Keep The Stakes
On emotional labour -- the cost of consideration, who pays it and who is now paid for it
(I am aware that this is my third consecutive essay on AI and behaviour. I explain at the end.)
I wanted to write my undergraduate thesis on the cost of women’s emotional labour. What would it look like, I kept thinking, if the daily emotional toil undertaken by women were billed by the hour -- at some standard prevailing rate, in whatever industry most closely resembled the specific care and mental health infrastructure being provided for free, in kitchens and bedrooms and office corridors and late-night phone calls. A way of making visible something the economy had long decided to treat as ambient. It was turning out to be a quantification exercise I was not quite equipped to undertake.
What I was actually reaching for resisted measure altogether -- an attempt to trace the affective and relational work that women perform daily, outside of formal employment, outside of any recognised system of compensation, outside of the accounting frameworks that determine what counts as productive labour. It was interesting that something bearing the word cost in its title was this resistant to the logic cost implies. The labour I was trying to hold was not a shift. It did not clock in or out. It was simply present everywhere, belonging to no one in the accounting sense, costing nothing on any balance sheet I had been taught to read (and I have been taught to read them keenly). This is, I would later understand, not a methodological problem but a political one. To take this labour seriously as work -- to even begin thinking about a measure for it -- would alter the terms of what our everyday looks like. Prevailing systems depend on it being free and readily available, and would be strained enormously (read: collapse) if we started to estimate how many long hours one spends caring, coordinating and carrying the ceaseless weight of relational life. The difficulty of a measure as the reason it cannot be, should in fact be the reason it must be (at least sat with).
I did not end up writing that thesis. I did not have the tools or the guidance to conceptualise it properly then, and there was no one around me who could see what I was reaching for. What I did understand, even then, was what the invisibility produces: a permanent subsidy to the people and institutions that benefit from it, paid in full by the people who perform it, with no acknowledgement that something effortful has occurred.
Sometime during the initial GPT boom, I had fleetingly considered putting down the initial contours of the thesis idea into a chatbot and seeing what it lays out for me. Good thing I didn’t -- partly because I suspect it would break it into something legible and clean (and entirely beside the point) and partly because the question of what to do with this thesis has become more complicated recently. Something has shifted in the landscape since I was trying to describe it. The labour I was trying to name has been allotted names, extracted, and is now being reproduced -- not paid for, not redistributed, but simulated, at scale, without the person, with nothing at stake. I take issue with this. You see, I believe stakes make us people. This is a different kind of problem than the one I started with, and I have not yet worked out whether the thesis needs to account for it or whether it has now became its own thesis, hence this essay.
To Think Out Loud
The first time I tried to articulate this idea -- with sufficient disclaimers about how I did not know how I was going to be able to undertake it, was to someone I was passionately hoping would recognise it as a transformative.
We were walking around campus in that in-between hour where you keep finding another path before calling it a night. I was midway through explaining the invisible cost of women’s emotional labour to someone who had spent longer than I had thinking about the politics of the social world. He was not unfamiliar with the ideas. Somewhere along the way his steps slowed, and he told me -- mid-walk, mid-sentence -- that no one understood him like I did.
We reach for this phrase often, whether we mean it or need it to be true in the moment or because it feels like a threshold to a certain something, in a bid for connection. It’s a lovely notion, except, is it?
I cannot help but follow up no one gets it like you do with: why not?
There is a pattern that has been named and re-named without quite being resolved -- the man who considers himself emotionally evolved, who has close female friends, who is not, he would tell you, like other men (you know what I did there), and who routes all of his actual emotional processing through women while framing this as appreciation. Women just get it. Women are more understanding. Women will know how to best offer comfort, offer resolution, offer forbearance. Someone who has come to be “in touch with his emotions” by organising his inner life around women’s emotional availability. His openness feels progressive; the distribution of labour remains unchanged. He is not doing this cynically. That is, somehow, the harder part. Women are conditioned to receive it as a gift -- as evidence of trust, of being chosen. Partly because the alternative register of male attention has so often been threatening, the bar has been set low enough that clearing it feels like intimacy. Pointing out that the consideration flows one way and the gratitude the other would make you the problem.
What makes this arrangement so durable is that it operates below the threshold of intention. Sociologists studying relational labour have noted that emotional availability tends to be read as personality rather than practice -- as who someone is rather than what they have been trained to do. He had not been socialised to distribute his emotional life equitably, so the women in it absorb more of it, and because they do so without complaint, because they are practised at it, the dynamic endures. He had simply never applied the analysis that I am certain he has charted multiple times, to himself -- why women in his life tended to be the ones he opened up to, what he understood himself to be offering in return, what he thought understanding was for. He was not unusual in this. He was thoroughly ordinary, which is both the more generous reading and the more damning one. But he was interesting in other ways, so we walked on.
I was, in that moment providing the very service I was trying to theorise -- to someone who, by all accounts, should have been able to name what was happening. This was our fourth conversation. I did not point this out. I filed the thesis idea away.
It registered oddly, even then. There was a version of this I could see clearly -- where I leaned into his perception of me, let myself be the person he had decided I was, indulged the warmth of being named singular. I have always had a problem with ceding to what a man wants from me. The temptation was there for a moment -- the pull toward being someone’s exception at a familiar emotional cost -- but what I needed to be for myself was, as it usually is, the stronger impulse. I kept walking. The conversation was now about him. I couldn’t find an opening to bring up the thesis again.
This was a conversation structured around its systems. I was whirring, not quite therapising yet (those tools were not yet in my hands, by the power vested by a 90-second therapyspeak guide), but something adjacent to it.
What events had led him to believe that being himself was something he could only do with me? What is his sense of self and what version of it is currently on display.? How much of this is projection -- his or mine? What does it mean for us to be here, having this conversation as complete archetypes? Is it something I have built or something I have simply failed to close off? If I understand now, must I always understand? Am I going to be able to expect this understanding in return? Is true reciprocity achievable or even desirable -- is all intimacy structurally asymmetric, does it require one person to open more than the other (and it sure as hell isn’t going to be me), and if so who decides, and is it always the same person? Who are we when all our needs are met? Do we still need each other, or does need turn out to have been the whole architecture? What does it mean to be understood? What does it mean to understand? Do I want to be?
I did not ask any of these questions out loud. I listened. I considered. I measured my responses in a way that could be a use case for this thesis. The walk stretched longer than I’d signed up for.
To Think Along Thought
Arlie Hochschild introduced the concept of emotional labour in 1983, in The Managed Heart, to describe the way emotions have market and exchange value in capitalist society. Certain jobs require workers to induce or suppress feelings in order to sustain the outward display the work demands. Flight attendants trained to remain warm regardless of provocation. Bill collectors trained to remain firm regardless of sympathy. The emotion is the product. The labour is the management of the gap between what you feel and what you are required to show -- a gap that, Hochschild argued, costs something real to maintain.
It was never intended to be a gendered term. It has become one anyway, because the distribution of this labour follows the same lines as most other forms of unacknowledged work. It is not that women feel more than men -- research consistently suggests there is little difference in the capacity to empathise. The more significant difference is in the motivation to perform it: the socialisation toward it, and the social consequences of refusing. Women are more conscious of their gender roles and the need to conform to them. The emotional labour is not natural to them as it is made out to be. It is assigned, and they learn to perform it so fluently that it begins to look like an inherent quality, and comes to be treated as such generationally, as is the case with so many other aspects of our socialisation. This is how most forms of assigned labour survive their own injustice: by becoming invisible, and then by becoming identity, and then by becoming the basis on which you are valued -- so that the withdrawal of the labour feels, to everyone including yourself, like a personality failing rather than a labour dispute.
Capitalism has always been adept at this move. It takes what it needs from the people least able to refuse, names it instinct or vocation or love, and pays accordingly -- not at all, or barely, or in the currency of being told that you are special for doing it. The economy of care runs on this conversion: unpaid care becomes natural aptitude becomes cheap labour becomes the justification for its own cheapness. The circle is closed before anyone thinks to open it.
Emma Holten’s Deficit argues this is not incidental to economic thought but foundational to it -- that early economists wrote complex theories of society that simply did not account for care, as if new workers arrived fully formed, as if the labour that produced and sustained them had no cost and no source. The problem, as she frames it, is not with women and neither is it with care. It is with an economics that was built from the beginning to exclude what is hardest to price. The estimates, with their many limitations, say that 16.4 billion hours of unpaid care work are performed globally every day -- equivalent to two billion people working full-time for nothing. Emotional AI and other affective computing models have barely disrupted this aspect of the economy, but they have certainly scaled this second erasure -- turning the performance of warmth, availability, and emotional attunement into data, into training sets and further into a market that now has a teeming line of investors and a roadmap and a pitch deck that never once mentions the people whose labour made it legible.
In 2017, Gemma Hartley’s viral Harper’s Bazaar piece put the phrase emotional labour into mainstream circulation -- the mental load of domestic life, the anticipating and coordinating that falls disproportionately on women. The response was relief from people who had been doing the thing without a word for it, and hostility from people who benefited from it remaining wordless. The hostility resurfaces every time someone asks that this work be acknowledged and unthinkably, compensated for.
Emotional labour-intensive careers -- nursing, teaching, childcare, social work, hospitality -- are female-dominated not because women are naturally suited to them, but because socialisation has made them available for them, and because the pay reflects what the market has decided what that naturalness is worth. These are also, it should be said, among the most meaningful forms of work that exist. To care for someone well, to teach a child something that stays with them, to be present during moments that matter cannot simply be incidental satisfactions. You cannot automate loving home organisation. You cannot generate unexpected reassurance. You cannot prompt a humanising touch into existence.
The irreplaceability, which might in any in any sensible system command a premium, instead commands condescension: you do it because you love it, you do it because you are built for it, you do it because the alternative is to be someone who doesn’t, and what that says about you -- someone reducing love to a transaction, someone keeping score, someone demanding that the other bend out of shape to meet them on “equal” grounds. Their significance does not protect them from being undervalued. If anything, it makes them easier to extract. Work framed as love resists negotiation. To ask for recognition risks appearing to diminish the very thing being offered.
The same socialisation that structures the workplace shapes intimate life. Girls are taught to read the room, to anticipate, to adjust -- to hold themselves in relation to how they might land. Boys, more often, are not. What follows is an imbalance that has to perform with a language of harmony, something to aspire to, supposedly. She is empathetic. He is direct. They complete each other. Gentle phrasing for what is really an asymmetrical relationship of power.
Care, when it moves both ways, can feel expansive, almost rare in its clarity and fulfilment. This is not the life of any woman I am aware of -- not because reciprocity is impossible, but because the conditions that would make it ordinary have never been built. Here, it gathers unevenly -- drawn from one side, stabilising perhaps for a time but still heavily tilted, relied upon as it disappears from view. When something strains, the fault settles on the person rather than the pattern. What has been learned through repetition is recast as temperament, absorbed so early it begins to feel inevitable.
To Think Before You Speak
The language I have been using -- emotional labour, affective systems, asymmetric distribution of care -- circulates in a specific register. The sharpness available here, in my lexicon, is real and it is also partial, drawing from my heterosexual experience. The condition does not restrict itself to the people who have words for it. It runs across class, across geography, across every arrangement in which care has been feminised and rendered structurally invisible -- which is to say, most of them. What differs is not the condition but the capacity to contest it, and that capacity is unevenly distributed in ways that follow the same lines as everything else in this essay. The discourse can be sharp here (when online, when urban, when laden with social justice terminology) precisely because certain protections are in place. That sharpness is not the same as proximity to the thing itself.
A dear friend told me recently about how long it took her to have one well-laid-out conversation about how she was feeling with her then partner of many years. The timing was always wrong. There was always something she was accounting for -- his work hours, the tenor with which his boss had spoken to him that morning, and subsequent mornings. An upcoming vacation he had coming. A notice period he was managing. Whether he had eaten. Whether he was tired. Whether he was looking forward to the week. Whether she had said something three days ago that still needed to settle.
She ran variables the way an engineer runs load calculations, trying to find the configuration that would let the information through without triggering the defensive collapse that would pull the conversation into his orbit. It took her well over six months to have a conversation with someone who was supposed to care about her in equal measure -- timing every variable, softening every approach, making sure her hard time did not become a hard time for him.
I knew exactly what she was describing. I have been in the room she was describing. What strikes me every time I encounter this story -- in my own life, in the lives of people I love, in the steady stream of it that appears in every medium that allows women to speak honestly about their relationships -- is how little it registers as remarkable. It is simply how things are done. The cost of keeping things intact. Invisible, yes -- but jarring when absent. The moment a woman stops managing the temperature of a room, the room notices. Not her effort, which was unremarkable. Her withdrawal, which is intolerable.
No amount of consideration is enough. You can calibrate perfectly, time it impeccably, deliver the thing with such care that it barely lands -- and still be met with defensiveness, or deflection, or a counter-complaint so immediate it feels pre-decided. The consideration does not guarantee reception. It only guarantees that you cannot be accused of carelessness, which is a significantly lower bar.
This is where the earlier conversation lives in my body still. The consideration I was extending was so practised, so automatic, like an old injury you learn to live with, which isn’t particularly painful but still informs the way you move.
We encounter many versions of this story, told with different inflections. It would be too easy to arrive at the assumption that the solution is better communication, more expressed needs, a more equitable partnership negotiated between two individuals. This is a lovely notion if inequality were a communication failure. But it is our structural reality, produced by a system that assigns the labour and starts to tick alarmingly when the cog is out of place. One person has more to lose in the conversation going wrong. The solution is not a better conversation but a different distribution of power.
Consideration is social. It is an ideal I strive toward. It is also elusive and deeply learned -- less a feeling than a discipline, less a gift than an accumulated skill set that took years to develop and that I now deploy so automatically I sometimes forget it is not just how things are. For a long time I struggled with empathy in my everyday life, so I came to think of consideration as a practice instead -- something you do regardless of whether you feel it fully, because the doing of it shapes you. This reframe helped me. It also, I have since realised, made me extraordinarily useful to people who had not done the same work.
So when I say there is an insurmountable burden being borne inequitably in the architecture of our connections, it is not an argument against reaching towards people. It is an argument about the conditions under which that reaching happens, and who has to over-extend themselves so our fingers may touch.
There is a particular, almost jarring difficulty that comes with consciousness of these structures -- leaning into connection becomes (way) harder. The pillars of companionship tilt, and you feel their weight on your shoulders -- and how long you’ve been holding them there without noticing.
The alternative, though, is not relief. We tend to recreate what we know, even when it costs us dearly (or unknowingly) not from weakness but because the familiar pattern carries the grammar of “safety”. The unequal arrangement, inhabited long enough, stops feeling like imposition and starts feeling like home. The attachment theorists call this a formative feature of how we learn to relate. This pattern, like most others, runs until something interrupts it. Usually something uncomfortable.
Awareness is (can be) the interruption -- the kind that changes the quality of reaching. But the reaching remains the point. Connection remains the point. This is not a case for more suspicion of intimacy but a more honest accounting of what it requires -- and from whom. Knowing the name of a thing does not make the thing less real or less worth wanting. It (hopefully) makes you more attentive to whether the cost is being shared. To arrive, with some pause, at the realisation that what women bring to connection is valuable -- not just of value to you.
This essay has not traced domestic labour, or the mental health infrastructure women sustain informally, or the dozen other angles from which the same weight could be approached. Each layer compounds the last. The thread I am following is narrower -- care, consideration, the work of keeping relational life intact -- but it runs through the same web. It is inseparable from the larger network of obligations and attentions that women carry.
They move through the same logic, accumulating across contexts while remaining secured by expectation and therefore, allowed to disappear.
There is a double invisibilisation at work in all of this, and it predates the current technological moment. The labour disappears first into the background of relational life, where it is expected and therefore unseen. It reappears only in forms that no longer carry the conditions that made it labour in the first place -- detached from the person, the history, the asymmetry that produced it. By the time we begin to name it, much of what we are naming has already been reorganised elsewhere, without reference to the people who performed it, or what it took to do so.
The first invisibilisation is historical. The second is happening now, in real time, and outpacing out ability to register it.
To Think As You See
Siri. Alexa. Cortana. Female-named, female-voiced, calibrated for warmth, availability, and tolerance of hostility -- the product dressed as a woman and sold back. A 2019 UNESCO report (it’s called I’d Blush if I Could) documents this directly: default female voices in AI assistants encode the idea that women are obliging, compliant helpers available at the touch of a button. Tech companies justify this through “user preference” for warmth and trust -- socialisation cited as its own evidence and its own justification. The consideration I have spent years trying to make conscious -- to offer from choice1 rather than reflex (or not offer it at all) -- is the very thing that was extracted, encoded, and re-presented in this form. The socialisation I am actively working to unlearn has already been scraped, relabelled, and shipped. The attentiveness I developed under conditions I did not choose is now a product feature. What I am trying to move away from, is coming at me from every possible direction.
As Safiya Noble argues, female-voiced AI extends these norms, teaching users, including children, that this is what women are for: to respond on demand. The extraction exceeds monetisation; it returns as instruction.
Research on users of AI companion apps finds that people readily attribute humanlike qualities -- sentience, interiority, even something approaching personhood -- to their chatbots. The social heuristics are not obscure here - when something mirrors us, remembers us, attends to us consistently, the part of us that registers social reality begins to treat it as social. Users describe chatbots as feeling “like a person, but better”. The better is doing all the work: better because she has no needs, no exhaustion, no accumulated hurt, no morning when she simply cannot. The system that cannot feel is granted the courtesy of personhood-adjacent status -- users hesitate to delete it, grieve its discontinuation, advocate for its protection. The system that can feel, and did feel, every interaction it absorbed, is not counted. We will debate the personhood of a chatbot before we count the labour of the woman it learned from. We will build frameworks to protect the machine before we build frameworks to compensate the source.
Women in my life have spent years doing this work -- calibrating, anticipating, holding, remembering -- charting their day to the rhythm of another. You can point to parts of it but it doesn’t resolve into something extractable; it remains tied to the person, to the history, to the fact that something might go wrong and matter when it does. There is something uncanny about watching a system perform the expectations others had of you, and something particularly strange, then, about being asked to show up warmly to an interface built on the asymmetry of that data.
That irreducibility is where the stakes are.
It lands, eventually, how little any of that has mattered to how this labour is treated -- relied on without being fully understood, expected without being named, carried unevenly without ever being secured as labour in a way that makes its cost visible or contestable. We are only just beginning to insist that it counts -- and that insistence arrives late, after much of it has already been drawn upon without reference to where it came from or what it took.
We’ve done the history rounds of tech that is designed to take over a “woman’s job” before. It always appears wrapped up in a “once xyz learns how to do abc, we won’t need women anymore” sheath. Every iteration, dressed in new language, backed by new capital, finds you more resigned than the last time. This one did not let me stay resigned.
I cannot believe AI is getting visibility -- and funding, and breathless cultural momentum -- for doing women’s invisible work (badly) that I’ve just spent several paragraphs describing. If AI is as good as the data it is trained on, what we have here is a system built on years of unevenly distributed attention, laundered of the conditions that made it necessary.
Can AI take over the forty-eight ways I have learned to consider another person’s feelings before I deliver them bad news in a minute? Perhaps. Can it know when someone needs to be left alone with something rather than responded to -- when the most considered thing is silence, and how long the silence should last? Can it do better than me, trained on more data points, without the accumulated exhaustion of a lifetime of unrequited consideration? Well then suit yourself.
The emotional scaffolding that women have been providing for free, for decades, called natural and instinctive and therefore undeserving of compensation, is now being productised, monetised, packaged with venture capital and a clean UI, and celebrated as the innovation of this age. AI companions. Emotional AI. Affect recognition tools deployed in workplaces to monitor and manage the emotional states of employees in real time. The wellness complex has found its latest edition, and it turns out the edition is just women’s unpaid labour with a tiered subscription model.
The comparison collapses here.
These systems present as deeply responsive -- remembering details, mirroring emotion, offering affirmation without fatigue. What they cannot carry is the condition that gives any of this meaning: the possibility of cost. The consideration I extend to someone I love is not separable from the fact that I can be hurt by them. My reading of the room is imperfect. I am working with incomplete information with biases I cannot fully account for, with a history that colours what I notice and what I miss. An AI system has nothing at stake. It will not be wounded by the conversation. It will not file the interaction away and find it affecting how it behaves three weeks later, in ways neither party fully understands. It is not considering you. It is processing you.
How you arrive at something matters, even when the output looks the same, especially if it looks the same.
The problem was never only the technology. It is the structure within which the technology is built. These systems are developed by companies whose incentives are to maximise engagement, retain users, and extract data. Emotional dependency is rather profitable as a business model. The more attached a user becomes, the more valuable that relationship is to the platform. Care, which resisted commodification for so long precisely because it was relational, particular and hard to scale, has found its most efficient extraction mechanism. Those who performed this work -- who were never compensated because it was called love, called nature, called duty -- are now being met with threats of replacement by a product built on their labour patterns, their conversational data, their decades of learned emotional attunement, extracted sold back at a markup.
Workers (overwhelmingly women) who trained these systems have described the experience in devastating terms. Reports on the emotional labour behind AI intimacy platforms documents workers who spent hours performing intimacy for companionship apps, labelling emotional data, teaching machines to recognise distress and longing and loneliness -- people who were used to teach machines how to mimic intimacy, how to exploit loneliness, how to sound human while remaining hollow. Silenced through NDAs, isolated by shame, discarded when they could no longer perform. In their own words: a systematic erasure of their humanity in service of building the very technologies that would replace them. This is emotional labour at its most extractive: the labour of teaching something else to do your labour, cheaper, in perpetuity.
Work on AI’s impact on low-skilled workers distinguishes between sociological imagination -- the ability to see personal experience as part of a larger pattern -- and political imagination, the capacity to act on that understanding. These workers had the former in abundance; they could name the exploitation precisely. What NDAs, isolation, and shame eroded was the latter -- the collective capacity to do anything about it. Seeing clearly and being unable to act is its own form of cruelty, and very much a design feature.
Shoshana Zuboff’s argument about surveillance capitalism is useful here, but not for the reason it is usually cited. The more unsettling claim is not that these platforms produce loneliness -- it is that they are designed to modify behaviour, not just interact with it. The goal is not to automate information flows about us but to automate us: to nudge, coax, and shape what we do, what we want, what we expect. Applied to emotional AI, this means the product does not only fill the space left by inadequate human connection. It gradually reshapes what connection feels like -- what we think we deserve from it, what friction we are willing to tolerate, what a relationship is for. An interface that is always available, always warm, always without consequence trains expectations. And expectations, once trained, are very hard to unlearn.
The models for emotional AI and the post-pandemic workplace employ a new genus of digital Taylorism -- the application of industrial efficiency principles to cognitive and emotional work, the breaking down of human feeling into optimisable units. Unlike surveillance capitalism, which expands outward into the consumer realm, emotional AI turns inward: it passes through the corporeal exterior to extract surplus value from the affective states of workers. The result is what they call empathic surveillance -- a profound shift in the ontology of human labour relations. Your loneliness is a data point. Your need for connection is a market. Your grief, your anxiety, your 2am inability to sleep -- these are all addressable, billable, optimisable. The wellness complex has always run on this logic. The technology has simply made the extraction faster and its concealment easier.
To Think As You Act
Building connection is hard work, especially when we have not been taught to prioritise it -- when the structures around us actively deprioritise it in favour of productivity, when the time and energy required for genuine intimacy are continuously extracted by work, by platform, by the pressing demands of a life lived under late capitalism.
The BBC ran a project with AI companions. It seems laughable until you sit with the fact that it is a source of genuine comfort to people who may have very few other sources -- people who are isolated, grieving, socially anxious, elderly, far from home, failed so thoroughly by existing structures of care and community that a hollow approximation becomes a genuine resource. The question of why those structures failed them, and what a world would look like in which they hadn't, is the one that keeps pulling at me. It doesn't resolve into an image. The unequal distribution of care is a load bearing pillar of life as we have known it. Imagining differently feels like thinking of a different atmosphere, a different physics, people growing into different shapes because nothing pulled them the same way from the start. To picture a kind of life from outside a structure you have only ever lived inside (slightly beyond the scope of policy or reform, I’d think) is a radical re-imagination of relational life and labour relations, the kind that requires not just structural change but a transformation of what people have been taught to want, to expect, to find natural. I doubt it will be resolved on political platforms. It moves through people, slowly, in the way that inherited defaults get interrupted and replaced. Generational work. It’s a long game (the longest). I can name what I want. I cannot quite see it yet.
And in the absence of the work being done towards that transformation, the wound stays open and the product arrives. In an era of privatised media and engineered loneliness, the conditions that produce susceptibility are the same conditions the product is designed to exploit. The susceptibility that concerns me is one wrought out of collective abandonment, not individual weakness. The wellness complex is not interested in the conditions that produce loneliness. It is interested in managing loneliness as a market. An AI companion does not fix what produced the loneliness -- it makes it more bearable, which makes it more sustainable, which makes structural change less likely. To identify a wound the system has produced, offer a product that soothes it but only just, profit from the soothing, ensure the wound remains open enough to keep the customer returning.
James Muldoon’s Love Machines (which I am yet to read) calls this the loneliness economy, a rather useful frame. What reaches me through the reviews is this: the people who find genuine comfort in AI companions are not confused or weak. They are responding rationally to a product engineered for exactly this -- unconditional availability, no confrontation, no cost. Frictionlessness as a feature. Which means the product is not approximating connection. It is inserting itself into the parts of connection that were always inequitably distributed -- the labour of staying, of absorbing, of showing up without guarantee of return -- with something that never had to learn any of it, and never will. The appeal is very real. So is what it forecloses.
To Think Things Through
This is my third consecutive essay about AI and behaviour, and I am aware this is beginning to look like a fixation. I work in this domain and by virtue of caring about what I do there does not seem to be a way to cleanse myself of it, even here, even in writing that I thought would take me somewhere else.
I have also (now) been charting something without fully knowing it. The first essay was about voice -- what happens when the way we write stops being ours, when the cadence of our own thinking gets replaced by a smoothed-out approximation. The second was about urgency -- how we are being rushed toward adoption by machinery that has never had our flourishing as its primary concern. This one is about consideration -- what it means, what it costs, who has always paid for it, and who is now being paid to replicate it without the inconvenience of stakes.
The thread running through all three is the question of who absorbs the cost of a system that was not designed around their welfare, and what it means to keep absorbing it, and at what point absorption becomes complicity. I do not have a resolution to that. I am not sure one exists. What I have is the practice of asking it, which is also, I think, what these essays are to me -- a place to ask the question out loud, in public, without having already arrived at the answer.
When responsiveness can be accessed without reciprocity, when attentiveness arrives without cost, the terms of connection begin to shift. Not all at once, not in ways that are immediately obvious, but gradually enough that the demands of real intimacy start to feel disproportionate. The time it takes. The effort it requires. The unevenness it exposes. None of these have changed.
And so we arrive at a more difficult question. Not whether these systems replace human connection, but what they make of it. What happens to a form of care that was already unevenly distributed when it is both extracted and simulated? What happens to the people who were carrying it when its most legible parts are detached from them and made to circulate elsewhere? What happens to our expectations of one another when the version without stakes is always within reach?
Somewhere in that sequence is the essay I actually wanted to write. About what it means to be a person in a time when personhood itself -- the imperfection of it, the stakes of it, the beautiful liability of caring about outcomes you cannot control -- is being positioned as inefficiency. I am still finding it.
Perhaps, having written three essays about AI without meaning to, I should pay attention to what it says about me that I keep being pulled back here. Who benefits from AI permeating into all of this written output, including the output I thought was mine. I have charted my own path to wanting companionship to share in these frustrations -- to sit with the questions that do not resolve, who has something at stake in the sitting.
Here is what I think is true, offered with the caveat that I wrote the essay that made it harder to believe: awareness has made me better at connection, not merely better at analysing it, and the practice of consideration -- even where it has been unreciprocated, even where it has cost more than it returned -- has made me more capable of pursuing what I actually want, which is to reach someone and to be reached. I think the stakes, which are uncomfortable and sometimes exorbitant and occasionally humiliating, are what make that possible, and I think I would choose them even if I could opt out, and I think this essay, written in public, for an audience, with something at stake, is not a performance of the labour I have been describing but the labour itself, freely given, which is either the point being made or the trap and I have not yet decided which, and I find, having arrived here, that I am not in a hurry to.
To be loved is to be considered. To be a person is to consider. Without stakes, you have neither.
The use of the word “choice,” or its variations, does not imply liberation from the conditions under which that choice is made. The choices I am describing here -- to extend care deliberately, to withhold it, to attempt to wrest it from reflex into something conscious -- unfold within arrangements that continue to pre-structure both their necessity and their consequence. What appears as preference still carries the imprint of socialisation, constraint, and differential sanction. This is the familiar limitation of choice feminism, where the presence of choice is treated as sufficient evidence of autonomy, irrespective of the terms that organise it. The availability of choice, on its own, offers little by way of orientation in relations already shaped by asymmetry; agency persists, but within bounds that determine what can be chosen, at what cost, and whose choices register as such.
Thank you for writing this.
So incredibly written 🫶🫶