More on “Chatfishing” on Dating Apps and in Texting

I blogged previously about the role and ethics of AI use in the online dating communication context. The Guardian published over the weekend a piece discussing the proliferation of this practice. One of the problems it highlights is the mismatch that individuals encounter between who they thought they were texting and the person that shows up on a date and is significantly less articulate or attuned. In that sense and on average, it likely makes online dating (as well as getting to know each other via texting generally) an even less efficient process than already.

Some of the uses of AI mentioned in the interviews conducted for the Guardian article lean toward the comical, such as this one:

As 32-year-old Rich points out, though, “it’s not like using ChatGPT guarantees success”. When he met someone in a bar one Friday night and swapped social media handles, he asked AI what his next move should be. ChatGPT discerned that sending an initial message on Monday midmorning would set the right pace. “Then it gave me some options for what the message could be,” says Rich. “Keep it light, warm, and low-stakes so it reads as genuine interest without urgency,” the bot advised. “Something like: Hey Sarah, still laughing about [tiny shared moment/reference if you’ve got one] – good to meet you!” Rich went back and forth with ChatGPT until he felt they’d hit upon exactly the right message (“Hey Sarah, it was lovely to meet you”) but sadly she never replied, he says. “It’s been two weeks now.”

Rather shocking that such a witty line wasn’t an instant winner (though to be fair, whatever happened before that probably left a negative or lukewarm impression enough not to inspire desire for another meeting likely couldn’t be overcome by ChatGPT anyway…).

Some other AI uses, however, bring up heavier subjects, such as here:

Still, there was one date that pricked his conscience. He was doing the usual copy-and-paste, letting ChatGPT do the heavy lifting, “when a girl started talking about how she’d had a bereavement in her family”. ChatGPT navigated her grief with composure, synthesising the kind of sympathy that made Jamil seem like a model of emotional literacy. “It said something like, ‘I’m so sorry you’re going through this, it must be really difficult – thank you for trusting me with it,'” Jamil recalls. When he met the girl in real life, she noted how supportive he’d been in his messages. “I felt bad – I think that was the only time I thought it was kind of dishonest. I didn’t tell her I’d used ChatGPT but I really tried to message her myself after that.”

In this kind of setting, ethically speaking, motive matters. Was Jamil mainly being lazy, manipulative, or just insecure about how to approach the situation and thought ChatGPT would help him to do right by his interlocutor’s grief? There’s no way to know from a brief journalistic set of quotations, but it brings us closer to one of the central guidelines about when use of AI may be acceptable.

At the heart of it, it may come down to the Platinum Rule, which is to treat others the way one believes that they would want to be treated. And in a situation of bereavement, most people would probably not find it acceptable for someone to use AI out of laziness but would at least tolerate it if it was done in a good-faith attempt to comfort in an appropriate tone. Whether the behavior fell into column A versus B is likely to reveal itself once in-person interactions begin or intensify. It is fair to say, however, that the existence of modern AI tools has made it more key than ever to place a lot less stock in what people (now potentially more assisted by technology than previously) say as opposed to what they do.