I began with a small and frustrating experience. When my phone was beside me, I repeatedly wanted to check Instagram, Zhihu, or WeChat. I could know perfectly well that I had more important work to do and that much of what I would encounter would contribute little to it. That knowledge did not reliably prevent the next check. Sometimes a brief inspection became a much longer session, and I emerged tired, dissatisfied, and aware that the time I had intended to spend thinking had disappeared.
There was an additional contradiction. I reasoned that something genuinely urgent would probably announce itself through a notification; I did not need to keep opening a feed to discover whether it existed. This was my working assumption about how important information would reach me, rather than a guarantee that notifications captured everything. Nevertheless, it made the repeated checking harder to explain. If I was not waiting for a particular message, and did not expect most of the material to help me, what was I repeatedly going back to obtain?
The question initially seemed straightforward: why did I keep doing something I did not, on reflection, particularly want to do? I wondered whether this was addiction, a failure of self-control, or the predictable effect of applications designed to make continued use unusually easy. I had already tried changing my phone’s display to black and white, which helped. Yet I wanted something beyond a practical adjustment. I wanted to understand whether I could keep the applications available and still decide, actively and reliably, when to use them.
I explored these questions in a conversation with an AI assistant. The discussion developed through repeated qualifications and objections. I would recognise something in an answer and then explain why it still did not reach the difficulty. An apparently practical solution could raise a question about agency; an account of agency could return me to the design of the application; a discussion of design could make me ask again what I could do myself. Eventually, I compared social media with the very activity through which I was examining it: sustained dialogue with AI. That comparison emerged from the earlier questions rather than replacing them.
Knowing better did not stop the next check
The first useful formulation was a “checking habit.” Research by Oulasvirta and colleagues described brief, repeated inspections of changing smartphone content, reinforced by the occasional discovery of something rewarding. Such checks could also lead into longer periods of use (Oulasvirta et al., 2012). I recognised that pattern immediately. I was not always opening an application to answer a question or accomplish a particular task. Often, I was checking whether something had appeared. My response was essentially: yes, that is exactly the pattern—but I already know it is interfering with my work. In frustration, I described almost everything I encountered as useless to what I needed to do, while allowing that Zhihu could sometimes be more useful. That was a judgment about relevance to my purposes, not a measured percentage of worthless content. The puzzle was that even this judgment did not stop the action. Recognising the habit explained something about its form; it did not explain why knowing better was insufficient, or show me how to change it.
The assistant also suggested measuring what mattered: unplanned checks, uninterrupted reading, and how I felt after using the applications. These measures addressed important outcomes, but my response returned to the same difficulty. Rationally, I already knew what mattered. I wanted to do my research, and I knew that it needed concentration. This was why I began asking whether the problem was psychological, whether addiction was the right word, and whether my own will was somehow failing. Those were questions prompted by a conflict between intention and action. The conversation did not establish a clinical diagnosis, and I did not need such a diagnosis before taking that conflict seriously.
A disagreement over the phone’s physical presence made the problem more concrete. The assistant cited a meta-analysis of 33 studies that found no significant overall effect of a smartphone’s mere presence on cognitive performance (Hartanto et al., 2024). I objected immediately that, in my case, having the phone nearby plainly mattered. I gave the example of waiting for an AI reply: when there was an interval of several minutes, the phone offered an obvious activity, and I could slip into checking it. At that stage, I wondered whether switching to another task would help; I had not yet settled on the response I would propose later. Looking back, the pooled research finding and my objection concerned different things. A silent phone’s average effect in a cognitive task is not identical to the sequence of seeing it, opening an application, and abandoning the task. My experience did not refute the meta-analysis. It showed why that result, by itself, did not answer the question I was asking.
The assistant’s suggestion to move the phone out of sight and reach for a manageable period generated a different objection. I wanted to know whether I could keep it nearby and still choose to continue working. Uninstalling the applications felt too passive to me: the action would become impossible, but would I have become more capable of choosing? My experience with a black-and-white display complicated my own position, because changing the environment had helped. I was therefore holding together two things that did not fit neatly: evidence that an environmental adjustment worked, and a desire for a capacity I could exercise while the temptation remained. This tension returned later. It would misrepresent the discussion to resolve it here simply because an argument for environmental support is available.
What fragmentation was taking from me
Before I had a satisfactory account of the behaviour, my frustration widened into sadness about the technology itself. The phone was supposed to make life better. Why did so much of my use seem to lead toward entertainment, repeated stimulation, and time I later regretted? I invoked Amusing Ourselves to Death and Brave New World because they expressed the disturbing possibility that attraction could help produce submission. I also worried about the values and relationships encouraged by these environments. These comparisons expressed a fear I wanted to examine; they were not evidence that a particular dystopian future had already been established.
I then asked whether fragmentation was one of the primary culprits, whether the issue was attention itself, or whether something deeper lay behind both. I also corrected the emphasis on “breaking news.” Much of what pulled me into Instagram was entertainment, something amusing, or an invitation to change the topic. It did not need to be urgent or politically consequential. This correction mattered: an explanation focused on wanting to remain informed would miss the attraction of the next interesting item, even when I had no informational need for it.
X became my clearest example of the pattern. I described a torrent of information that made me eager to find something new while leaving me with fragments and a sense of exhaustion. Sometimes two hours could pass in this way. Afterwards, I could feel that I needed to rest or nap, which meant that the cost extended into the time when I had hoped to resume work. I was describing the experience of those sessions, rather than measuring the platform’s effects on every user. But the experience sharpened my question: why could an activity feel continuously attractive while I was doing it and so unsatisfactory when I considered what it had given me?
I put my diagnosis strongly: focus is vital to meaningful work, and fragmentation seemed to be directly eroding it. Deep thinking, as I experience it in research, requires holding a problem long enough to develop it: remembering what an objection challenged, seeing how a new detail changes the whole, and allowing an initially unclear idea to become more precise. A stream can contain individually worthwhile items while repeatedly interrupting that process. The distinction worth making is between a short piece of information and a fragmented activity. A brief observation can contribute to a sustained inquiry; an endless succession of unrelated observations can prevent one from developing. This gives more precision to my original complaint without withdrawing its central point. I wanted the continuity through which information could become understanding.
Recognising my experience in Williams’s questions
When the assistant mentioned James Williams’s Stand Out of Our Light, I recognised the book. I had already been thinking about the kinds of concerns it raised, and I said so. I then brought Williams’s publication list and substantial texts into the conversation: the Banksy essay, the Nautilus interview, the critique of Facebook’s advertising model, and the Guardian essay (Williams, 2018c). My reaction was a sense of recognition: were these authors describing precisely what I was experiencing? The texts became material with which to examine the experience, rather than names added afterwards to give it authority.
In “Orwell, Huxley, Banksy,” Williams places two images beside one another: surveillance around a telephone booth, and an embracing couple looking at phones behind each other’s backs. His interpretation distinguishes control over information from the direction of attention, and contrasts our impulses with our intentions (Williams, 2014). This helped articulate something in my initial puzzle. Information could be available to me while my capacity to attend to what mattered was repeatedly diverted. The question therefore could not be settled by asking whether I had access to useful content. I also needed to ask what drew me toward some content, away from other work, and back again the next time.
The book’s distinction among immediate attention, longer-term purposes, and the reflective capacities through which we examine those purposes gave the concern a broader scale (Williams, 2018a). The navigation-device analogy in the interview offered a concrete version: a tool can be easy to use while repeatedly carrying someone away from their intended destination (Gallagher, 2017). I could now connect the immediate irritation of a check with the accumulated effect of many such interruptions. The question was becoming whether the tools through which I encountered information supported the intellectual life I wanted to sustain.
The denial-of-service metaphor in the same interview prompted a further question of my own. To understand and address the problem, I needed to read these arguments carefully. Yet the habits being analysed could make that reading difficult. I described this as a vicious circle: the attention required to investigate distraction was itself being interrupted. That was more than agreement with Williams; it was a consequence I drew from placing the text beside my situation. It would go beyond the evidence to infer a permanent loss of the ability to read. The practical circularity was enough to matter. Better explanations would remain incomplete if I could not sustain the activity through which I examined and used them.
The problem became a question about design
My frustration made me ask whether these applications were designed with bad intentions. I had already singled out infinite scrolling as a feature that seemed fitted to continued use. Later, I asked more explicitly whether companies employed psychological expertise to study how to capture attention and produce addiction. I was pressing against the idea that the entire difficulty belonged to my character. Yet the question about expertise did not establish the motives of particular employees, and the word “evil” could not substitute for an account of what the system rewarded. The relevant distinction was between investigating a design’s effects and assuming a private intention behind every one of them.
Williams’s critique of Facebook helped separate these issues. He argues that advertising can create organisational priorities that conflict with users’ interests even when individual employees care about those users (Williams, 2018b). This preserved the force of the design criticism without requiring a story in which every designer wanted to harm people. It also returned me to a concrete question: what behaviour counts as success for the organisation, and how closely does that correspond to what I wanted from the application?
I then pushed the question of responsibility into an awkwardly quantitative form: could the fault be divided into percentages between the company and the user? We had no basis for calculating such a division, but the question should not disappear simply because its numerical form was inadequate. I wanted to know how responsibility could be understood when a person acted within conditions deliberately arranged by someone else. Users make decisions; organisations determine many of the options, cues, rewards, and stopping opportunities surrounding those decisions. Causal contribution, moral responsibility, and the practical power to change a situation are different considerations. My wish to become more capable of choosing did not relieve designers of responsibility for the conditions of choice.
The assistant introduced an ethical qualification that I accepted: entertainment is a legitimate purpose, and a designer should not assume that scholarship is valuable while games or comedy are inferior. But my agreement was immediately followed by another objection. Surely people should retain the freedom to choose; could the legitimacy of entertainment justify drawing them into it without an effective limit? I was questioning the inference from enjoying an activity to endorsing its continuation. If I entered for a brief diversion and repeatedly emerged much later, exhausted and dissatisfied, the fact that individual items had entertained me did not settle whether the whole episode expressed what I wanted. The possibility of reconsidering and stopping belonged inside the account of agency.
I connected this directly to human–computer interaction and ethical design. When Value Sensitive Design entered the discussion, I said that I had studied it before and knew why it mattered. My objection was that it could remain optional: knowing how to consider values does not ensure that an organisation gives those values practical authority. I was not offering a survey of how many companies followed VSD. I was asking why a framework I recognised as important seemed insufficient to prevent the experience we had been analysing. That shifted the question from whether ethical design methods existed to how their requirements could shape products, incentives, and accountability.
The assistant later offered a concrete example of regulatory attention. On 6 February 2026, the European Commission announced preliminary findings concerning TikTok’s addictive design. It questioned the effectiveness of easily dismissed time-management controls and said changes to the service’s basic design were needed. The announcement explicitly left the investigation’s final outcome open (European Commission, 2026). This was relevant evidence against the impression that design escaped every form of scrutiny. It was not evidence that the problem had been solved, or that a preliminary finding had already produced the changes users needed.
My response was that this still seemed weak, although it was a step forward. That reaction mattered because I had also asked whether there was a possible solution or whether the trajectory would simply continue toward something resembling Brave New World. A regulatory example answered part of the question about what was being attempted; it did not satisfy the question about whether those attempts matched the scale of the problem. I remained dissatisfied with both easy reassurance and inevitability. The issue was how meaningful control could become a requirement of design rather than a possibility left for the user to discover.
Why I returned to active choice
After discussing structural responsibility, I returned to the personal question. The assistant suggested keeping the phone nearby, acknowledging the urge with a phrase such as “I want novelty,” and returning to a chosen action, initially for five minutes. This addressed my stated preference more closely than simply moving the device away. Nevertheless, I again insisted that I wanted an active choice: something I could positively do, with the applications still installed, rather than a solution defined only by preventing access. This was a recurrence of the earlier objection, now made after considering design, ethics, and regulation. Explaining the system had not removed my wish to act differently within it.
I even suggested that leaving the applications available and choosing not to use them would represent more self-control than uninstalling them, because the latter could leave me controlled by the environment. There is a counterargument: choosing an environment can itself be an exercise of agency. My experience with grayscale supported the practical value of such a choice. But it would be too neat to present that counterargument as something that settled the discussion. Two aims remained distinguishable: reducing unwanted behaviour under present conditions, and developing a capacity that would persist when the opportunity to engage returned. I wanted help with the second aim as well as the first.
This is why the word “active” kept returning. I wanted to direct myself toward the work I valued. A useful response would need to connect the moment of temptation with an action that continued my inquiry, so that I had something to do with my attention beyond monitoring whether I was resisting the phone. Later, when I proposed writing down my own thoughts while waiting for an AI reply, I supplied one example of such an action. At this point, however, the more general problem remained open: how could deliberate engagement become more reliable than the checking routine?
I consequently asked again about mechanism. Why was fragmented content so effective at attracting attention? Why did the attraction persist when I knew it was interfering with something more meaningful? Was this dopamine, a feature of cognition, or something like hacking a vulnerability? I also wondered whether the brain was poorly suited to this volume and kind of information. These were attempts to locate an explanation, not neuroscientific findings I had established. Naming a neurotransmitter would not by itself tell me why a particular cue prompted a particular action, or what could make a different action easier to sustain.
Research on reward learning helps delimit that question. Lindström and colleagues found that social rewards helped explain patterns of posting, and experimentally changing rewards influenced posting behaviour (Lindström et al., 2021). This supports a behavioural mechanism in a defined setting; it does not establish that all scrolling is addiction or identify the cause of each of my checks. The distinction I needed was between what readily attracts behaviour and what I endorse as worth pursuing. My question about why the brain seemed to “like” social media more than difficult thinking brought that distinction into view. Immediate attraction and reflective importance were plainly not functioning as interchangeable measures in my experience.
The phrase “exploiting a vulnerability” led me to ask about values as well as attention. Could the same environment that rewarded repeated checking also help explain the prominence of extreme ideas? This opened several possible questions: whether certain expressions received greater visibility, whether people adapted what they posted to obtain rewards, and whether continued exposure changed what seemed normal or important. Those possibilities should not be collapsed into a single established cause. VSD scholarship already recognises that values and their interpretations can change through technological development and use (van de Poel, 2021). My question was moving toward how environments might influence what people pursued, while the original concern about sustaining my own research remained in place.
Why AI entered the comparison
Only after developing these concerns did I draw a stronger comparison with AI. I argued that social media seemed to generate little or no social productivity or intellectual development, whereas productive AI use appeared to require a continuing contribution from the person using it. I listed what I meant: reflection, counterarguments, new parameters, critical thought, and further qualifications. On social media, a photograph of a meal, drinks with friends, a gathering, or a holiday could receive attention more readily than specialised or professional content. In a serious AI exchange, I thought, precisely the unusual details that narrowed a public audience could help produce a more useful answer. I followed this strong contrast with questions: was it actually the case, were other factors involved, and how did it look from an HCI–VSD perspective? It would be misleading either to repeat the generalisation as an established finding or to erase it in favour of qualifications I supposedly held from the beginning. I wanted to investigate whether different environments encouraged fundamentally different uses of attention and intellectual effort.
The broadest version of the claim needs qualification. A smaller audience for specialised work may reflect the background knowledge it requires. Ordinary photographs can sustain relationships, and social platforms can host demanding exchanges. Popularity does not measure intellectual value, but neither does everyday sociability become worthless because it is not scholarship. These considerations limit what can be inferred from engagement figures without dissolving the observation that prompted the comparison. The more precise question concerns which contributions a particular environment rewards, how it responds to complexity, and whether it helps people develop an inquiry beyond the first encounter.
What interested me about thinking with AI was the possibility of remaining with a question while changing the way I understood it. This conversation supplied a concrete example. I recognised the description of a checking habit, then insisted that recognition did not solve the difficulty. I challenged the relevance of the claim about a phone’s mere presence. I connected the personal problem to Williams’s writing and then to HCI and VSD. Later, I questioned whether the comparison with AI exposed a different relationship between attention and intellectual effort. The assistant could propose a formulation, but my objections affected what the next formulation needed to address. Continuity allowed the previous exchange to become material for further thought.
This does not establish that AI inherently demands deep thinking. It can supply a fluent answer that a person accepts without examination, and an elaborate exchange can still contain errors. The stronger formulation concerns a possibility: a person can use AI so that sustained questioning contributes to the development of an answer, while continuing to examine its evidence and assumptions. An impressive output, an enjoyable interaction, and an improvement in the person’s understanding remain different outcomes. Returning to HCI–VSD made this distinction consequential. If intellectual agency is a design value, the evaluation should ask what the person can understand, challenge, and do afterwards, as well as whether the system produced something persuasive during the session.
Asking what was still unresolved
At this stage I explicitly asked how far we had moved from the original topic and what the core question now was. I also asked whether there was already research on it or whether a gap remained. This was a request to orient the inquiry after several changes of scale. We had discussed individual habits, the organisation of attention, commercial incentives, ethical design, the formation of values, and the possibility of intellectual work with AI. Having accumulated those connections did not automatically tell me which question was most important or where further work would make a contribution.
The literature gave several parts of the inquiry an existing context. Marin examines social networking platforms as environments that can support or undermine epistemic agency—the capacities involved in using information and participating in knowledge practices (Marin, 2022). Buijsman and colleagues examine how AI decision support can preserve human autonomy, including competence and the formation of values (Buijsman et al., 2025). These frameworks do not experimentally establish that my preferred use of AI produces better thinkers. They do show why discovering a question through personal experience should be distinguished from being the first person to formulate it.
I then asked again: if so many of these questions had already been considered, what might still be new in our discussion, and what future directions could matter? This was different from the earlier request for relevant research. A map of existing work helped me identify the subject, but I was now asking what to do with that map. I did not want a declaration of novelty simply because several familiar concepts had been placed together. I wanted a question with enough substance to justify pursuing it.
The assistant returned to the problem of distraction while waiting for an AI response and cited research on short response delays. I rejected that as the main direction. Earlier, waiting had been my example of a situation in which I drifted toward the phone. Now I offered a possible response myself: write down my thoughts for the next round while the system was working. I did not report that I had tested and solved the whole habit. I judged this particular difficulty to be manageable and insufficiently fundamental for the inquiry I wanted to pursue. The same example had first helped explain the problem and later helped me set a boundary around the research question. My proposal also supplied the positive action I had been asking for: writing during a pause would keep me engaged with the question. That left the larger concerns about attention and design intact. A useful practical idea and a significant research direction need not be the same thing, and I asked the assistant to continue toward questions I considered more important.
What the comparison with AI opened up
One suggestion did interest me: “intellectual depth might itself become a means of holding attention.” It introduced a possibility that the original contrast with social media had left open. An engaging system could hold someone through coherent, demanding discussion as well as through repeated novelty. I wanted to consider that possibility, but it did not follow that deep thinking had become suspect. An extended period of thought can be exactly what I want. The question would concern the relationship between that engagement, my purposes, and my ability to reconsider its direction. The duration alone would not tell me whether the activity had served me.
I also found a second suggestion interesting: “private intellectual gains could coexist with losses to shared knowledge.” This introduced a different scale of evaluation. My initial comparison had concentrated on what happened to the person using the system. Here, an interaction could benefit that person while changing what remained available to others. I asked whether there were still more fundamental questions beyond these suggestions. That request should be preserved alongside my interest in them: neither had yet become the settled conclusion of the conversation.
The question about shared knowledge complicated my enthusiasm about AI. An individually useful exchange might remain between one person and a system. A public question could receive corrections from several people, remain available to later readers, and become part of a community’s accumulated understanding. If some of that activity moved into private conversations, the consequences would extend beyond the quality of the answers received by individual users. This also made the broad dismissal of online social activity harder to sustain: the public exchanges through which people develop and correct knowledge were among the activities whose possible loss now concerned me.
There is evidence relevant to this concern. Del Rio-Chanona and colleagues estimated that Stack Overflow activity declined by about 25 per cent relative to comparison platforms within six months of ChatGPT’s release. They found declines among more and less experienced contributors, without a significant change in their peer-feedback measure of post quality (del Rio-Chanona et al., 2024). Stack Overflow is a question-and-answer community, so this finding does not establish the effects of replacing an Instagram feed with AI dialogue. It supports a more specific concern about public contribution. The observational comparison did not measure the quality of the private conversations that might replace participation or demonstrate a corresponding loss of human understanding.
The finding nevertheless made the question more concrete. Public knowledge includes practices of questioning, attribution, correction, and disagreement, as well as the information that results. Choi’s argument for protecting Wikipedia as a knowledge commons similarly directs attention toward participation and collective stewardship in the age of generative AI (Choi, 2026). A system could be useful to an individual while changing the conditions under which those shared practices continue. That possibility deserved investigation alongside the benefits I associated with sustained private inquiry.
Bringing this question into the article gives the writing a further purpose. Turning a private conversation into an essay can make an inquiry available to others, provided its evidence and uncertainties remain visible. Publication does not make it reliable by itself. Readers need to distinguish my experience from a research finding, an AI suggestion from something I accepted, and a possible explanation from an established conclusion. Preserving the history of thinking offers something open to examination: others can challenge a transition in the argument instead of encountering only the conclusion produced after its difficulties have been removed.
Keeping deep thinking at the centre
The distinction between an interesting possibility and my central concern became necessary again during the writing. A proposed title, “When Deep Thinking Is No Guarantee of Freedom,” made one speculative branch of the discussion sound like its destination. I objected. Deep thinking remained necessary to what I wanted to do; my focus was social media and distraction, followed by a comparison with AI. A compelling formulation could give the wrong weight to what I had said. Correcting the title was therefore also a correction of the argument. Interest in a suggestion did not authorise turning it into the position from which the entire inquiry should be narrated.
Even after that correction, I asked for another revision because too much of the interaction had disappeared. The draft retained many topics but compressed the iterations through which I had reached them. It mentioned active choice without fully showing how I returned to it after practical advice and again after structural criticism. It presented a comparison with AI without giving enough space to my questions about what was already researched, what might be new, and why I rejected one proposed direction. My objection was that these exchanges were part of the thinking itself. Removing them could make a provisional, contested development appear to have been an orderly sequence of conclusions.
This experience places a concrete condition on the benefits I associate with AI dialogue. The assistant can help articulate connections, but I still have to judge whether a connection answers my question, whether an example deserves to become a research agenda, and whether a fluent reconstruction remains faithful to what happened. My contribution did not consist only in supplying prompts or approving the final prose. It included recognising a pattern, challenging the relevance of evidence, introducing texts, reopening unresolved questions, rejecting a minor direction, and correcting the account of the inquiry. Those acts were how I kept the conversation connected to its purpose.
The questions I would now take forward concern how digital environments affect the development and continuity of attention. What helps a person move from noticing something to understanding it? Which forms of support help someone choose deliberately when the competing activity remains available? How could design be evaluated through the capacities a person develops, as well as through what happens during a session? A comparison between a feed and an AI dialogue would need to examine the user’s purpose, what they could explain or apply afterwards, and what the exchange contributed to shared knowledge. These are directions for further investigation. The conversation does not establish their novelty or answer them.
I have not resolved my checking habit simply by giving it a more elaborate account. I still want the practical freedom I wanted at the beginning: to have the phone nearby, leave the applications available, and continue with the work I have chosen. The discussion has made that wish easier to articulate, while leaving real disagreements and uncertainties in place. I want enough continuity to develop a question, enough judgment to challenge an attractive answer, and enough control to decide when to continue and when to stop. My insistence on restoring this conversation’s history belongs to the same concern. I want the sequence through which I thought something to remain available for thought, rather than disappear behind the next convincing formulation.
References
Buijsman, Stefan, Sarah E. Carter, and Juan-Pablo Bermúdez. 2025. “Autonomy by Design: Preserving Human Autonomy in AI Decision-Support.” Philosophy & Technology 38, article 97. https://doi.org/10.1007/s13347-025-00932-2.
Choi, Helen. 2026. “Towards a Practice of Knowledge Commoning: Academics’ Ethical Duties to Defend Wikipedia in the Age of Generative AI.” AI and Ethics 6, article 511. https://doi.org/10.1007/s43681-026-01373-z.
del Rio-Chanona, R. Maria, Nadzeya Laurentsyeva, and Johannes Wachs. 2024. “Large Language Models Reduce Public Knowledge Sharing on Online Q&A Platforms.” PNAS Nexus 3 (9): pgae400. https://doi.org/10.1093/pnasnexus/pgae400.
European Commission. 2026. “Commission Preliminarily Finds TikTok’s Addictive Design in Breach of the Digital Services Act.” February 6. Official announcement.
Gallagher, Brian. 2017. “Modern Media Is a DoS Attack on Your Free Will.” Interview with James Williams. Nautilus, September 20. https://nautil.us/modern-media-is-a-dos-attack-on-your-free-will-236806.
Hartanto, Andree, et al. 2024. “The Effect of Mere Presence of Smartphone on Cognitive Functions: A Four-Level Meta-Analysis.” Technology, Mind, and Behavior 5 (1). https://doi.org/10.1037/tmb0000123.
Lindström, Björn, Martin Bellander, David T. Schultner, Allen Chang, Philippe N. Tobler, and David M. Amodio. 2021. “A Computational Reward Learning Account of Social Media Engagement.” Nature Communications 12, article 1311. https://doi.org/10.1038/s41467-020-19607-x.
Marin, Lavinia. 2022. “How to Do Things with Information Online. A Conceptual Framework for Evaluating Social Networking Platforms as Epistemic Environments.” Philosophy & Technology 35, article 77. https://doi.org/10.1007/s13347-022-00569-5.
Oulasvirta, Antti, Tye Rattenbury, Lingyi Ma, and Eeva Raita. 2012. “Habits Make Smartphone Use More Pervasive.” Personal and Ubiquitous Computing 16: 105–114. https://doi.org/10.1007/s00779-011-0412-2.
van de Poel, Ibo. 2021. “Design for Value Change.” Ethics and Information Technology 23: 27–31. First published online in 2018. https://doi.org/10.1007/s10676-018-9461-9.
Williams, James. 2014. “Orwell, Huxley, Banksy.” Oxford Internet Institute, May 24. https://www.oii.ox.ac.uk/orwell-huxley-banksy/.
Williams, James. 2018a. Stand Out of Our Light: Freedom and Resistance in the Attention Economy. Cambridge: Cambridge University Press. https://doi.org/10.1017/9781108453004.
Williams, James. 2018b. “Facebook’s Fundamental Problem? Mark Zuckerberg Can’t Innovate.” WIRED, June 12. https://www.wired.com/story/alternative-business-models-for-facebook/.
Williams, James. 2018c. “Technology Is Driving Us to Distraction.” The Observer/The Guardian, May 27. https://www.theguardian.com/commentisfree/2018/may/27/world-distraction-demands-new-focus.