I’m standing in the pouring rain trying to retrieve a parcel from a locker wall that just won’t open. I call the help desk and am greeted by Matt, a customer service chatbot. To get my parcel, I must follow his instructions, describing the problem and speaking in full sentences. But it’s not working, so I ask for a human. Matt asks for my phone number. I give it to him. He says ‘Thank you’ and asks how he can help, again. Minutes pass. The rain is unrelenting, as is the bot’s cheeriness. When the locker finally clicks open, I hang up while he’s still speaking.
A week later, I find myself still thinking about that exchange – about my irritation, and about hanging up on him like that. My discomfort about how I treated the bot might not be unusual. In 2003, the media researchers Byron Reeves and Clifford Nass documented what they called the ‘media equation’: people apply genuine social reflexes to computers automatically, even when we know they aren’t conscious. Some of us say ‘Please’ and ‘Thank you’ to chatbots or feel sorry for robot vacuum cleaners. The researchers’ explanation is not that we’re naive, exactly. Our brains process a voice – any voice – as a social presence, triggering social responses before conscious thought has had a chance to intervene.
But this social reflex isn’t always kind. When people talk to chatbots, we often seem to strip basic courtesy out of the exchange. Our messages become shorter, simpler and more profane, and in one analysis, more than half of conversations included at least one form of insult or verbal aggression.
Chatbots are increasingly woven into daily life, which means that many of us are likely having these automatic – and frequently brusque – social responses multiple times a day. They are becoming habitual. Habits take root, after all, when a behaviour is repeated in stable contexts and driven by consistent cues. You can see this with other ways of using technology: if you pick up your phone to check a message, only to lose yourself for 10 minutes as you scroll your favourite apps, you probably didn’t deliberate about whether to use those apps; you just did it, because that’s what you usually do when you look at your phone. You’ve developed a habit.
Chatbot interactions are, for habit formation, close to ideal. When you have a question and open an AI-based chatbot app, there is little friction or waiting. Often the reward is nearly instantaneous. The register that many of us use when we get there – direct, brief, transactional – is already familiar from voice assistants: instructing Siri to set a timer, commanding ‘Alexa, stop!’ when the smart speaker gets it wrong. The little chatbot prompt box invites the same: cut the fluff and demand what you need. It seems to get results, too. Some studies on prompt tone suggest the systems themselves often respond better to it. And when the machine fails or frustrates the user, the user gets irritated, or even abandons the conversation mid-sentence, as I did.
For me, and for those who’ve been studying how we relate to artificial beings, it’s hard not to wonder if this increasingly normal mode of interacting will change what we’re like in the rest of our lives.
In a study published last year, the sociologists Ashley Harrell and Margaret Traeger tested whether people’s interactions with artificial agents might spill over into those they have with humans. They conducted two experiments with more than 4,000 participants. In each study, the participants played a simple online cooperation game with what they thought was either a bot or another human being (in reality, these other players were all programmed). After 10 rounds, the participants were paired with a new partner – this time, a real human they had never interacted with before – and played again.
People were more instrumental and demanding toward an AI chatbot than another human
A couple of things emerged from the data. The bot’s behaviour mattered: if participants first interacted with an uncooperative bot, they were less cooperative themselves with the human who followed it, compared with participants who started with a cooperative bot. More to the point, participants who began the game with a bot partner were less cooperative afterward than those who started with a human. The researchers traced this to empathy: people felt less empathy towards a bot partner (less compassion, less curiosity about their point of view), and that didn’t immediately reset when a human partner took the bot’s place. The empathy dial, once it was turned down, stayed down.
The bots in this study were simple: just a programmed image on a screen, not a conversational AI. Yet the effects still showed. In another experiment with an AI chatbot, people who briefly collaborated with a bot were more instrumental and demanding toward their collaborator compared with those who thought they were working with another human – and they later judged other people’s work more harshly. What immersive chatbot systems might do, over time, to how people treat each other is still hard to know, but these are early signs of how spillover can happen. As we contemplate where it goes from here, a different kind of thinking could be helpful.
The ethicist Tae Wan Kim at Carnegie Mellon University has spent years thinking about what we owe to robots and what we owe to ourselves – two things that he considers inseparable. Kim draws on the Confucian concept li, which translates roughly as ‘rite’ or ‘ritual’. In Confucian thought, li refers not only to formal ceremonies but also to everyday forms of interaction such as a greeting or a gesture of gratitude. The philosopher Herbert Fingarette described what this looks like in practice in Confucius: The Secular as Sacred (1972): two people meet on the street, one extends a hand, and the other turns, smiles, reaches out – a moment of ‘spontaneous and perfect cooperative action’ so habitual that people barely notice it. That unremarkable exchange, Kim and his co-author Alan Strudler write, is li. And it matters, because each time someone does it, they become slightly more the kind of person who does it.
As Kim told me, li ‘includes countless everyday routines and forms of speech, and it treats participation in proper ritual as something that shapes people into more civil, mutually respectful agents.’ Exchanges with an AI are becoming another kind of everyday interaction, and, seen this way, every message we send is practice for the kind of people we become.
This takes us beyond the obvious argument that it doesn’t matter how we treat bots because, unlike another person or animal, bots can’t be harmed. The question of how we should treat a dog or a chimpanzee is a question about the animal. With chatbots, the question is about how the ill-treatment affects the human. Kim’s point, and what the cooperation-game research suggests, is that something is at stake for us. If we develop a habit of talking to chatbots in an instrumental, impatient and dismissive register, we are rehearsing a way of interacting with others.
Technology has always shaped the social norms and expectations of its time, often in ways that are noticed only in retrospect. Cinema set standards for romantic love, ones that real life rarely measures up to. The internet made it easier to meet people – and to avoid them. Social media changed the way many of us present ourselves to the world and how we feel about our own lives. These effects followed from specific design choices, business incentives and the habits people settled into. Chatbots are the latest instance in this pattern.
The instrumental register suits the companies building these systems just fine: short exchanges are convenient for users and cheaper for AI companies to process and, in this business, greater efficiency means greater profits. That’s one reason the tone could readily spread. Many millions of people are now regularly using chatbots and rehearsing it.
Enough people rehearsing the same instrumental register multiple times a day could influence what ordinary conversation sounds like
The sociologist Erving Goffman used the term ‘interaction order’ to describe the unwritten rules that govern how people behave with each other in everyday life. This includes things like when to speak and pause, how much warmth and patience to show. These rules of everyday interaction can feel so natural we don’t realise they exist until they start to change.
Think of how potentially awkward conversations are often moved to text or email, or how read receipts have imbued a recipient’s silence with meaning. Nobody decided one day that that’s the way things would be. It just came to be part of how many people interact. The concern here is similar: that enough people rehearsing the same instrumental register multiple times a day could influence what ordinary conversation sounds like and how much patience we bring to exchanges that don’t go smoothly. What do we carry over to the call-centre worker who can’t fix our problem; the colleague who misunderstands us; the partner who needs us to say it again?
But the direction of change is not necessarily fixed. As Kim put it: ‘chatbot design can also rehearse patience, warmth, and civility.’ One study found that a bot that met abuse with expressions of empathy – rather than avoidance or retaliation – left people feeling less angry and more guilty. A chatbot that pushes back gently or redirects hostility is doing something different from one that is optimised for speed or engagement. The interaction feels different, and it may leave the user slightly different too. The texture of these interactions is, in part, a design choice, one that the people building these systems are making right now. The rest is up to us – a choice we make every time we open the app. We don’t need to pretend the chatbot has feelings. Maybe it’s enough to notice when impatience gets the better of us, and to pause before we let that become our normal way of speaking.
This brings me back to Matt. I still don’t know why I hung up while he was still talking. Irritation, probably. The rain didn’t help. I’m not sure it mattered to anyone but me. But that’s the point: it did matter to me, because of what I was practising, and what I might carry with me.