Exocrasis

1The environment, not the screen

Much of the public debate about computers and the mind asks one question about one object: does the screen damage attention? Both the object and the question are too narrow.

The object is too narrow because a computer is not a screen. It is a keyboard that turns language into keystrokes, a pointer that turns intention into position, a memory that holds what we no longer need to, a processor that calculates for us, and a network that connects all of it to everyone else. Typomora, an earlier paper in this series, dealt with one part of that environment, the input layer, and the shift from typing back to voice. This paper deals with the rest.

The question is too narrow because "harm" is the wrong frame. McLuhan's tetrad asks four questions of any medium: what it enhances, what it pushes out, what it retrieves, and what it reverses into when pushed to an extreme. A computer can make some capacities stronger and others weaker at the same time. That is not a contradiction to be resolved. It is what a medium does.

The argument here is that the computing environment moves part of the configuration of the senses, and of memory, outside the body. We call this exocrasis. Memory that once had to be held now sits in a device. Navigation that once required an inner map now follows a voice. Attention that once set its own pace now meets an environment that sets the pace for it. In each case, something is extended outward and something internal is used less. McLuhan's word for that pairing was extension and amputation.

One methodological point shapes everything that follows. Most research on computers and cognition measures performance on a task in a lab: the figure, in McLuhan's terms. But the claim here is about the ground: the environment people live inside for hours a day, for years. Lab tasks are blind to the ground almost by design. The paper therefore gives special weight to studies that measure whole environments, and it is explicit about which findings come from which kind of study.

2Extension and amputation, measured

McLuhan said every extension is also an amputation. Research on what psychologists call cognitive offloading has made that claim measurable.

Offloading as a choice

Risko and Gilbert (2016)1 defined cognitive offloading as using physical action or external tools to reduce the mental demand of a task: writing a note, setting a reminder, letting a phone hold a number. They framed it as a cost-benefit decision people make, often without noticing. Offloading usually improves performance on the task at hand. The question is what it does to what is left inside.

What is lost: photographs

The clearest evidence comes from photography. In museum studies, people remembered objects they photographed less well than objects they only looked at (Henkel, 2014)2. The effect has replicated (Soares & Storm, 2018) and extends to screenshots. The leading explanation is not distraction but release: once something is saved externally, the mind lets it go.

Saving is not only a loss. In other experiments, saving one set of information freed memory for the next (Storm & Stone, 2015)3. Offloading is a reallocation: less held of what is stored, more room for what comes after. That is the tetrad in miniature, an enhancement and an obsolescence in the same act.

What is lost: the inner map

Navigation shows the effect over years. People who rely on GPS walk further, stop more and draw worse maps of routes than people using paper maps (Ishikawa and colleagues, 2008)4. In a study by Dahmani and Bohbot (2020)5, habitual GPS users had worse spatial memory when navigating on their own, and in a three-year follow-up of a smaller group, more GPS use predicted a steeper decline, not explained by a poor sense of direction to begin with. The follow-up sample was small, and the finding needs replication. But it is the clearest evidence that an extension, used for years, changes the capacity it replaced.

The newest extension

Artificial intelligence extends thinking itself. A widely reported 2025 study from MIT found that people writing essays with ChatGPT showed lower brain connectivity, remembered their own essays less well, and felt less ownership of them (Kosmyna and colleagues). The study was a preprint, used typed rather than spoken interaction, and has drawn a published methodological critique. It should be treated as a question, not a finding. But the question is the same one GPS and photographs raise: what happens to a capacity once a machine reliably does it for us?

3Reading by the screen's rules

The best-established finding about screens and cognition concerns reading. A meta-analysis of 54 studies with more than 171,000 participants found that people understood text slightly less well on screens than on paper (Delgado and colleagues, 2018)7. The overall effect was small (g = −0.21), larger for expository text such as articles and textbooks (around −0.32), absent for narrative, and stronger under time pressure. Two other meta-analyses reached almost the same size (Clinton, 20198; Kong and colleagues, 2018).

Two details matter more than the average.

First, the gap grew over the period studied, by about 0.01 standard deviations a year between 2000 and 2017, and it showed no difference between younger and older participants. Younger people, who grew up with screens, did not read better on them. If familiarity were closing the gap, the opposite would be expected.

Second, the mechanism appears to be calibration rather than the display. Readers on screens tend to be more confident in their understanding than their results justify (Ackerman & Goldsmith, 2011), and under time pressure they fail to slow down. One experiment found that screen readers under time pressure did not reduce their mind-wandering and understood less as a result (Delgado & Salmerón, 2021)9. The screen does not make reading impossible. It sets a faster, shallower default, and readers follow it.

This is the clearest case of the environment setting a mode. In the terms of this series, it is a setting of gain and persistence: a lower-effort processing level applied by default, and a habit of not holding text long enough for it to settle. The screen is not only where reading happens. It sets the rules for how reading happens there.

The argument that sustained, linear deep reading is a cultivated capacity that can weaken with disuse is made most fully by Maryanne Wolf (Reader, Come Home, 2018)10. Her synthesis draws on the evidence above. Whether screen habits carry over to reading on paper, as she suggests, is less well established.

4Enhancement toward the medium

If the computing environment only weakened capacities, the case would be simple. It does not. Some capacities get stronger, and which ones is revealing.

Action video games are the clearest case. A meta-analysis of the research found that people who play them perform better on tests of perception, spatial cognition and top-down attention. The difference between players and non-players was moderate (g = 0.55), and, more importantly, training non-players on action games produced a smaller but real improvement (g = 0.34) (Bediou and colleagues, 2018)11. Because the training studies assign people to play, the gain is at least partly caused by the games, not only by who chooses to play them. A later update found similar effects (g = 0.30 for training).

Gamers also judge the timing of sights and sounds more precisely. In a test of whether a flash and a beep happened at the same moment, players showed sharper multisensory timing than non-players (Donohue, Woldorff & Mitroff, 2010)12. Part of that advantage may come from sharper visual timing alone rather than better integration of the senses.

What is enhanced is telling. Fast visual tracking, rapid shifts of attention and precise timing are exactly what screens and games demand. The environment trains attention for itself. This is not evidence against an exocrasis; it is the enhancement half of the tetrad. An environment that strengthens the capacities it uses, while others go unpractised, is doing what McLuhan said every medium does.

This reframes a common argument. People sometimes cite gaming research to show that screens are good for the mind. The research does show real gains. But a gain toward the medium is not a neutral improvement; it is a configuration tuned to the environment that produced it. The question that matters is not whether screens make people better or worse, but what they tune people toward, and what falls out of practice to make room.

5Capture, and measuring the ground at scale

Lab studies measure the figure. To see the ground, a study has to change or compare whole environments. A small number of studies have done this, mostly in economics, and they give the strongest causal evidence available.

Changing the environment

Facebook launched at different US colleges at different times between 2004 and 2006. Braghieri, Levy and Makarin (2022)13 used that staggered rollout as a natural experiment. Where Facebook arrived, students' mental health got worse, by about 0.085 standard deviations on a poor-mental-health index. That is roughly a fifth of the effect of losing a job, and the mechanism they identified was unfavourable social comparison.

Allcott and colleagues (2020)14 ran the reverse experiment: they paid people to deactivate Facebook for four weeks. Subjective well-being rose by about 0.09 standard deviations, and people spent less time online overall. A later study of Facebook and Instagram deactivation found effects about 40% smaller.

Broadband offers a third natural experiment. In Italy, cohorts exposed to high-speed internet in adolescence were more often diagnosed with mental disorders, by about 0.08 standard deviations (Donati and colleagues, 202215, a working paper).

Capture by design

Allcott, Gentzkow and Song (2022)16 asked why people use social media more than they say they want to. In a randomised experiment with nearly 2,000 US adults, temporary incentives to cut use had lasting effects, and self-set limits worked. Their model attributes about 31% of social media use to self-control problems. That is a measurable estimate of capture: use the person would not choose on reflection, driven by habit and design.

Short-form video shows a narrower mechanism. In a lab study of 60 people, interruptions by TikTok-style videos impaired prospective memory, the ability to remember to do something later, while longer-form videos and rest did not (Chiossi and colleagues, 2023)17. The likely cause is rapid context switching. Later studies have conceptually replicated the effect.

The scale of the effects

These effects are small per person, around a tenth of a standard deviation. Across hundreds of thousands of adolescents, digital technology use explains at most 0.4% of the variation in well-being (Orben & Przybylski, 2019)18. Any claim that screens transform individual minds dramatically has to reckon with that.

But small per person does not mean small. These effects apply to entire populations, every day, for years. McLuhan's point about media was never that they hit individuals hard. It was that they change the environment everyone lives in, and that the change is hard to see precisely because it is everywhere.

6What does not hold up

Several of the most widely repeated findings about computers and the mind have weakened or failed under replication. A theory of exocrasis is stronger for setting them aside.

The "Google effect." Sparrow, Liu and Wegner (2011, Science)19 reported that difficult questions made people think of computers, and that people remember where information is stored rather than the information itself. The first finding failed in the large Social Sciences Replication Project (Camerer and colleagues, 2018), and a further replication found no conclusive evidence for it. The broader idea of offloading survives on other evidence (section 2); this particular demonstration does not.

"Brain drain." Ward and colleagues (2017)20 reported that the mere presence of one's own smartphone reduced working memory capacity. A pre-registered direct replication failed, and the largest meta-analysis, of 33 studies and over 4,000 people, found no effect of mere presence (d = −0.02). Effects of actually using a phone, or of notifications, are better supported than effects of its presence.

Laptops versus longhand. Mueller and Oppenheimer (2014)21 reported that students who took notes by hand understood lectures better than those using laptops. A pre-registered direct replication found that laptop notes were longer and more word-for-word, as the original said, but found no difference in learning (Urry and colleagues, 2021).

Handwriting and brain connectivity. A 2024 EEG study reported widespread brain connectivity during handwriting but not typing (Van der Weel & Van der Meer). Critics noted that participants typed with one finger, that typing skill was not measured, and that connectivity is not the same as learning (Pinet & Longcamp, 2025).

School phone bans. The best observational study so far, of more than 1,200 students in 30 English schools, found no association between restrictive phone policies and well-being, sleep or attainment. Students used their phones about 30 minutes less at school but made up for it outside (Goodyear and colleagues, 2025)23.

"Digital natives." The idea that young people raised with technology think and learn differently has little empirical support. Reviews describe it as closer to a moral panic than a finding (Bennett, Maton & Kervin, 200824; Kirschner & De Bruyckere, 2017).

Why this matters

The failures share a pattern. Most measured a single, short exposure in a lab and claimed a large effect. The findings that hold up either measure whole environments over time or measure a specific, repeated practice such as photographing, navigating by GPS or reading on screens. That is what McLuhan's distinction between figure and ground would lead one to expect. Brief exposures to a figure tell us little. The ground acts slowly, through repetition, and has to be measured at the scale it operates on.

The school-ban result fits the same logic. A partial restriction inside a total environment changes little, because the environment is the whole day.

7Exocrasis across the five dimensions

The table maps the evidence onto the five dimensions of sensory syncrasis. It is interpretation: none of these studies used the framework, and each row names what it leans on.

The inside–outside row carries the paper's name. In the earlier papers, the inside–outside balance meant attention to the body versus the world. In exocrasis, more than attention moves outward: storage does. Part of the configuration that used to live in the person now lives in the environment, and the person's own capacity adjusts to its absence.

8Objections and replies

"The effects are tiny." Per person, yes: around a tenth to a third of a standard deviation, and well-being associations at most 0.4% of variance. The paper does not claim dramatic individual transformation. Its claim concerns environments: small effects applied to everyone, daily, for years, which is the scale at which McLuhan said media act.

"Offloading is just efficiency." Often it is, and saving can free memory for new material (Storm & Stone, 2015)3. The paper treats offloading as reallocation, not loss. The point is that it is a reallocation, with a measurable cost to the offloaded capacity, as the GPS and photography studies show.

"Gaming shows screens make people sharper." It shows that action games strengthen the skills they demand. That is the enhancement half of the tetrad, and the paper counts it as such. It does not show that the environment is neutral.

"Many of these effects come from lab tasks too." Some do. The paper gives most weight to studies of repeated practice and whole environments, and it sets aside the findings that rested on single brief exposures (section 6).

"Social media effects may reflect who uses it, not what it does." That is why the rollout, broadband and deactivation studies matter. They change exposure for reasons unrelated to the person, which is the closest available test of cause.

"This overlaps with Typomora." Deliberately at the edges. Typomora deals with the input layer, the keyboard and voice. Exocrasis deals with storage, processing, reading and network, and with how the environment as a whole moves capacities outward.

"Exocrasis is just the extended mind." Clark and Chalmers's extended-mind thesis argues that cognition can include external tools. Exocrasis accepts that and adds the other side McLuhan insisted on: the internal capacity that adjusts when a tool takes over. Extension and amputation, not extension alone.

*"This is Nicholas Carr's The Shallows again."* Carr argued in 2010 that the internet makes thinking shallower, and he deserves credit for raising the question. Exocrasis differs in three ways. It does not argue harm; it counts enhancements, such as the gaming findings, alongside costs. It sets aside several findings popular arguments have relied on that later failed replication (section 6). And it places the effects within a five-dimensional framework rather than a single dimension of depth.

9Conclusion

The computing environment is not a screen that harms or helps attention. It is an environment that moves part of the mind outside the body. Memory sits in devices, navigation follows a voice, reading follows the screen's pace, and attention meets an environment designed to hold it. Each extension improves something, and each leaves an internal capacity less practised. That is exocrasis.

McLuhan's tetrad gives the shape of it:

The evidence for all of this is real but modest per person, and much of the popular case against screens rests on findings that have not held up. The case that survives is narrower and, for that reason, more serious: a small, steady re-tempering applied to almost everyone, every day, through repeated practice.

The open questions are about the ground. Do capacities lost to offloading return when the device is set aside, and how quickly? Can the screen's default mode for reading be overridden by habit or design? What is the right scale at which to measure an environment that never switches off? And as artificial intelligence takes on thinking itself, the newest and largest extension, what internal capacity will adjust to its presence? That last question is where the next decade of this research will be decided.

References

1. 1Risko & Gilbert (2016), Trends in Cognitive Sciences 20(9): 676–688
2. 2Henkel (2014), Psychological Science; Soares & Storm (2018), J Applied Research in Memory and Cognition 7: 154–160
3. 3Storm & Stone (2015), Psychological Science
4. 4Ishikawa et al. (2008), Journal of Environmental Psychology
5. 5Dahmani & Bohbot (2020), Scientific Reports 10: 6310
6. 6Kosmyna et al. (2025), arXiv:2506.08872, preprint, with published critique (arXiv:2601.00856)
7. 7Delgado, Vargas, Ackerman & Salmerón (2018), Educational Research Review 25: 23–38
8. 8Clinton (2019), Journal of Research in Reading; Kong, Seo & Zhai (2018), Computers & Education
9. 9Delgado & Salmerón (2021), Learning and Instruction
10. 10Wolf (2018), Reader, Come Home
11. 11Bediou et al. (2018), Psychological Bulletin 144(1): 77–110; 2023 update
12. 12Donohue, Woldorff & Mitroff (2010), Attention, Perception & Psychophysics 72: 1120–1129
13. 13Braghieri, Levy & Makarin (2022), American Economic Review 112(11): 3660–3693
14. 14Allcott, Braghieri, Eichmeyer & Gentzkow (2020), American Economic Review 110(3): 629–676
15. 15Donati, Durante, Sobbrio & Zejcirovic (2022), IZA Discussion Paper 15202 (working paper)
16. 16Allcott, Gentzkow & Song (2022), American Economic Review 112(7): 2424–2463
17. 17Chiossi et al. (2023), CHI '23
18. 18Orben & Przybylski (2019), Nature Human Behaviour 3: 173–182
19. 19Sparrow, Liu & Wegner (2011), Science 333: 776–778; Camerer et al. (2018), Nature Human Behaviour 2: 637–644
20. 20Ward et al. (2017), J Association for Consumer Research 2(2): 140–154; Ruiz Pardo & Minda (2022); Hartanto et al. meta-analysis
21. 21Mueller & Oppenheimer (2014), Psychological Science 25(6); Urry et al. (2021), Psychological Science 32: 326–339
22. 22Van der Weel & Van der Meer (2024), Frontiers in Psychology; Pinet & Longcamp (2025), Frontiers in Psychology
23. 23Goodyear et al. (2025), Lancet Regional Health – Europe 51: 101211
24. 24Bennett, Maton & Kervin (2008), BJET 39(5); Kirschner & De Bruyckere (2017), Teaching and Teacher Education 67: 135–142
25. 25Clark & Chalmers (1998), The extended mind, Analysis 58(1): 7–19