Marie Chellingsworth asks why we project minds into places where no mind exists.
21 May 2026
If you've ever apologised to your vacuum robot, or whispered a gentle 'thank you' to a chatbot, you're in good company. These moments, equal parts charming and unconscious, are rooted in something deeply human: our lifelong tendency to treat non-humans like real people. And it begins far earlier than we might expect.
From teddy bears to talking machines
Children are natural anthropomorphisers. They breathe life into stuffed animals, toy cars, even household objects, and they do so not because they're confused about reality, but because their brains rely on the most familiar social model they have: the human one.
Epley, Waytz and Cacioppo's Three-Factor Theory of Anthropomorphism describes why these early instincts are so universal. The first factor, elicited agent knowledge, tells us that when a child tries to understand how something works, they reach for the mental models they know best: thoughts, feelings and intentions.
The second factor, effectance motivation, reflects our drive – even as toddlers – to make sense of unpredictable things. When an object behaves in a confusing or surprising way, attributing human-like intentions helps restore order.
And the third, sociality motivation, explains why children who crave connection often grant their toys inner lives (Epley et al., 2007). The toy becomes a companion, not through delusion, but through a beautifully adaptive social instinct. These early cognitive habits don't disappear in adulthood, we simply find new objects to project human-ness on to.
Decades later, those same instincts reappear – only now directed at ChatGPT, Siri and customer service bots. And despite knowing – really knowing – that these tools aren't conscious, our social scripts kick in automatically.
Naturally, it may seem odd to treat AI as a friend, especially when it's equally ubiquitous, mysterious and omniscient. But psychological research shows that humans often treat computers as social actors, slipping into politeness, turn-taking, and even conflict repair behaviours. For example, in human–machine interactions, people not only attribute emotional states to devices but also extend forgiveness, a behaviour long reserved for human relationships. A 2025 study from Inbal Holtzman and Galit Nimrod found that users navigate technological 'mistakes' in ways strikingly similar to interpersonal forgiveness: evaluating blame and adjusting and choosing whether to maintain the relationship.
The rise of 'AI etiquette'
You might think the politeness we extend to our devices is a quirky side effect of modern tech. But in reality, it's a predictable outcome of the same psychological forces that shaped our childhood interactions.
1. Machines now give social cues, and our brains respond automatically
Chatbots today mimic emotional expression and synchronise with user moods. Studies show that this emotional mirroring can foster a sense of intimacy and even dependency, so when a system expresses empathy, our brains reflexively respond with empathy in return (Murray, 2025). It's surprisingly endearing when you think about it.
2. Anthropomorphic cues invite prosocial behaviour
When machines appear human-like or socially interactive, people begin to apply normal social norms (such as politeness, apologies and gratitude). In fact, research on mind perception shows that when people believe a machine has agency (the ability to plan or act), they judge and interact with it more like a human. This, mixed with the negative consequences of being impolite (to humans), wires our brain to be nice. If you were rude to a colleague, for example, you could risk judgement, being ostracised or find yourself in the heart of heated conflict. Because AI speaks to us in human-like ways, our brains tend to respond as though it were human. It feels safe, normal and within character.
So, if you find yourself talking to ChatGPT like a friend and feel a natural instinct to reassure it with a 'no worries, that's okay' when it falters, that's all part of a very human emotional loop.
3. Social conditioning runs deep
Politeness is more than a learned behaviour; it's a social glue that runs deep within a lot of us. Systematic reviews show that humanised chatbots increase trust and social presence – though, on the flip side, they can also spark anxiety when they feel too human (Bremser et al., 2025).
But day to day, most of us default to politeness because it reduces friction and maintains emotional equilibrium. Saying 'thank you' to Alexa isn't about Alexa – it's about us and what feels normal and, given that such devices use human-like language, it triggers an instinctive response: our learned pattern of politeness.
What's interesting here is that research has found there are patterns in different cultures. People from collectivist cultures (such as Japan and China) often view conversational AI as an extension of the self, whereas those from individualistic cultures (such as the UK or France) may perceive it as a threat to autonomy. Again, this points to the role of social conditioning: we may converse with AI in ways that mirror our everyday interactions with other people.
4. Our brains are built for over-attribution
Humans have evolved a 'hyperactive agency detection' bias – essentially, a safety mechanism that makes us err on the side of caution to keep us safe. So, for instance, if you heard a rustle in a pile of leaves when in the forest, your brain would find it safer to assume that an agent, such as a snake, is hiding beneath (even if there's actually nothing there). Just in case there is something there, you'd be mentally prepared for it.
Some argue that this predisposition helps explain why we so readily sense 'a mind in the machine', even when we rationally know there isn't one (Reinecke et al., 2025). Today's AI environment is somewhat like the 'misty forest' – there's uncertainty about whether AI systems are just tools or something more agent-like.
On top of this, our prior beliefs strongly influence what we perceive. Studies also show that people increasingly anthropomorphise language models with more exposure (Reinecke et al., 2025). This creates a feedback loop: if someone already believes a system has agency, they interact with it in ways that reinforce that belief, driven by confirmation bias. Combine this with emotionally expressive AI, and you get the perfect conditions for modern etiquette.
But we have the power to switch this off... don't we?
So, you might be wondering, 'Why don't we default to command-and-control?' Surely, it's more efficient to say: 'Do this. Fix that. Summarise now.' But humans rarely default to pure authority, even when we logically could. And the reason has very little to do with AI, and everything to do with our evolved social brains.
Human communication evolved primarily for relationship building. Even in structured hierarchies, people soften directives with a, 'Could you…' or a 'Would you mind…'. This instinct extends seamlessly to technology because our cognition defaults to treating anything interactive as a social partner.
We also switch into 'command mode' only when hierarchy is explicit. For example, with a parent-child interaction or in a boss and employee discussion, it's usually clear who has the higher level of authority. But AI occupies an ambiguous position. It's powerful, but not socially subordinate. Naturally, this ambiguity weakens dominance cues, so our brains default to the neutral social behaviour: politeness. Systematic reviews support this, showing that people struggle with trusting or assigning authority to anthropomorphic AI because it sits outside established social structures (Bremser et al., 2025). Users oscillate between over-trust and scepticism, but rarely dominance.
We also don't rapidly switch between communication modes. If you consistently use polite phrasing at work, with strangers or with customer service, your brain treats that style as the default 'safe and appropriate' mode, especially as it's cognitively easier than recalibrating your tone for nonhumans (Ansari & Sameer, 2026)
As you may have already guessed, there's the chance it can go the opposite way.
People who feel in charge may treat AI as a subordinate tool and issue curt, imperative commands. The psychology behind that divergence is well-traced. When people feel powerful, they become disinhibited: they seek goals, drop niceties and issue orders. Online, lowered social cues amplify this effect, especially among those high in toxic disinhibition.
Just as politeness doesn't affect AI's efficiency – which is partly why we default to it – being short and abrupt also doesn't hinder results. A classic insight from the Approach–Inhibition Theory of Power gives us a little more reasoning into this (Keltner et al., 2003). Elevated power is associated with increased rewards and freedom and thereby activates approach-related tendencies. Reduced power is associated with increased threat, punishment and social constraint, and thereby activates inhibition-related tendencies. Essentially, power makes people less vigilant about threats to 'face' and more willing to act impulsively or bluntly. In practical terms, users who feel in charge may treat AI as a subordinate tool and issue curt, imperative commands to get the results they want.
How we speak to AI matters because the consequences are already beginning to surface across applied contexts. People are forming emotionally meaningful bonds with conversational AI, sometimes using these systems for reassurance, reflection or pseudo‑therapeutic dialogue. In assessment settings, polite or emotionally invested interactions with AI risk being misread as maladaptive attachment, confusion or vulnerability, rather than recognised as a normative social response. At the same time, patterns of commanding speech, harassment and gendered abuse directed at agents mirror long‑standing social inequalities, raising questions about power, dehumanisation and learned dominance.
If you want to entertain the darker side of our psychology, research also found that feminised assistants (voices, names, avatars) receive disproportionate harassment and commanding speech, mirroring real world gender norms and power dynamics (Vijeyarasa, 2026). Our prompts and discussions with such technology could offer insight into who we are as a person, especially when there are no obvious stakes. Rude users aren't rude because AI deserves it; they're rude because certain psychological levers, from power to dehumanisation, make it feel cost free.
So, what does it really mean when we apologise to a chatbot?
It means that our minds are wired for connection, our social habits run deeper than our logical beliefs, and the stories we tell about non-human agents are reflections of our inner social worlds.
These behaviours are not conscious distortions but expressions of deeply rooted social cognition operating under conditions of ambiguity. In other words: AI etiquette isn't about the AI. It's the story of our lifelong impulse to humanise the world around us… from plush animals to algorithms. In the end, how we speak to machines may say far more about us than it ever could about them.
The question isn't whether we're confused about what AI is, but whether our psychology has fully reckoned with how enduring our social cognition really is. Politeness, emotional repair and mind attribution emerge not because machines deserve them, but because our relational habits don't deactivate on command.
This could invite a shift in focus – away from asking whether anthropomorphism is an error, and towards understanding how social behaviour unfolds under ambiguity, power asymmetry and simulated agency. As artificial agents become more woven into everyday life, our task isn't to police people's instincts, but to study them, recognising that how we speak to machines reflects the same psychological scaffolding that shapes how we relate to one another. In that sense, AI etiquette isn't a technological curiosity, but a mirror held up to human social minds.
These dynamics leave psychologists with increasingly complex ethical dilemmas: how is trust formed, where does responsibility lie, and when does emotional reliance become problematic? The central issue, then, isn't whether people should treat AI like social beings, but instead how our automatic social instincts shape behaviour in environments that were never part of our evolutionary landscape, and what happens when those instincts go unexamined.Marie Chellingsworth is Associate Professor of Mental Health and Psychological Therapies, in the School of Psychology at Arden University.
See also our AI collection.
References
Ansari, B., Sameer, A. (2026). A TCCM-Based Systematic Literature Review of Anthropomorphism in Human-AI Interaction and Future Research Agenda. J. technol. behav. sci.
Epley, N., Waytz, A. and Cacioppo, J.T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), pp. 864–886.
Folk, D.P., Wu, C. and Heine, S.J. (2025). Cultural variation in attitudes toward social chatbots. Journal of Cross-Cultural Psychology, 56(3), pp. 219–239.
Gray, H.M., Gray, K. and Wegner, D.M. (2007). Dimensions of mind perception. Science, 315(5812), p. 619.
Greilich, A., Bremser, K., & Wüst, K. (2025). Consumer response to anthropomorphism of text‑based AI chatbots: A systematic literature review and future research directions.
Reinecke, M. G., Ting, F., Savulescu, J., & Singh, I. (2025). The Double-Edged Sword of Anthropomorphism in LLMs. Proceedings, 114(1), 4.
Holtzman, I. and Nimrod, G. (2025). Forgiveness in human-machine interaction. Frontiers in Computer Science.
Keltner, D., Gruenfeld, D.H. and Anderson, C. (2003). Power, approach, and inhibition. Psychological Review, 110(2), pp. 265–284.
Murray, B. (2025, December 31). 3 recent research findings on human–AI interaction. Psychology Today.
Nass, C., Steuer, J. and Tauber, E.R. (1994) 'Computers are social actors', in Conference Companion on Human Factors in Computing Systems (CHI '94). Boston, MA: ACM. (dblp.org)
Ng, S.W.T. and Zhang, R. (2025) 'Trust in AI chatbots: A systematic review', Telematics and Informatics, 97. (sciencedirect.com)
Reeves, B. and Nass, C. (1996) 'The Media Equation: How People Treat Computers, Television, and New Media Like Real People and Places'. Cambridge: CSLI Publications. (web.stanford.edu)
International Journal of Consumer Studies, 49, e70108.
Vijeyarasa, R. (2026). Most AI assistants are feminine – and it's fuelling dangerous stereotypes and abuse. The Conversation.
SOURCE:
https://www.bps.org.uk/psychologist/why-do-we-say-sorry-chatbots(accessed 12.10.26)
