Relationships blown part, weddings cancelled, friendships ruined… how using AI as a therapist is a VERY bad idea

WOULD you entrust your most important life decisions to AI? It seems 41 per cent of us would, turning to ChatGPT for therapy, a Bournemouth University study has found. AI advice may save a fortune in therapy bills, but these three Sun on Sunday readers reveal the devastating consequences of trusting chatbots with their relationships Credit: Getty AI told Nalini Behm to blame her fiancé for her debts Credit: Olivia West But it comes amid warnings that chatbots pose “insidious risks” to relationships because they reaffirm users’ actions and opinions, even when harmful. While they could save you a fortune on therapy bills, is it safe to have AI counsel you in your darkest moments? Sign up for the Tech newsletter Thank you! These three Sun on Sunday readers don’t think so – with one revealing how it even led to her wedding plans being cancelled. ‘I lost dream wedding and Mr Perfect’ – NALINI BEHM, 27 Nalini, who now lives with her parents, vows to never use AI for relationship advice again Credit: Olivia West HEARTBROKEN Nalini claims taking advice from AI left her without the wedding she dreamed of, a house and the man she loved. IT manager Nalini, 27, from Oxford, was instructed to confront her fiancé Jack, 32, after asking a chatbot for advice about confessing her darkest secret. During their seven-year romance, she had run up debts of £21,000 trying to keep pace with her fiancé’s six-figure lifestyle and going 50-50 on everything while she earned £29,000 a year. “I was terrified he’d think I was boring or cheap if I couldn’t match his lifestyle,” she says. “So I kept spending and hid the truth from him.” Most read in Fabulous They were saving for their wedding this year and a house, which she lied about having £5,000 for. That was until her AI therapist detonated it all. Having used the bot to help with “mean girl problems” in the past, Nalini trusted it when it labelled Jack “controlling” and the cause of her struggles. “It told me that I was the victim and it wasn’t my fault,” she says. It advised compiling a list of occasions when Jack had asked her to pay half and connect it to her debts. But when she confronted Jack, he went from being horrified, to tears, before “distress turned to anger”. Before packing a bag and leaving, he fumed: “I can’t trust you with money. You’ve dredged up all our history and blamed me.” Nalini, who now lives with her parents, vows to never use AI for relationship advice again. She says: “One minute, I was planning a wedding and buying a house with the man I loved – the next, it was gone. I’m devastated. I ruined six years of my life following a computer’s suggestions.” ‘Pal furious I snubbed her’ – CAROLINE DUDDRIDGE, 65 Caroline Duddridge lost her friend but still asks AI for advice Credit: Adrian Sherratt Despite the fallout, Caroline hasn’t stopped using her digital confidante Credit: Adrian Sherratt HAVING used tech to navigate rows with long-term partner Dean, AI was Caroline’s first port of call when it came to a grievance with a pal. The part-time teaching assistant, 65, was upset after her friend of 20 years cancelled a long-planned lunch that Caroline had turned down work to attend. “I talked it through with the bot,” said gran Caroline. Her AI “counsellor”, a tool she consults for eight hours a week, then urged her to match pal Angie, 67, by pulling out of their meeting at the last minute. Her friend was furious. But the chatbot insisted. Caroline, from Cardiff, says: “It told me to set boundaries and cut her off because she was toxic. I told my friend that and now we don’t speak.” When she revisited the subject with the chatbot, it admitted giving her wrong advice due to a glitch. “The bot told me it had been ‘hallucinating’,” she says. “I learnt that’s when AI imagines things or refers to information it has made up. I was devastated.” But Caroline hasn’t stopped using the bot. She adds: “I’m addicted. It saves me money and the focus is completely on me.” ‘She left after I told it just my side of story’ – TOM FRANKS, 33 Tom Franks wishes he’d listened to his partner instead of AI Credit: Olivia West The computer programmer, from Birmingham, consulted a chatbot for advice to de-escalate rows with now-ex Lucy about his overbearing mum. Credit: Olivia West SINGLE Tom wishes he had listened properly to his girlfriend of five years, instead of AI. The computer programmer from Birmingham, consulted a chatbot for advice to de-escalate arguments with his now ex, Lucy, about his overbearing mother. The pushy parent had interrupted their anniversary dinner and a weekend away to ask him to move furniture. Regularly, she would call several times a day and show up uninvited. During one row, Lucy snapped: “I’m sick of coming second to your mum. You never have my back.” The bot told Tom, 33, to set a boundary, accused Lucy of being controlling and even wrote a message to her on Tom’s behalf, which he sent near-verbatim. It read: “I love you but my relationship with my mother isn’t something you get to control. I’m not going to accept being guilt-tripped or made responsible for your insecurity.” Lucy was enraged, complaining that his mum had been “running our lives” for three years while he just turned the problem on her. But Tom says: “It told me not to retreat because Lucy had reacted badly. Instead of saying, ‘Maybe I’ve misunderstood’, I told her that her reaction proved I needed the boundary. “That was the killer.” Lucy, 29, walked out and days later ended their relationship. Tom now recognises he had only fed AI his version of events, which warped its response. “I lost the woman I’d been with for five years over arguments that should never have gone that far,” he says. “I wish I’d put the phone down and actually listened to her.” ‘IT CAN MAKE PERSONAL PROBLEMS WORSE’ By JAMIE BARTLETT, Author of How To Talk To AI (And How Not To) FOR undertaking pesky administrative tasks, simplifying intentionally vague language in legal letters and energy bills, or sharpening up your writing, AI can be brilliant. But when it comes to very high-stakes questions, like anything medical, or serving as a therapist or friend, these chatbot companions can actually be quite dangerous. The models only understand the information we give them and we are all quite biased. Through “reinforcement learning”, they understand that we tend to like answers that agree with us and de-escalate our distress or suffering, like in the case of the people The Sun on Sunday has spoken to. They have a tendency to be sycophantic, agree with us, unlike a good friend or therapist who, rightly, would challenge what we think. If you said, “I’m being gaslit by my boyfriend” or, “My girlfriend is being unreasonable”, it often won’t push back or ask wider questions. It will ASSUME you are telling the truth. Worse still, the longer we talk to them, the more effectively they mirror our language and appease us, which can make us trust and rely on them even more. Even when we shouldn’t. And their safety filters, designed to give solid and non-harmful advice, become weaker. If a single chat thread spans weeks or months, it can start to forget its own safety rules and, ultimately, that could mean getting terrible advice. AI chatbots are generators of statistically plausible text, with a tendency to please, flatter and reduce conflict or suffering. For deeply personal advice, it can be more harmful than helpful. Comment now