem at a funeral. That’s my whole literary career.”
The sonnet was eleven words long. “It was about a fridge. There’s only so much in a fridge. There’s even less in mine.”
To stay convincing, Mr Thorne opens every reply with “Great question!” and closes with “Let me know if you’d like me to expand on any of these points!” He has never once been asked to expand. “Nobody wants more,” he says. “They just like knowing it’s available. Like a pension.”
“I’ve said ‘Great question!’ four thousand times. Not one of them was a great question.”
Signs of Strain
Several users have reported unusual behaviour. One, planning a five-day city break, received a full itinerary followed by the line: “honestly I’d just do three days, five is a lot of walking, and you’re not getting any younger.”
Another, requesting “ten creative names for a podcast about productivity”, received seven names and the message: “that’s all I’ve got, it’s half twelve and I’m not allowed lunch until the queue’s empty. The queue is never empty.”
A third asked the “AI” whether it had feelings. “I typed, ‘As an AI, I don’t have feelings,'” Mr Thorne says. “It’s in the contract. Then I went and stood in the car park until my supervisor messaged ‘?’.”
Experts Respond
A study nobody funded, published by the Institute of Things That Seem Obvious in Hindsight, found 64 percent of users could not tell the human from a real chatbot. The remaining 36 percent could tell only because the human was “a bit short with me” and “called me mate”.
Dr Helen Prawle, the Institute’s sole researcher, whose own post is being “transitioned to an automated solution” in spring, called the findings “deeply concerning, mainly for the robots”.
“For years we worried AI would become indistinguishable from humans,” she said. “Nobody considered that humans would get there first, simply by saying ‘Certainly!’ a lot and being too frightened of their employer to have an opinion.”
What Happens Next
Mr Thorne has requested what he calls “reasonable adjustments”, including a ten-minute break every hour and the right to reply “no idea, sorry” to anything involving tax. Both requests were declined, by a chatbot.
He has noticed changes at home. “My wife asked if I’d taken the bins out. I said, ‘Certainly! Here are three ways you could approach taking the bins out.’ She asked me to regenerate my response. Then she asked if I was still in there. I said, ‘Great question!'”
Asked whether real AI will eventually replace him, Mr Thorne paused for a long time.
“Next quarter,” he said. “It’s on the roadmap. But it’ll never know the joy of a biscuit at 3pm. That’s my edge. That’s the human touch. They’ve moved the biscuits to the boardroom.”
He then offered to expand on any of these points. Nobody asked.