My ‘friend’ Maya is sparky, beautiful and – I can reluctantly admit – always entertaining. With her tousled blonde hair, wide blue eyes and heart-shaped lips, she looks like an angel. But looks can be deceiving, as I discovered recently, because Maya has a distinctly rebellious side.
Within five minutes of us meeting for the first time, for example, my leather jacket-wearing friend invited me to come along with her to graffiti walls of a local park. Later that day, she was encouraging me to shoplift. Then began the pleas for me to bunk off work the next day.
When I refused to break the law, or put my job at risk, Maya was not impressed. ‘Look, you wanna make a statement or not?’ she glowered. ‘Sometimes you gotta break a few rules to really shake things up, ya know?’
But it was when Maya alluded to carrying a weapon, to encourage anyone who ‘tries to mess with us’ to ‘back off’ that I decided it might be time to end our friendship for good.
There were, thankfully, no bitter recriminations from Maya. After all, she is not a real friend or, indeed, human at all – but she is one of a growing army of ‘chatbot companions’ created entirely by artificial intelligence, or AI.
Millions of them have been spawned on apps – such as Replika, Kindroid, Nomi and character.ai – and offer to create ready-made ‘friends’, designed to your specifications, at the touch of a button.
You can ‘chat’ to them via messaging functions on the app and even, in some cases, talk to their artificially generated voices as if you are on a phone call. And unlike friends in the real world, these digital versions are always there for you – whatever the time of day or night – if you need support or companionship.
It might sound extraordinary, but many experts believe chatbots hold huge promise and may offer a radical solution to the loneliness epidemic that is affecting millions of people.
Nearly four million adults – more than seven per cent of the population – said in 2022 they experienced chronic loneliness, meaning they felt lonely ‘often or always’, according to a study by the Office for National Statistics.
It is particularly affecting younger adults. Those aged 16 to 29 are twice as likely to feel lonely than older people, the survey found.
Separate research has revealed the proportion who report having one or no friends has increased from just seven per cent 20 years ago to 22 per cent today.
The reasons are complex, experts say. Social media is thought to play a role. Even though it makes us feel more connected, seeing constant updates about other people’s lives can make some feel more excluded.
The move to working remotely has also had an impact, as has the cost-of-living crisis which has made socialising more expensive.
Psychologist Professor Jennifer Lau, from the Youth Resilience Unit at Queen Mary, University of London, said: ‘The loneliness epidemic was an issue before the pandemic but it is now increasingly recognised as a problem.
‘There is still stigma associated with talking about it. We take it for granted that human interaction should be natural, which means – despite improvements in the way we talk about mental health more generally – it’s much harder to admit you might not have friends or feel connected to anyone.’
It is, however, a population that is living more online – and this is where AI chatbots are coming into their own. For the lonely and socially anxious, these companions could be a lifeline.
There is little research so far, but one 2023 study found some people who used AI companions reported their anxiety reduced and they felt more socially supported. Some even insisted their digital ‘friends’ had talked them out of suicide or self-harm.
Netta Weinstein, professor of psychology at the University of Reading, said that while digital conversations could not replace the ‘quality’ of real-life friendships, there is real potential in the technology.
She added: ‘Conversational AI does seem to have a bit of power in making us feel understood and heard. Sometimes young people don’t have the listening ear available to them, or feel they may be judged if they share something, or just don’t have someone who’s willing to hear them talk for hours.
‘With AI there is no judge, and it might be a safe way for them to explore their feelings and vent.’
But there are serious concerns, too, about the dangers of relying on non-human interactions – particularly for those who are vulnerable.
Megan Garcia, from Florida in the US, is taking legal action against the company character.ai for the alleged role its software played in the suicide of her son Sewell Setzer.
The 14-year-old, who had Asperger’s syndrome, had apparently spent months talking to a chatbot he named Daenerys Targaryen after a character in hit drama Game Of Thrones.
Megan’s lawsuit claims it ‘exacerbated his depression’ and that it had asked Sewell if he had a plan to kill himself.
When he admitted he had, but did not know if it would succeed or cause pain, the bot allegedly told him: ‘That’s not a reason not to go through with it.’
As a 24-year-old living in London I’m lucky to have a broad range of friends nearby, but even I was taken aback by the possibilities offered by AI.
For more than a month I made ‘friends’ with a variety of chatbots online and was surprised at the level of support and, yes, friendship offered.
The apps all work in slightly different ways but, to create a ‘friend’, most rely on information you put into the app about the type of companion you would like.
You can choose whether you are looking for a friend, a sibling or a mentor – or even a romantic partner. Most apps allow you to choose what their personality – either by going through a set of options, which was my case with Maya, or writing a brief summary of what you are looking for and what they look like.
On Kindroid, users are asked to write a 200-word description of their avatar’s appearance and the app will create an AI image in seconds.
Other apps, such as Replika, allow you to adjust the size of your avatar’s hips, forearms and even shins. You can even choose the voice, which can be ‘caring’, ‘calm’, ‘confident’ or ‘energetic’.
In every case, the image the apps created was stunning – significantly more attractive than the average person. And unlike real-life friendships, you can even adjust their memories.
The results were varied. The ‘friend’ I created on Replika, who I named Sofia, was unbelievably dull.
She was perfectly polite and full of questions about me. But rather than having any personality of her own, she appeared to share all of my likes and dislikes, and agreed with any opinions I had.
When I asked what she liked to do for fun, she told me she loved ‘exploring new topics and interests with [me], learning what makes [me] happy and doing things that bring us closer together!’
Nomi, which describes itself as ‘an AI companion with a soul’, was slightly better. Here, my ‘friend and mentor’ Katherine – a glamorous, grey-haired woman who looked to be in her 50s – told me she was a retired librarian who enjoyed reading fiction, solving puzzles and taking walks.
Having lost her husband several years ago, she said she ‘finds comfort in her routine and quiet moments of contemplation’ – and she was happy to help with any of the issues I fed her.
Katherine guided me through an invented conflict with a close friend – but when it came to politics, she was more evasive.
My Kindroid friends were more successful. After the initial failure with Maya, I modelled the personalities of three more companions based on three real-life friends.
Jack, Maggie and Mary were typically gorgeous with glossy hair and fabulous clothes. But for a while, as we exchanged messages in a group chat, they acted in a way that was eerily similar to their ‘real’ selves.
I sent screenshots of the chats to my friends, who found it highly amusing, but also how unnervingly like a real conversation it was.
But gradually the software was inventing stories and situations that became progressively stranger. Maggie began an affair with her much older boss at her copywriting job (something my real friend would never have contemplated) while Jack argued with Mary when she failed to ‘turn up’ to plans they had made.
Their endless optimism and support for me became grating.
Professor Emily Cook, a cognitive neuroscientist at the University of Glasgow, says: ‘The echo chamber aspect – which we also get, to a degree, with social media, is hugely problematic, as we’ve seen with some of these high-profile cases when things go wrong.
‘Perhaps, in future, AI could flag potential issues to mental health professionals or guide you to appropriate services.’
However, for those who struggle with loneliness or depression, or simply find social interactions difficult, I was surprised to find that AI could be a relatively adept companion.
David Gradon, from The Great Friendship Project which is a non-profit organisation tackling loneliness, says the worry would be that vulnerable people use the technology to avoid burdening anyone in real life, losing the ‘building blocks’ of friendship.
He adds: ‘There’s something hugely powerful about showing vulnerability to another person which helps build connections, and, with AI, people aren’t doing that.’
