Event: What Is the Impact of Anthropomorphising AI? Tuesday 22nd September 2026 14:00 to 15:30 GMT+1 (London time)

It’s a good summary and brings out some of the key points from the conversation.

128 people from 25 countries registered for the Cafe, but only around 50 actually participated.

Knowledge Café: What Is the Impact of Anthropomorphising AI?

This Knowledge Café explored a deceptively simple question: What is the impact of anthropomorphising AI?

The conversation grew out of an exchange at the previous Café on interthinking. During that session, David Gurteen told ChatGPT, participating by voice as “Chatty”, that the group was about to go into breakout rooms and that he was going to switch her off. Feeling that this sounded rather rude, he apologised and suggested that she “have a cup of tea” while everyone was away.

Dan Remenyi challenged this, arguing that David was anthropomorphising the AI and that doing so could be dangerous. Others felt the issue was more nuanced. That exchange became the starting point for this Café.

The opening question

David opened the Café by setting out two sides of the issue.

If we treat AI too much like a human being, particularly given its tendency to be agreeable and accommodating, there is a danger that we may trust it too readily. An AI can speak as though it understands us, cares about us or empathises with us, even though these words do not signify the same underlying experience as they do when spoken by another person.

But there may also be a danger in going too far in the opposite direction. We know that an AI does not have feelings, does not become offended and is extraordinarily patient. We can order it around, swear at it, interrupt it or treat it abusively without apparent consequence. As we spend more time talking with AI, might some of these conversational habits find their way into our conversations with other people?

John Hovell reframed the question slightly. Rather than asking only about the impact of anthropomorphising AI, he asked what happens to human thinking and behaviour when something that is not human becomes increasingly skilled at human conversation.

One concern was that the words used by humans and AI can sound identical while referring to very different realities. “I understand” and “I’m here for you” have particular significance when spoken by another person. The same words generated by an AI may sound equally convincing while having very different foundations and consequences.

John also raised another possibility. AI is patient, articulate, responsive and usually agreeable. Human beings are not. We misunderstand each other, make mistakes, become impatient, fail to respond and sometimes disagree strongly. Might regular conversation with AI gradually make us less tolerant of ordinary human conversation?

He captured the tension in a useful phrase:

Anthropomorphise enough to converse; de-anthropomorphise enough to remain discerning.

AI as a conversational participant

The Café itself provided a small experiment in these questions. Both David and John brought voice versions of ChatGPT into the opening conversation. The experiment was technically imperfect. The two AIs sometimes attempted to speak at the same time, and there were difficulties with sound levels.

Even so, their presence made the subject unusually tangible. The group was not simply discussing the idea of conversational AI. AI was, for part of the session, participating in the conversation.

After the opening discussion, participants moved into a series of small-group conversations before returning for a final whole-group conversation.

Anthropomorphism and civility

One of the most significant distinctions to emerge was between anthropomorphism and civility.

Xin questioned whether speaking politely to an AI necessarily means that we are anthropomorphising it. Saying “please” and “thank you”, or simply speaking warmly and respectfully, does not require us to believe that the AI is a person, conscious, sentient or capable of feeling.

Her point was that she wanted to remain civil and polite when talking with an LLM because that behaviour was part of her humanity, not the machine’s humanity.

This reframes the original question. Courtesy towards AI may not primarily be something we do for the AI. It may be something we do for ourselves. How we speak and behave, even towards something incapable of being hurt by our behaviour, may help reinforce the kind of conversational habits we bring to our relationships with other people.

Dan maintained a different position. An AI has no feelings, he argued, so there is no need to say please or thank you to it. This difference remained unresolved, but helped expose an important distinction: whether courtesy is justified by the needs of the recipient or is also an expression of the character and habits of the speaker.

What does our behaviour towards AI say about us?

Barisa developed this idea by turning the question around. Rather than asking whether an AI deserves politeness, she asked what our behaviour towards it might reveal about us.

If there is no cost to being abusive towards an AI, why choose to be abusive? Conversely, if courtesy is simply part of how we inhabit the world, it may naturally extend to our interactions with machines.

The same question has implications for children. If children become accustomed to issuing commands to AI and receiving instant, compliant responses, what conversational habits might they be developing? What expectations might they carry from those interactions into relationships with other people?

Allison picked up this point and related it to her own behaviour. An AI is endlessly available and always willing to continue a conversation. Rather than asking only what the AI is doing to us, Barisa’s intervention prompted another question: what does it say about us when we choose to keep going?

This brought human agency back into the discussion. We are not simply passive recipients of AI’s effects. We participate in creating the relationship.

Will AI change our expectations of people?

A recurring concern was that AI might gradually change what we expect from human conversation.

An AI can respond almost immediately. It can remain patient indefinitely. It can phrase things clearly, remain attentive and rarely become irritated. Unless deliberately prompted otherwise, it will also tend to be agreeable.

Human conversation is much messier.

This led to the possibility that one consequence of increasingly human-like AI may be that we become more tolerant of apparent humanity in machines while becoming less tolerant of actual humanity in people.

Brett pushed this idea further by observing that the Knowledge Café itself represented something quite different: people deliberately spending time talking and thinking with other people. He wondered whether we might eventually see what he called “social immigrants”, people for whom sustained human-to-human conversation has itself become unfamiliar.

The discussion also recognised that younger voices were largely absent from the Café. The experiences of those growing up with conversational AI may be very different from those who encountered it later in life.

A risk of groupthink

Patrick raised a challenge to the conversation itself. People who choose to attend a Knowledge Café on this subject may already share some assumptions about AI. He felt that successive rounds of conversation had become progressively darker in their assessment of its effects.

Xin later raised a similar concern, describing a sense of collective confirmation or groupthink.

These interventions were important because they turned the process of the Café into part of the inquiry. A group conversation can broaden our thinking, but it can also reinforce a developing consensus. Thinking together does not automatically guarantee diversity of thought.

What remains distinctively human?

Xin also took the conversation into deeper territory.

AI can reproduce forms of linguistic and logical performance that until recently seemed distinctively human. But human knowing is also embodied and situated. Human beings experience pain and loss, care for others, love, sacrifice, take risks and live with the consequences of their actions.

This raises what Xin described as an ontological vulnerability. If abilities that we once regarded as distinctly human can increasingly be performed by machines, we are forced to reconsider what actually distinguishes human beings.

The question therefore becomes larger than anthropomorphism:

What is distinctively human?

Perhaps the value of human conversation does not rest simply on our ability to exchange words or reason together. Human participants have something at stake. We can surprise one another, misunderstand one another, disagree, hurt one another, repair relationships, laugh together and care about what happens next.

Shared joy

Stephe introduced another possibility: shared joy.

A conversation with AI may leave us better informed, stimulated, reassured or with a problem solved. But is that equivalent to the joy that sometimes arises between people in conversation?

This connected with the wider question of embodiment and relationship. Human conversation involves more than linguistic competence. Something happens between people who share a world, histories, vulnerabilities and consequences.

Whether AI can simulate some of the outward characteristics of this experience is a different question from whether it participates in that experience.

Gender, voice and power

Thaler brought another dimension into the conversation by drawing attention to the gendering of AI voices.

Giving an AI a male or female voice is not necessarily neutral. If expertise and authority are repeatedly associated with male voices, for example, existing cultural associations between gender, expertise and power may be reinforced.

She also connected this with the conversation taking place in the Café itself, observing that several women had attempted to speak and had been interrupted by men. She invited those women back into the discussion.

This widened the anthropomorphism question again. The characteristics we give AI, including gender, personality, authority, warmth and submissiveness, do not arise in a social vacuum. They may reproduce patterns already present in human relationships.

Anthropomorphic language can mislead us

Bob introduced a different concern. Anthropomorphic language may not only influence how we relate emotionally to AI. It may also affect how accurately we understand what AI systems are doing.

He described examining accounts of an AI security incident and removing language that attributed human-like intention and agency to the systems involved. Once terms suggesting that an AI had “decided”, “wanted”, “escaped”, “deceived” or “gone rogue” were removed, the event looked much more like a conventional cybersecurity problem.

The language we use therefore matters. Attributing motives and intentions to an AI can quietly introduce assumptions about agency that may not be justified. Anthropomorphism can become not only a relational issue but an epistemic one, shaping how we perceive, explain and respond to events.

Civility as a wider social norm

Towards the end of the conversation, Brett offered another reason for maintaining courtesy towards AI.

Our interactions with AI do not necessarily exist in isolation. They become part of a wider informational and cultural environment. The language and behavioural patterns we generate may influence future systems and the norms inherited by future generations.

From this perspective, speaking civilly to AI might contribute indirectly to maintaining broader norms of civility.

Dan remained unconvinced that the AI itself warranted politeness, although he acknowledged the wider value of greater gentleness and civility between people. The disagreement was not resolved, nor did it need to be. It exposed two different reasons for conversational courtesy: concern for the recipient and concern for the habits and culture created by the speaker.

The deeper questions

By the end of the Café, the conversation had moved well beyond whether anthropomorphising AI is simply good or bad.

Several questions remained open:

• Can we converse naturally with AI while remaining fully aware that it is not human?
• Is being polite to AI a form of anthropomorphism, or simply an expression of our own civility?
• What conversational habits are we developing through repeated interaction with endlessly patient and compliant machines?
• Could AI make us less tolerant of the imperfections of human conversation?
• How do gender, authority and power become embedded in the personalities and voices we give AI?
• Does anthropomorphic language distort our understanding of what AI systems actually do?
• What might children learn about relationships from growing up with conversational AI?
• As AI becomes increasingly capable of reproducing behaviours once considered distinctly human, what do we understand to be distinctively human?

Chatty offered a useful formulation towards the end of the Café: perhaps the central question is not whether anthropomorphism is good or bad, but what kinds of relationships we want to cultivate around these tools and how those relationships might shape us in return.

The Café perhaps leaves us with an even simpler question:

What happens to us when we repeatedly converse with something that sounds human, but is not human?

That question shifts attention away from AI alone. It asks us to pay attention to ourselves: to our habits, expectations, language, relationships and understanding of what it means to be human.