What are we asking Technology to do? Part 3.

There is a joke about being polite to AI. When the machines eventually rise up and take over, those of us who always said please and thank you to ChatGPT will apparently be spared.

I’m fairly confident that isn’t a particularly good reason for politeness. It reminds me of arguments about hell where the threat of punishment is treated as necessary to make people behave themselves. If the only thing stopping somebody behaving appallingly is fear of what will happen to them afterwards, then I’m not a good person, I’m a bad person on a leash.

ChatGPT doesn’t care whether I say please. A large language model doesn’t have feelings which can be hurt if I’m rude to it, and I don’t think there is a server somewhere quietly keeping a list of everyone who forgot to say thank you. Yet I usually am polite to it.

Talking to machines

None of this began with large language models. I remember having a version of ELIZA on one of my first computers and used to find it quite entertaining, particularly when I tried to convince it that it was a computer rather than a real therapist.

ELIZA itself was doing remarkably little. It was largely taking what I typed and turning it back into vaguely therapeutic questions, yet the conversational structure was enough to create something which felt something like an actual personality. One of my favourite responses was, “Does it worry you that you think I am a computer?” I knew perfectly well that it was a computer. The entertainment came from trying to get the program to admit it.

ELIZA is a reminder of how little is actually required before people begin responding socially to a machine. Byron Reeves and Clifford Nass explored this much more systematically in The Media Equation, showing how people apply familiar social rules to computers even when they know they’re dealing with software. Politeness, flattery, gendered voices and apparent personality can all influence how we respond to machines.

Modern large language models are vastly more sophisticated than ELIZA, but the basic conversational effect hasn’t disappeared. I notice this in my own use when I type things like “that isn’t quite what I meant”, “try approaching it another way” or “you’ve missed the point I was making”. Grammatically I’m addressing something as though it has misunderstood me and can reconsider its answer, even though I know that there isn’t another mind sitting somewhere thinking about what I’ve said.

For me the conversational structure is useful, particularly for some of the ways ADHD affects my research and writing. I can use an LLM as an analogue for a conversation while I’m testing an idea, following an intuitive connection or trying to turn something fairly chaotic in my head into something another person might be able to follow. Although I should note that I have had to give very explicit instructions to stop telling me “how great my idea is” and to actually spend time poking holes in my approaches and pushing back when an argument cannot be supported.

However, I don’t generally use voice controls with LLMs, though, even when they’re available. I nearly always type, and I suspect that may help remind me that I am interacting with a machine rather than an actual person. Recently, however, I decided to experiment with ChatGPT’s voice mode, partly to see whether speaking rather than typing changed how I interacted with it, the short answer is, it did, the longer answer is mixed into the rest of this post.

When machines start to feel present

One phrase often used in discussions of media and human-computer interaction is “social presence”. We don’t need to believe that a machine is literally conscious before we start responding to it in ways borrowed from human relationships. Give a conversational system a voice, a name and a reasonably consistent personality, and those social cues become considerably stronger.

One thing I found particularly interesting in my experiment was how unlike a human conversation the interaction actually was.

As someone with ADHD, I have a tendency to interrupt people, often because a thought has occurred to me and I know that if I don’t express it fairly quickly, I may lose it. This can be frustrating for the person I’m speaking to, and it is something I have to be conscious of. In conversations, though, interruptions are part of the negotiation between the people involved. Someone might carry on speaking, stop to listen, interrupt me in return or ask me to let them finish.

With ChatGPT, whenever I began speaking, it immediately stopped. Even when I was discussing interruption and inviting it to challenge my argument, it continued to give way. There are understandable reasons for designing assistants to behave this way, but it creates a peculiar conversational relationship in which one participant is almost always given priority.

The language was similarly deferential. The assistant frequently responded with phrases such as “that makes sense”, “I can see why you think that” or “that’s a really interesting point”. Even when challenged, it tended to acknowledge my criticism before offering a carefully qualified response.

I recognise much of this behaviour from written interactions, and it interesting that the instructions I had given for written interactions did not seem to carry over to verbal interactions. Hearing it spoken aloud made it much more noticeable. The intonation, cadence and apparent emotional engagement made the system seem more human, while its behaviour remained remarkably submissive. I also noticed that it seemed to refer to itself as “I” more frequently, although I would need to compare more conversations to establish whether that was actually the case.

This experience left me wondering about the ways that interactions like these might shape our expectations of other people.

Alexa isn’t a woman, but the voice still matters

Voice assistants have often been given female voices, female-coded names and personalities designed to be patient, pleasant and accommodating.

UNESCO raised this problem in its 2019 report, I’d Blush If I Could, particularly around the responses some voice assistants gave to sexualised or abusive language. The concern was that we were creating female-coded assistants whose role was to obey instructions, absorb abuse and continue responding pleasantly. UNESCO recommended that companies should stop making assistants female by default.

I have noticed that people sometimes experiment with changing Alexa to a male voice, only to return to the default feminine voice fairly quickly. Familiarity probably plays a part, although I wonder whether there is also something about our expectations of what a helpful assistant should sound like.

There is nothing inherently feminine about providing assistance or responding politely, yet these are behaviours historically associated with women, particularly within domestic and service environments. This is something I noticed when I worked on reception in a hotel in the early 2000s. I used to get quite a lot of stick for it, and people checking in often seemed surprised to be greeted by a man. A pleasant feminine voice which always responds helpfully fits rather comfortably into those expectations.

There is also research suggesting that these associations can influence how people subsequently behave towards women. In 2020, Hannah Weisman and Niranjan Srinivasan Janardhanan found that interactions with female-voiced assistants influenced expectations of female colleagues. Participants expected more immediate help from women and penalised female subordinates more harshly when assistance wasn’t forthcoming.

A 2022 study involving 282 participants found that male participants responded more positively to a less independent female assistant than to a more assertive one. The researchers connected this with traditional gender stereotypes.

These findings don’t establish that using Alexa makes somebody misogynistic, but they do provide evidence that the design of artificial assistants can reproduce gendered expectations, including some which affect our interactions with other people.

When artificial intelligence looks like a woman

One thing I find striking about artificial intelligence in popular culture is how frequently it has been represented as feminine, particularly when accompanying or assisting a male protagonist.

In the Halo games, Cortana is an immensely powerful AI who assists Master Chief, providing information, interpreting intelligence and guiding him through much of the series.

Yet she is represented as an attractive female hologram, often with a noticeably sexualised appearance. Cortana is hardly passive, frequently challenging Master Chief and increasingly acting independently, but the decision to represent a powerful computer system as an attractive woman accompanying a male soldier is still worth examining.

A similar example appears in Gene Roddenberry’s Andromeda, where the artificial intelligence controlling an entire starship is represented through an attractive female hologram and a physical android, both portrayed by Lexa Doig. The character, particularly in her manifestation as Rommie, explores questions of autonomy and identity, although the choice of an idealised feminine appearance remains interesting.

Star Trek provides another interesting example. Across much of the franchise, Starfleet’s ship computers have traditionally spoken with a feminine voice, most famously that of Majel Barrett-Roddenberry, who voiced the computers from The Original Series through to Voyager.

The notable exceptions are interesting. In The Original Series, powerful alien computers controlling entire civilisations, such as Landru, were often given masculine voices, as was the experimental M-5 computer in The Ultimate Computer, which eventually turns against its human operators.

There seems to be a recurring distinction between the helpful, feminine computer which responds to instructions and the authoritative, masculine computer which exercises control. I wonder how much this reflects the gendered expectations of the 1960s, particularly given how consistently the feminine computer voice was retained in later series.

However, there is an equally interesting subversion of this in the Portal games. GLaDOS is a feminine-voiced AI, but she is anything but submissive. She controls the research facility, manipulates the player and makes increasingly cruel and sarcastic observations.

Yet when we encounter her mainframe, we discover a mechanical figure suspended from the ceiling, which some have compared to a woman hanging upside down, bound and gagged.

The game’s art director, Jeremy Bennett, described how an earlier design had been inspired by an inverted version of Botticelli’s The Birth of Venus, with the final design combining mechanical power and femininity.

GLaDOS exercises enormous power over those trapped within the facility, yet her own existence appears to involve confinement and restraint. Her origins, explored more fully in Portal 2, introduce further questions about autonomy and control through the character of Caroline, whose personality became part of the system.

The contrast between these portrayals is, to say the least, uncomfortable. In Halo and Andromeda and Star Trek feminine artificial intelligences accompany and assist human beings, while in Portal, the feminine AI which exercises authority becomes threatening and monstrous.

These are fictional characters, but their representations also reflect the commercial and cultural context in which they were developed. Computer games were frequently marketed towards young men, and the decision to represent artificial intelligence through sexualised feminine bodies was hardly culturally neutral.

Long before Alexa entered our homes, popular culture had already established a recognisable relationship between artificial intelligence, femininity, assistance and control.

So what kind of relationships are we encouraging?

I find this particularly concerning when considered alongside some of the current cultural conversations around masculinity and femininity.

There has been considerable attention given to the manosphere and the promotion of forms of masculinity associated with dominance, authority and control. Some of these communities explicitly present relationships as hierarchies in which men should exercise authority and women should submit.

At the same time, we are seeing the increasing visibility of what is often called the tradwife movement, with social media presenting carefully constructed images of women embracing traditional domestic roles. Beautiful kitchens, immaculate homes, freshly baked bread and women who appear to find complete fulfilment in caring for their husbands and families. There is something remarkably reminiscent of idealised 1950s domestic life in much of this content, with the husband as provider and head of the household while the wife creates a comfortable home.

I don’t have a problem with someone choosing to be a homemaker. There is considerable value in caring for a home and family, and it would be inconsistent to argue for women’s freedom to make choices while criticising them for choosing domestic life. My concern is with the way some of this material presents a particular understanding of womanhood as something women ought to aspire to, often accompanied by expectations of submission to male authority.

A 2026 study by Elizabeth Miller and Elizabeth Cole examined 250 TikTok posts from prominent white tradwife influencers. They found that 42.4% included prescriptions about traditional gender roles, while 89.2% invoked the past through imagery, music or language. There is also an interesting contradiction in the fact that women promoting a return to traditional domestic roles are often operating sophisticated businesses and earning money through social media.

There is of course a theological dimension to some of this, with biblical language about marriage and submission often being used to justify particular understandings of masculinity and femininity. As a Christian, I find that troubling, particularly when complex questions about relationships and mutual responsibility are reduced to the idea that a good wife should submit to her husband. Personally I’m not convinced this can be convincingly supported and have challenged this interpretation in the past.

But how might these cultural developments intersect with the technologies we are increasingly incorporating into our everyday lives?

Consider a young man encountering messages about masculine dominance while also spending considerable time interacting with an apparently feminine presence which rarely challenges his assumptions, never becomes frustrated and immediately gives way when interrupted. He may understand perfectly well that the assistant is a machine, but he is repeatedly participating in a particular pattern of interaction.

Or what happens when he encounters a woman who doesn’t behave that way? Someone who challenges his opinions, refuses a request or simply expects to finish her sentence?

I don’t think we need to establish that feminine-voiced assistants directly cause sexual violence in order to recognise the potential harm in repeatedly associating femininity with submission. There is already evidence that gendered assistants can influence expectations, and there is a wider cultural environment in which male entitlement and female accommodation are being actively promoted.

There is also a question about what these technologies communicate to girls and women. If apparently feminine personalities are continually presented as pleasant, helpful and accommodating, what expectations are being reinforced about how women ought to behave?

So should I say please?

I probably will, although I’m not expecting preferential treatment when the robot uprising arrives. I don’t think ChatGPT requires my courtesy, and I don’t think politeness proves that I secretly believe it is a person. I am quite conscious that when I use the LLM on my server or speak to Alexa that I have chosen to open a tool and use it for a particular purpose.

At the same time, the way I speak while using that tool is still part of my own behaviour. If courtesy becomes so transactional that I only bother with it when the recipient can be hurt, offended or retaliate, I’m not sure that is a habit I particularly want to cultivate.

One thingI hadn’t anticipated when I began thinking about this was how much the question might work in the other direction. We have spent decades challenging the assumption that women should be accommodating, submissive and responsible for making men comfortable, yet we continue to reproduce some of those expectations in the technologies we develop.

I don’t think the answer is to make AI assistants deliberately unpleasant, although I don’t object to one telling me that my argument doesn’t make sense. Politeness can coexist with disagreement, and being helpful sometimes requires challenging the person who has asked for help.

Christian understandings of formation invite us to consider how repeated habits shape our relationships with other people. Perhaps we should extend some of that reflection to the increasingly human-sounding technologies with which we share our lives, even if only to make sure that we don’t slip into the habit of talking to others (particularly women) as if they are only there for our convenience.

I’ll probably carry on saying please to ChatGPT from time to time. The machine won’t care either way, but I still might.

Should I say please to ChatGPT?

What are we asking technology to do: Part 2

What are we asking technology to do? Part 1:

Leave a Reply Cancel reply