one failure mode keeps recurring across these notes: delegate a capacity to a tool, and the capacity goes slack. comprehension debt is that mechanism applied to a codebase. an engineer who lets an agent author and merge what no human still reads has not saved the work. the work moved into the apparatus, and the understanding left with it.

the same mechanism has a sharper test case, and it is not about code. it is about the act of figuring out what one feels and putting it into words. that faculty is now delegable to a language model, in real time, in fluent prose, which makes it a clean test of the de-skilling pattern, since the skill being handed off is definitionally interior. nobody else was ever doing it.

call the result techno-alexithymia: the technologically mediated erosion of the capacity to identify, describe, and feel one's own emotions. not a disorder caught from a device. a skill let go slack, because a machine holds it on request.

one disclosure up front, because the rest depends on it. techno-alexithymia is not an established construct. no scale, no validation study, no clinical entry, no source in the literature that uses the term. what the literature holds is three separate fields, each gripping one corner of the idea, none reading the others. that gap is the reason for the note. the phenomenon is arriving faster than the words for it.

the act

the starting point is a familiar one: a vague sense that something is wrong, before it resolves into any specific thought. the older way of handling that was to sit with the feeling until it produced a word for itself: grief. boredom. the specific loneliness of a sunday. naming the feeling was itself part of the work, and doing that work was, in part, how a person learned to feel it clearly.

the alternative is to open a chat window and type why do i feel like this. the model answers in seconds, patiently, and offers a few possibilities plus a follow-up question. usually one of them lands, and the feeling resolves into language the person did not generate themselves. the relief is real. it is also a transaction: naming an interior state, a task that used to require doing it yourself, traded for something cleaner and faster. the trade tends to repeat, because every incentive points that way.

delegation de-skills

bernard stiegler spent a career on a single claim: human capacities live outside the body, in tools, as much as inside it, and anything placed in a tool can be drawn back out of the person. he took proletarianization from marx. the factory absorbed the artisan's know-how into the machine, and the worker shrank to the size of what the machine now held.

stiegler's move was to say the absorption does not stop at the hands. in a 2017 essay he argued that the culture industries proletarianized the capacity to feel: "a loss of knowledge." if feeling is a knowledge, a practiced and embodied skill, then like any skill it can be moved into an apparatus and left unused. he died in 2020, before conversational models, so he did not live to see the technology that fits his argument most precisely: an apparatus that speaks in the first person and appears to understand.

this is the same pattern described elsewhere in these notes under another name. a capacity that gets exercised stays strong. a capacity that gets delegated goes soft. the variable that matters is whether the tool leaves the operator more capable or less.

the correlation is solid. the arrow is not.

people who struggle to name their feelings use their phones more compulsively. the link replicates. three independent meta-analyses land in the same neighborhood, a moderate-to-large positive correlation of r ≈ 0.40 to 0.42 between alexithymia, or its cousin emotion dysregulation, and problematic phone use. one pooled 26 studies and 23,387 participants at r = 0.41. a second put emotion dysregulation and problematic use at r ≈ 0.416 in the relevant dysregulation subdomains, not across the whole reviewed literature. a third found smartphone addiction and alexithymia at r = 0.40. the phones and the muteness travel together. that part is not in dispute.

the arrow is. every one of those analyses is cross-sectional, a snapshot, and every one says so in its own limitations section. the data cannot tell whether the muteness drives the phone use, the phone use erodes the capacity, or a third thing drives both. the standard caution about correlation applies in full here.

a methodology note, since it's a good small example of the larger problem. an earlier version of this piece cited a meta-analysis of "110 studies, 96,680 participants" as the evidence for the alexithymia link. that total is real, but it describes the whole meta-analysis, not the alexithymia-only relationship, which rests on the smaller figures above. the number had been carried over and re-scoped without checking. the error is left visible here rather than quietly fixed, because the underlying story is compelling enough that the pull toward bigger numbers than the evidence supports is real, and staying accurate mattered more than staying polished.

the only longitudinal evidence comes from a single 2025 paper, and it splits. in a two-wave study following 901 young adults over three months, distress tolerance and problematic phone use each predicted the other, both directions at once. in a 14-day daily diary of 77 people, the arrow pointed one way: phone use on a given day predicted diminished emotional capacity the next, and not the reverse. read narrowly, that one diary suggests the device takes something day to day, while the three-month study shows the two trading influence. one small study is not proof, and the daily effect could still run through sleep, mood, or a dozen unmeasured paths. but the shape, if it holds, is the shape of a skill atrophying through disuse: small daily concessions hardening into mutual dependency.

one construct caveat that the rest of the field tends to skip. alexithymia (cannot identify or describe feelings), emotion dysregulation (cannot manage them), and distress tolerance (cannot sit with discomfort) are three instruments measuring three overlapping but distinct things. most of the strongest data measures dysregulation and distress tolerance, not alexithymia proper. the slippage is not a footnote. it is the first thing a real theory of techno-alexithymia would have to pin down.

the outsourcing already happened

greg wadley, wally smith, peter koval, and james gross coined digital emotion regulation in 2020: the strategic use of digital technology to influence one's own affective states. gross is the psychologist whose process model defines the field, which makes the framing an established one rather than an alarmist one. the observation is almost banal once seen. the phone consolidated every mood tool a person owns, music and messaging and games and the voices of distant friends, into one portable platform available at essentially all times.

everyday use maps cleanly onto gross's strategies. a comfort show instead of the taxes is situation selection. scrolling to numb a spike of anxiety is attentional deployment. texting a friend to turn a humiliation into a funny story is cognitive change. for a great many people the phone has become the primary channel through which feeling gets managed at all.

this is measured, not asserted. in a one-week diary study, 21 of 23 office workers reported deliberately using ordinary technology to manage their own or others' emotions. twenty-one of twenty-three. the same researchers named the worry in their own measured language: psychologists have flagged "over-dependence on technology," some participants raised "over-reliance," and "the boundaries between productive and counterproductive uses of technology for personal emotional well-being are presently not well understood." that reads as a field that has located a boundary it cannot yet see past.

the menu flattens resolution

a quieter version of the problem lives in the interface. affective computing relies on categorical models: a fixed, small menu of discrete emotions. happy, sad, angry, surprised, afraid, disgusted. the machine has to sort the user into a bin, because bins are what it can compute, and the bins are how it hands the feeling back.

real people hit the wall fast. in one reviewed system, users complained that the sixteen emotions on offer "were not sufficient to express their emotional experience." sixteen, already far past the six-emoji reaction bar under a message, and still a cage. the texture of an actual interior state, with its blends and contradictions, gets quantized down to the nearest available label.

picking a feeling off a menu is recognition. finding the word unaided is generation. recognition is easier, which explains its appeal, and also why relying on it exclusively is worth being cautious about. the capacity that atrophies under recognition-only practice is the capacity to generate. emotional granularity, the ability to draw fine distinctions between feelings, is one of the better predictors of psychological health in the affect-science literature. the menu is convenient, and it may also be shaping users toward feeling in lower resolution, though that specific link, from reaction menus to a measurable loss of granularity, has not been demonstrated. it remains a testable hypothesis nobody has tested, itself part of the gap.

built to repair the same faculty

the same technology is also being built to repair the faculty it may erode: apps that train emotion-recognition skills and, in small pilots, appear to reduce alexithymia; just-in-time adaptive interventions, like a wristband that reads skin conductance, notices arousal climbing past a personal threshold, and prompts a coping strategy in the moment. in that form the technology works as a prosthesis, restoring a faculty rather than absorbing it.

the evidence is thin and young. a scoping review of 65 hci emotion-regulation papers found only four randomized controlled trials in the entire set; most findings the reviewers called "speculative," resting on indirect self-report rather than any direct measure of improved skill. many prototypes, little proof.

a tool that trains a skill and a tool that replaces it look identical from outside: both have the user handing feelings to a device. the difference is whether the user walks away more capable or less, which is exactly the boundary the researchers say they cannot locate yet. it's the same boundary that separates an agent that teaches an engineer a codebase from one that quietly becomes the only thing that understands it.

the unmeasured frontier

every mechanism above predates large language models. outsourcing emotional management to a playlist leaves the feeling intact and only changes its weather. the reaction emoji flattens expression to other people. the conversational model does something different in kind. it performs introspection itself, the inward act of identifying and naming a feeling, on the user's behalf, in language, in real time, and often better than the user can.

that is the precise faculty alexithymia is defined as the absence of. not managing the feeling. not expressing it. identifying and describing it to oneself. accepting a chatbot's fluent, plausible, well-organized answer to "why do i feel this way" outsources the single operation the entire construct is built around.

a deliberate search for any study of what llm-mediated introspection does to a person's own capacity to identify emotions returns essentially nothing. no verified study addresses it directly. the experiment is effectively running on the whole user base already, with no instrument pointed at the outcome. absence of measurement is not reassurance, it just means the question has not been asked yet.

the structure matches every other de-skilling pattern in these notes. the chatbot is more available than any friend, more articulate than a half-formed inner sense, never tired, never judging. delegation is frictionless and the relief is genuine every time, which means the conditions for atrophy aren't just present, they're reinforced by how convenient the tool is. if feeling is a knowledge that can be proletarianized in stiegler's sense, a model that performs introspection looks like an unusually efficient vector for it, since it absorbs the highest and most inward layer of the skill, the layer that used to be definitionally the person's own.

what would make it real

the current shape of the field, summarized: a strong but directionless correlation between emotional muteness and compulsive device use; a theory of digital emotion regulation that describes the outsourcing precisely and cannot find the line between healthy and harmful; a philosopher who predicted, before the fact, that any faculty handed to an apparatus can be hollowed out; interfaces that train recognition over generation; an under-proven set of tools that might repair what the rest of the system causes; and, brand new and entirely unstudied, a machine that does the introspecting.

each piece lives in a different building. the psychometricians do not read the continental philosophers. the hci researchers do not talk to the affect scientists who study granularity. nobody has put the pieces on one table and asked whether they describe a single thing.

for techno-alexithymia to move from coinage to construct, the unglamorous work remains undone: defining it as distinct from ordinary alexithymia and ordinary dysregulation, building a measure for it, and running the longitudinal studies that could settle the direction of the arrow, namely whether sustained delegation of emotional naming actually de-skills the namer, and on what timescale. until that work exists, this stays a hypothesis wearing the clothes of a diagnosis.

what this suggests for practice

the same discipline that applies to agent work applies here: the model gets used after an attempt, not instead of one. reaching for the chat window before trying to find the word oneself, even briefly, skips past the part that keeps the faculty in shape. not because the model's answer is wrong, but because the finding is the faculty, and a faculty stays sharp only through use. an engineer who never reads what an agent merged and a person who never names what they feel are running the same pattern, and accumulating the same kind of debt.

related notes

• context is a budget: delegation has a cost, and the scarce capacity is best spent where it changes the next decision
• reward hacking in coding agents: a fluent, plausible answer is not a correct one; relief is not a measure of fit
• agents.md effectiveness: comprehension debt as the cost of offloading understanding to an apparatus

sources

1. meta-analysis of 26 studies (n = 23,387), alexithymia-mobile-phone-addiction r = 0.41, 95% ci [0.37, 0.45]. pmc8866180
2. shahidin et al. (2022), emotion dysregulation-problematic smartphone use, pooled r ≈ 0.416. clarity measured via the ders; the bridge to alexithymia's "difficulty identifying feelings" facet is this note's, not the paper's. pmc9740505
3. smartphone-addiction-alexithymia meta-analysis, r = 0.40, 95% ci [0.36, 0.43]. the inflated "110 studies / n = 96,680" version of this claim did not survive verification; these are the verified figures. pmc9481561
4. wang et al. (2025), addictive behaviors. study 1: two-wave longitudinal, n = 901, bidirectional prediction between distress tolerance and problematic use. study 2: 14-day diary, n = 77, use predicted next-day distress tolerance but not the reverse. distress tolerance is adjacent to, not identical with, alexithymia. sciencedirect
5. wadley, smith, koval & gross (2020), "digital emotion regulation," current directions in psychological science. coins the term; maps device use onto gross's process model. sage
6. smith, wadley, webber, tag, kostakos, koval & gross (chi 2022). seven-day diary, 23 office workers; 21 reported deliberate use of technology to manage emotions. source of the over-dependence, over-reliance, and "boundaries not well understood" quotations. acm
7. bernard stiegler, "the proletarianization of sensibility," boundary 2 44:1 (2017). "i use the term proletarianization to refer to a loss of knowledge (savoir)." duke university press
8. antle et al., scoping review of 65 hci emotion-regulation papers: only 4 rcts; most findings non-causal and self-report-based; users found 16 discrete emotion categories "not sufficient to express their emotional experience." sciencedirect
9. conceptual review of digital emotion-regulation interventions, including an eda-sensorband just-in-time adaptive intervention found feasible and acceptable over three months (feasibility, not efficacy). pmc8846444

note: "techno-alexithymia" is a coinage, not an established scientific construct. no source in the literature uses it. this note assembles adjacent, validated findings into a hypothesis and marks where the evidence ends and the speculation begins.