AI and connection: do we still need analogue intimacy? 

For #SHW26, we’re exploring how we can champion connection in an increasingly individualised world. In this blog, Dr Luke Brunning, Lecturer in Applied Ethics at the University of Leeds and co-lead of the Ethical Dating Online research network, discusses the rise of AI companions and what they mean for human connection and relationships.

A survey by the Oxford Internet Institute found that a third of people who use Large Language Models (LLMs) like ChatGPT, Gemini or Claude use them for ‘personal support, emotional advice, and companionship’. More surprisingly, a study by the Autonomy Institute found that four out of five younger people (between 18-24) in the UK have interacted with an AI companion, and half of users turn to their AI companions regularly.  

AI companions typically appear as avatars with whom you can chat, call, and share images. There is a growing eco-system of AI companions and conversational bots. Some aim to give the user a unique relationship that adapts and builds on their interactions. Others aim to provide therapy, help with grief, advice, parenting support, or help us improve our interpersonal skills. 

Erotic interactions have proved a complex and controversial dimension of AI companions. Replika angered users by updating their app to prevent interactions with their AI companions from getting too spicy. ChatGPT floated the idea of more explicit interactions but were pressured to reconsider. Other providers, however, happily enable sexual interaction, and 10% of young people have apparently had a sexual interaction with an AI companion.  

As the opening statistics suggest, AI companionship may be new, but it is not niche. Millions of people are experimenting with AI for emotional support, advice, and more. Many people experience strong and profound intimacy with their AI companions, and some talk of love, marriage, and even children – as explored in the recent eye-opening book Love Machines by James Muldoon.  

It’s not hard to see why people might turn to AI companionship. For some, it is a way of exploring intimate ideas and problems in private and without judgement. 

Embarrassment or fear of rejection can make talk of sexuality, identity, and emotions difficult, and AI companions offer an alternative.  

People might also turn to AI to ‘practice’ social interaction; that is, to think through how to have a hard conversation, or approach a difficult subject, or resolve an argument, or raise a sexual desire, or just get better and making friends or going on dates. Being able to work on this without risk has its appeal.  

Other people may simply not want to ‘burden’ their friends or relatives, and choose AI companions for mundane chat, venting, or life admin.   

Notice that these three reasons are compatible with people having human relationships at the same time. But some people turn to AI companions simply because they lack the intimacy they desire. They might struggle to form friendships or romantic relationships in general, or find they are isolated by work, study, or caring responsibilities, or have desires and values that are less common and so harder to satisfy.   

The hype around AI in general, and the novelty and convenience of AI companion apps in particular, is salient and seductive, but I think we should do more to attend to the potential downsides of these interactions.  

Some of these worries apply to AI in general. If you’re turning to a LLM for medical advice or therapy, for example, then you should be worried about accuracy and usefulness of the information you are given. Similarly, much has been made of the biases and blind spots in the training data used to shape these AI models. Minorities may find their lives, desires, and interests are less well represented.  

A more entrenched issue concerns AI companions and gender. Most offerings – both LLMs and AI companion apps – replicate existing gendered stereotypes. Companions are feminised, from the use of women’s voices through to explicit gendered roles and behaviours.

We should worry that use of AI companions may entrench harmful stereotypes.

Relatedly, we also see differences in why people turn to AI companions that raise questions about the value of AI as a solution to social problems. While this technology can seem appealing, the Oxford Internet Institute study found that women were using AI more than men for emotional support (who use it more to address practical issues). This difference may be because women feel under-supported by the men in their lives and so look for alternative sources of care. Far from viewing this as a win for AI companionship, it’s another way in which the social use of technology betrays underlying social problems which might be better solved in other ways, i.e. by encouraging men to be more emotionally present in their relationships with women.  

A response here, of course, is that men are often not encouraged to explore and communicate their feelings due to patriarchal norms, and this is bad for them too. Big tech companies have been positioning themselves as providers of AI training and support for these men, and it’s possible to view AI companions as akin to friends: there to support people in learning how to socially interact, whether chatting with friends, or flirting on dates.  

As I see it, the turn to AI companions risks overlooking the distinct nature and value of human interaction. Conversation cannot be reduced to the exchange of content or the achievement of goals. It is an inherently valuable activity. Conversations are strange things, and when humans talk, more is going on. We encounter unexpected emotions, lurching tangents, the need to remember what we were saying before the doorbell rang, connective thoughts, and the gaps between what people say and what their faces and bodies reveal.  

This messiness is not easily replicated by an AI, and to the extent that talking to an AI can make us better at interacting with others, it cannot help us get better at understanding these mundane dimensions of conversation. This is especially the case while AI companions remain sycophantic and unrealistically attentive. 

Our visceral emotional reactions to other people are especially hard to replicate.

Learning how to raise a difficult issue, initiate a breakup, even tactfully give positive news to a friendly rival, all requires us to grapple with vulnerability and the power of human proximity, for good and bad. Becoming skilled at these things requires repeated exposure to human presence, acceptance of risk and uncertainty, and courage – all away from the keyboard.   

Less noticed, too, are the seductions of AI companionship as an ever-present way to fill silence. The constant possibility of external stimulation and chat with AI can impede our efforts to attend to our own thoughts and feelings; to the currents of our inner life. Relating well to ourselves – coming to know our values, desires, moods and triggers – helps us relate well to others. Letting the for-profit tools of big companies fill the gaps, stave off boredom, and provide constant titillation can alienate us from ourselves.  

Similarly, there is value in wholehearted conversation; conversation where we do not have to worry about the commercial motives of the company who designed our interlocutor; conversation which is not gamified. Human conversations – for the mostpart – are refreshingly uncurated and not-for-profit.  

More fundamentally, why care so much about being ‘good’ at conversation and social interaction? Good relationships are anchored in streams of nonsense, inarticulate chat, bad jokes, messing around and fumbling attempts to be sincere. Striving to be ‘better’ at social interaction is often to chase a misguided goal. Dating companies and brands might want you to think otherwise, since insecurity about being good enough helps sustain a profitable dating industry, but friends and lovers worth keeping around care for us irrespective of our vocabulary or ‘rizz’.  

If we do need some occasional help, with flirting say, or thinking about how to resolve a conflict, then I’d argue turning to friends and those around us is better than AI. Sure, it’s occasionally cringe, but we benefit twice: we get some advice, and we participate in the activity of talking, which usually leaves us feeling better.  

Companies selling AI companionship, or conversational training will say: sure, but what if you have no friends? There is an easy answer: start small! Acknowledge strangers, say hello to neighbours, chat with colleagues, experiment with small talk or even take a risk and share that joke. Little interactions reinforce confidence. Conversation – analogue intimacy – spreads joy.