Robert Epstein was looking for love. The year being 2006, he was looking online.
Epstein was disappointed – he wanted more than a penfriend, let’s be frank – but she was warm and friendly. Soon she confessed she was developing a crush on him.
“I have very special feelings about you. In the same way as the beautiful flower blossoming in mine soul… I only cannot explain… I shall wait your answer, holding my fingers have crossed…”
The correspondence blossomed, but it took a long while for him to notice that Ivana never really responded directly to his questions.
She would write about taking a walk in the park, having conversations with her mother, and repeat sweet nothings about how much she liked him.
Suspicious, he eventually sent Ivana a line of pure bang-on-the-keyboard gibberish. She responded with another email about her mother.
At last, Epstein realised the truth: Ivana was a chatbot.
What makes the story surprising is not that a Russian chatbot managed to trick a lonely middle-aged Californian man.
In other words, one of the world’s top chatbot experts had spent two months trying to seduce a computer program.
In Turing’s “imitation game”, a judge would communicate through a teleprompter with a human and a computer. The computer’s job was to imitate human conversation convincingly enough to persuade the judge.
Turing thought that within 50 years, computers would be able to fool 30% of human judges after five minutes of conversation.
Like Ivana, Goostman managed expectations by claiming not to be a native English speaker. He said he was a 13-year-old kid from Odessa in Ukraine.
One of the first and most famous early chatbots, Eliza, would not have passed the Turing Test – but did, with just a few lines of code, successfully imitate a human non-directional therapist.
Named after Eliza Doolittle, the unworldly heroine of George Bernard Shaw’s Pygmalion, she – it? – was programmed in the mid-1960s by Joseph Weizenbaum.
If you typed, “my husband made me come here”, Eliza might simply reply, “your husband made you come here”. If you mentioned feeling angry, Eliza might ask, “do you think coming here will help you not to feel angry?”. Or she might simply say, “please go on”.
People did not care that Eliza was not human: they seemed pleased that someone would listen to them without judgement or trying to sleep with them.
Weizenbaum’s secretary famously asked him to leave the room so that she could talk to Eliza in private.
More things that made the modern economy:
Psychotherapists were fascinated.
A contemporary article in The Journal of Nervous and Mental Disease mused that “several hundred patients an hour could be handled by a computer system”. Supervising an army of bots, the human therapist would be far more efficient.
And indeed, cognitive behavioural therapy is now administered by chatbots, such as Woebot, designed by a clinical psychologist, Alison Darcy. There is no pretence that they are human.
Weizenbaum himself was horrified by the idea that people would settle for so poor a substitute for human interaction. But like Mary Shelley’s Dr Frankenstein, he had created something beyond his control.
Chatbots are now ubiquitous, handling a growing number of complaints and enquiries.
Amelia talks directly to the customers of some banks, but is used by US company Allstate Insurance to provide information to the call centre workers which they use while talking to customers.
And voice-controlled programmes like Amazon’s Alexa, Apple’s Siri and Google’s Assistant interpret our requests and speak back, with the simple goal of sparing us from stabbing clumsily at tiny screens.
Brian Christian, author of The Most Human Human, a book about the Turing test, points out that most modern chatbots do not even try to pass it.
It seems we are less likely to notice a chatbot is not human when it plugs directly into our libido.
Another tactic is to wind us up. The MGonz chatbot tricks people by starting an exchange of insults. Politics – perhaps most notoriously the 2016 US election campaign – is well-seasoned with social media chatbots pretending to be outraged citizens, tweeting lies and insulting memes.
But generally chatbots are happy to present as chatbots. Seeming human is hard.
Commercial bots have largely ignored that challenge, and instead specialise in doing small tasks well – solving straightforward problems, and passing on the complex cases to a real person.
The economist Adam Smith explained in the late 1700s that productivity is built on a process of dividing up labour into small specialised tasks. Modern chatbots work on the same principle.
That is what we observe, for example, with the digital spreadsheet, the cash machine or the self-checkout kiosk. Chatbots give us another example.
But we must be wary of the risk that as consumers or producers – and perhaps even as ordinary citizens – we contort ourselves to fit the computers.
We use the self-checkout, even though a chat with a shop assistant might lift our mood.
We post status updates – or just click an emoji – that are filtered by social media algorithms; as with Eliza, we are settling for the feeling that someone is listening.
Christian argues that we humans should view this as a challenge to raise our game. Let the computers take over the call centres. Is that not better than forcing a robot made of flesh-and-blood to stick to a script, frustrating everyone involved?
We might hope that rather than trying or failing to fool humans, better chatbots will save time for everyone – freeing us up to talk more meaningfully to each other for real.