Human Limits in Digital Systems

Why I am researching this

Digital systems have become extraordinarily good at removing the effort it takes to reach, interrupt, assist and act for another person. Most of that progress is genuinely useful. But it leaves a question that becomes more pressing the better the technology works: once we can remove a human limit, how do we decide whether we should?

This page collects the evidence behind two essays — The Right to Be Unavailable and The Value of Friction. They are separate arguments, but they share one proposition:

Human agency increasingly depends on deliberately putting some friction back.

Three working definitions

Availability

A condition in which another actor can legitimately expect access to a person's attention, response, labour, judgement or agency within a particular period.

The right to be unavailable

The principle that technical reachability must not automatically create an obligation of immediate human response, attention or action.

Friction

Any effort, delay, constraint, verification step or loss of convenience between an intention and its completion.

The first definition is deliberately narrower than reachability. A phone number makes someone reachable. It does not, by itself, entitle anyone to their attention. Yet once software displays “online”, “active” or “seen”, an inner human condition becomes a visible signal, and a quiet conversion can take place:

reachable → noticed → available → expected to answer → accountable for not answering

The right to be unavailable is broader than the legal right to disconnect, which is an employment-law mechanism about work outside working time. The broader principle also applies to friendship, family, public life, religious observance and, increasingly, relationships mediated by AI.

The definition of friction is neutral on purpose. A password-reset loop and a confirmation before deleting a database are both friction. What separates them is what they protect, who carries the cost and what happens if they disappear.

From reachability to delegated availability

The history of communication technology can be read as the steady removal of constraints on access to another person's attention:

1. The telephone (1870s). Synchronous reachability, tied to a fixed place.
2. Handheld mobile telephony (from 1973). Reachability begins to follow the person rather than the building.²
3. Digital messaging. Requests persist when sender and recipient are not present at the same time.
4. Presence (2000). The Internet Engineering Task Force formally models “presence” and instant messaging: availability becomes data a system can distribute.¹
5. Smartphones and workplace platforms. Notifications and presence become ambient infrastructure for work.
6. AI agents. Tasks and replies can continue while the responsible person is absent.

Each step solved a real problem. Together they expose a category error: technical availability is not human availability. A device can receive a message while its owner lacks the time, attention, capacity, responsibility or wish to respond. And as AI agents make people and organisations look permanently responsive, it becomes more important — not less — to say when human judgement is actually available.

Five kinds of friction

Two mistakes sit on either side of this table. Optimisation fundamentalism treats all friction as waste. Friction romanticism treats all difficulty as virtue. Neither survives contact with the evidence. A useful test for the last row: if removing the friction changes not just how easily something is done, but what it is, the friction may be constitutive.

Three dimensions of availability

The most important cell is technically available but humanly unavailable: online while praying, caring for a child, grieving, concentrating, sleeping or simply choosing not to engage. Current tools expose far more about technical state than about human capacity, which invites others to infer one from the other.

A useful category here is asynchronous availability: others may communicate now, while the recipient keeps control over when attention is given. Asynchronous tools only work when sending a message is decoupled from expecting a reply. Without that cultural decoupling, asynchronous technology simply makes synchronous pressure permanent.

What the evidence supports — and what it doesn't

The scientific case for limits is strongest when it is made carefully. Popular claims such as “every interruption costs 23 minutes” or “decisions drain a fixed daily battery” go further than the research safely allows.

Three implications follow. First, attention is not the same thing as time: a thirty-second interruption can cost more than thirty seconds because the task has to be rebuilt. Second, “people can simply ignore it” is not a complete account of notification cost. Third, a system can preserve visible output by consuming invisible human slack — faster working, skipped recovery and evening catch-up rather than lower output.

The case for these essays does not need decision fatigue. It rests on better-established findings: interruption, switching, attentional capture, recovery, telepressure and workload.

Organisations: the problem is expectation

Organisations often treat availability as personal self-management: turn off notifications, set a status, protect your time. The evidence suggests that this is not enough.

Eurofound surveyed workers and HR managers in Belgium, France, Italy and Spain.³ More than 80% of workers received work communications outside their contracted hours in a typical week, and almost nine in ten replied to at least some of them. Their reasons reveal the mechanism:

The most useful finding is a counter-intuitive one. Having a right-to-disconnect policy did not appear to reduce the likelihood of being contacted or of replying. Yet workers in companies with a policy reported better work–life balance (92% versus 80%) and less stress or anxiety (28% versus 38%), and over 70% judged the policy's impact positive. Because this is survey evidence, the policy may be a marker of healthier organisations rather than the cause. Eurofound itself concludes that a policy alone is insufficient: it needs awareness, training, an honest look at workload, effective limits and regular review.³

That points to a distinction I find useful:

• Policy availability asks what the handbook says.
• Cultural availability asks what senior people actually do, reward and punish.

A leader who says “no need to reply tonight” while rewarding the people who do has created coerced availability without writing it down. The design variable that matters is expected response time, not message delivery time.

The legal picture

There is no freestanding international “right to be unavailable”. It is better understood as an emerging principle built on established rights to rest and limited working time.

• International baseline. Article 7(d) of the International Covenant on Economic, Social and Cultural Rights recognises rest, leisure, reasonable limitation of working hours and periodic paid holidays.¹⁵
• France. The Labour Code makes the full exercise of the right to disconnect, and company measures regulating digital tools, a subject of workplace negotiation — to protect rest, leave and personal and family life. Without an agreement, the employer must draw up a charter and provide training and awareness on reasonable use of digital tools.¹⁶ France did not “ban email after hours”; it made the boundary something organisations must govern.
• Ireland. A statutory Code of Practice sets out how employers and employees should approach the right to disconnect.¹⁸
• European Union. In July 2025 the European Commission launched a second-stage consultation of social partners on a possible EU initiative on the right to disconnect and fair telework.¹⁷ At the time of writing I have not found an adopted EU-wide right-to-disconnect law.

The defensible conclusion is not “there is a human right to ignore messages”. It is that digital reachability does not erase existing rights to rest, and that labour institutions are being pushed to spell out what this means when work can reach people continuously.

What religious traditions add

Religious practices should not be reduced to productivity techniques. Their meanings are theological, ethical and communal. But they show something the optimisation mindset tends to forget: societies can treat limits, interruption of ordinary activity and non-market time as positive institutions rather than defects.

The Torah's Sabbath establishes a recurring cessation of work, and Deuteronomy extends that rest to servants, strangers and animals, linking it to the memory of liberation from slavery.¹⁹ Rest here is not a privilege for those with enough autonomy to claim it; it is collective and extends across hierarchy.

The Qur'an describes prayer as prescribed at appointed times; calls believers to leave trade for Friday congregational prayer; prescribes the Ramadan fast; and refers to iʿtikāf, a period of devotional seclusion in the mosque.²⁰ Commerce and ordinary schedules are paused because another value is given priority.

The lesson I take is modest but important: a boundary is stronger when what lies outside it does not have to justify itself in the language of productivity. Rest does not need to make Monday more efficient to matter.

Design: notifications are governance

A notification system repeatedly decides whose intention wins: the sender's wish to communicate, or the recipient's wish to continue what they were doing. Its defaults are therefore not neutral.

Mainstream tools now treat being unavailable as a normal state rather than a connection failure. Apple's Focus lets people choose which people and apps can interrupt, schedule Focus and share that notifications are silenced.²¹ Slack can pause notifications on a schedule, show that state to colleagues and still allow a deliberate override for urgent messages.²² Microsoft Teams offers recurring quiet hours and quiet days on mobile.²³ These are design examples, not evidence of outcomes: their documentation describes functionality, not effects on wellbeing.

Presence indicators solve a real coordination problem, but a system can observe connection state much more easily than it can know human readiness. “Active” may mean a laptop is open. It does not mean interrupt me. A better status communicates expectations rather than surveillance: “Routine replies tomorrow” says more than a green dot.

AI changes both sides of the boundary. It can multiply interruptions, because the cost of asking for someone's attention approaches zero. And it can create false availability: an agent can reply on someone's behalf, making an organisation look responsive when no responsible person has looked at the matter. I would distinguish three states — human available, agent available and human judgement required — and hold to one principle: AI may extend service availability without silently extending claims about human availability.

Two cases worth keeping

Guardrailed AI tutoring

The Bastani field experiment is unusually useful because it compares two designs of the same help.¹⁴ The lesson is not “make AI difficult”. It is: preserve the part of the difficulty that produces the capability the person came to build. Ease of performance and growth of capability are different optimisation targets.

A shared day of rest from screens

The National Day of Unplugging grew out of tech-free Shabbat dinners into an annual 24-hour communal disconnection. Its organisers report thousands of community partners; in their own survey, over 90% said they were highly likely to take part again and 62% said their community members were more likely to unplug afterwards.²⁴ These are organiser-reported figures, not an independent evaluation. The case matters conceptually: it turns disconnection from individual discipline into a shared ritual — and it is easier to be unavailable when the people around you agree that unavailability is legitimate.

A test for any friction

Rather than asking whether friction is good or bad, I ask four questions:

1. What value does it serve — safety, consent, learning, trust, something constitutive, or nothing?
2. Who bears the cost, and who receives the benefit?
3. Who controls it, and can it be escalated without destroying the boundary?
4. What would happen if it disappeared?

If the answer to the first is “nothing”, remove it. If the cost falls disproportionately on disabled, lower-power or time-constrained people, redesign it. If it protects something real and the burden is proportionate, keep it on purpose — and keep measuring.

Remove friction that wastes human capability. Preserve friction that protects attention, consent, safety, learning, relationships, dignity or meaning. Make availability explicit rather than inferred.

Open questions

• Causal evidence on right-to-disconnect rules is thin. Most findings are cross-sectional; longitudinal and quasi-experimental studies around policy changes are needed.
• We know more about interruption than about legitimate availability. How reachability turns into obligation is barely studied.
• Presence is technically mature but socially under-theorised. Connection state, willingness and legitimate access are still blurred.
• AI agents are a new frontier. Do they raise expectations of human responsiveness, hide working-time boundaries — or absorb routine demands and protect people?
• Beneficial friction needs a fairness lens. A step that protects one person can exclude another.
• AI-and-learning evidence is promising but early. One domain and one population cannot settle it.

Method and limits

This is a research-led synthesis of published studies, legal texts, religious sources and product documentation, prepared in September–October 2026. It is not a systematic review. I checked each source listed below against its original publication; where only an abstract or summary was available, I say so. Product documentation describes features, not outcomes. The taxonomies and the four-question test are my own proposals.

Sources

1. Day, M., Rosenberg, J. & Sugano, H. (2000). A Model for Presence and Instant Messaging (RFC 2778). Internet Engineering Task Force. rfc-editor.org. ↩
2. Federal Communications Commission. A Short History of Radio. Notes Martin Cooper's 1973 handheld cellular call. fcc.gov (PDF). ↩
3. Eurofound (2023). Right to disconnect: Implementation and impact at company level. eurofound.europa.eu. ↩
4. Stothart, C., Mitchum, A. & Yehnert, C. (2015). The attentional cost of receiving a cell phone notification. Journal of Experimental Psychology: Human Perception and Performance, 41(4), 893–897. doi:10.1037/xhp0000100. ↩
5. Altmann, E. M., Trafton, J. G. & Hambrick, D. Z. (2014). Momentary interruptions can derail the train of thought. Journal of Experimental Psychology: General, 143(1), 215–226. doi:10.1037/a0030986. ↩
6. Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7(3), 134–140. doi:10.1016/S1364-6613(03)00028-7. ↩
7. Mark, G., Gudith, D. & Klocke, U. (2008). The cost of interrupted work: More speed and stress. Proceedings of CHI 2008, 107–110. doi:10.1145/1357054.1357072. ↩
8. Barber, L. K. & Santuzzi, A. M. (2015). Please respond ASAP: Workplace telepressure and employee recovery. Journal of Occupational Health Psychology, 20(2), 172–189. doi:10.1037/a0038278. ↩
9. Tedone, A. M. (2022). Keeping up with work email after hours and employee wellbeing. Occupational Health Science, 6(1), 51–72. doi:10.1007/s41542-021-00107-3. ↩
10. Santos et al. (2023). Information and communication technologies-assisted after-hours work: A systematic literature review and meta-analysis of the relationships with work–family/life management variables. Frontiers in Psychology, 14. doi:10.3389/fpsyg.2023.1101191. ↩
11. Carter, E. C., Kofler, L. M., Forster, D. E. & McCullough, M. E. (2015). A series of meta-analytic tests of the depletion effect. Journal of Experimental Psychology: General, 144(4), 796–815. doi:10.1037/xge0000083. ↩
12. Hagger, M. S. et al. (2016). A multilab preregistered replication of the ego-depletion effect. Perspectives on Psychological Science, 11(4), 546–573. doi:10.1177/1745691616652873. ↩
13. Roediger, H. L. & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255. doi:10.1111/j.1467-9280.2006.01693.x. ↩
14. Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö. & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. PNAS, 122(26), e2422633122. doi:10.1073/pnas.2422633122. ↩
15. United Nations (1966). International Covenant on Economic, Social and Cultural Rights, Article 7. ohchr.org. ↩
16. Code du travail (France), Article L2242-17, 7°. legifrance.gouv.fr. ↩
17. European Commission (25 July 2025). Commission starts second-stage talks with social partners on right to disconnect and fair telework. ec.europa.eu. ↩
18. Workplace Relations Commission (Ireland). Code of Practice for Employers and Employees on the Right to Disconnect. workplacerelations.ie (PDF). ↩
19. Exodus 20:8–11; Deuteronomy 5:12–15. Exodus (Sefaria) · Deuteronomy (Sefaria). ↩
20. Qur'an 4:103; 62:9; 2:183–187. 4:103 · 62:9 · 2:183–187 (quran.com). ↩
21. Apple. Set up a Focus on iPhone. support.apple.com. ↩
22. Slack. Pause your Slack notifications. slack.com. ↩
23. Microsoft. Quiet time in Microsoft Teams for mobile devices. support.microsoft.com. ↩
24. Reboot. News from the National Day of Unplugging. rebooting.com. Organiser-reported figures. ↩