Accessibility and Inequality
Linas JuozenasShare
Intelligence Unleashed · Access & equity
Accessibility & inequality in the cognitive-technology era
The digital divide is no longer only about whether a connection appears on a coverage map. It is about whether a person can afford a suitable device, use and understand a service safely, access it as a disabled person, receive support when it fails—and retain a meaningful choice not to use it.
01 · Start with the right question
Who can turn availability into agency?
A headline about network coverage or device distribution can be true while many intended users remain excluded. Equity analysis follows the full route from infrastructure to an opportunity a person can meaningfully use, evaluate and sustain.
Technology can widen opportunity, but it does not distribute opportunity by itself. Existing income, education, language, disability, geography, age, gender, migration status and institutional power shape who receives the best tools, who bears their risks and whose difficulties are mislabelled as “non-adoption.”
Availability
A network, device or service exists in the relevant place. It says nothing yet about price, usability or value.
Accessibility
People with diverse sensory, physical, speech, cognitive and neurological needs can perceive, operate and understand it.
Equality
People receive the same offer. Identical treatment can still reproduce disadvantage when needs and starting conditions differ.
Equity
Resources and accommodations are distributed—and decisions governed—so people have a fair chance to benefit without coercion or avoidable harm.
An important boundary
Access is not a promise of identical outcomes
People differ in goals, preferences and circumstances. Equity does not require every person to use the same technology or produce the same result. It requires a fair route to valuable opportunities, reasonable accommodation, protection from discrimination and a genuine ability to decline. Nor does it imply that every experimental or elective intervention must immediately be universal. Scarce clinical resources still require transparent evidence, safety review and fair allocation.
The phrase cognitive technology is used broadly here: educational platforms, search and generative AI, communication aids, accessible interfaces, telehealth, neurotechnology and other systems that can support learning, memory, communication or decision-making. These are not interchangeable. An AI assistant is an external tool; an augmentative-communication device is assistive technology; an implanted brain–computer interface may be an investigational medical device. Calling all of them “enhancement” hides crucial differences in evidence, risk and entitlement.
02 · The access chain
Eight links must hold
The International Telecommunication Union’s universal and meaningful connectivity framework covers quality, availability, affordability, security, devices and skills. This guide extends that practical lens to accessibility, support and relevance.1
Availability
Coverage reaches the home, school, clinic or workplace—and survives outages, disasters and local congestion.
Quality
Speed, latency, data allowance and reliability fit the task. A text message and a live interpreted consultation need different service.
Affordability
The recurring cost of service, electricity, repair, travel and subscriptions does not crowd out food, housing or care.
Suitable devices
People have private, dependable equipment with the screen, input, storage, battery and software their task requires.
Accessible design
Interfaces work with keyboards and assistive technology, reflow cleanly, expose errors, caption media and avoid needless cognitive load.
Skills & support
Users and frontline staff can operate, evaluate and troubleshoot the system, with human help available when something breaks.
Safety & trust
Privacy, security, redress and clear accountability make participation safe, rather than forcing people to trade intimate data for an essential service.
Relevance & agency
Content fits local language and goals; users can understand automated decisions, leave a provider and choose a human or offline route.
The weakest-link rule: a programme should not call itself inclusive because one input was delivered. Count the people who can complete the intended task, at acceptable quality and cost, over time—and examine who still cannot.
03 · A global snapshot
More people are online, but meaningful use remains sharply unequal.
ITU’s 2025 estimates show enormous progress and a stubborn pattern: the people still offline are concentrated in lower-income economies, rural areas and groups already facing structural disadvantage.2
Read the numbers correctly: ITU describes these as estimates and nowcasts, not a live census. Household surveys can arrive years late, missing demographic values may be modelled and earlier totals are revised as national data improve. They measure whether someone used the internet—not whether the connection was private, affordable, accessible or good enough for a particular task.
| What is measured | Best current signal | What it cannot tell us |
|---|---|---|
| Mobile-network availability | In 2025, 3G-or-better service covered an estimated 96% of the world; 5G covered 55%, ranging from 84% in high-income to 4% in low-income economies.2 | Whether a person owns a compatible device, can buy service or receives the advertised performance. |
| Service affordability | Only 54 low- and middle-income economies met the Broadband Commission’s 2%-of-GNI-per-capita affordability target for at least one benchmark basket in 2025. The median fixed-broadband basket in low-income economies exceeded one-quarter of GNI per capita.3 | ITU expresses benchmark prices as a share of GNI per capita; this does not measure the burden on a low-income household and excludes devices, electricity and repair. |
| Device ownership | ITU estimated that 82% of people aged 10 or older owned a mobile phone in 2025, but only 53% did in low-income economies.2 | Whether it is a smartphone, is privately controlled, can run required software or is suitable for sustained study or work. |
| Digital skills | Only 88 countries had submitted any ICT-skills data to ITU since 2020, and 48 provided comparable skill-level data.4 | A defensible single global “digital literacy rate.” Sparse, differently collected data do not support one. |
| Gendered mobile use | GSMA estimated that women in low- and middle-income countries were 12% less likely than men to use mobile internet in 2025.5 | The experience of every country or every woman. The estimate covers adults, combines surveys and modelling, and differs from ITU’s global measure. |
Different organisations use different denominators and definitions. “People who use any internet,” “mobile internet on one’s own device,” “subscriptions” and “households covered” are not interchangeable.
Regional statistics can fill part of the skills gap without becoming a global estimate. Eurostat reported that 60.4% of EU residents aged 16–74 had at least basic digital skills in 2025—up from 55.6% in 2023, but still below the EU’s 80% target for 2030.6 Quality gaps also persist inside wealthier economies: OECD analysis found median tested fixed-download speeds at the end of 2024 were 43.8% higher in metropolitan regions than in regions far from metropolitan areas.7 Neither statistic describes every locality; both show why national averages conceal who receives a capable connection.
The coverage gap
No adequate network reaches the user
This remains acute in remote, low-density, conflict-affected and disaster-prone places. It requires infrastructure, resilient power, backhaul, spectrum and maintenance—not a new app.
The usage gap
A network exists, but people do not use it
Cost, device access, language, relevance, safety, literacy, documentation requirements and social restrictions can all block adoption. This gap is now larger than the bare mobile-coverage gap.8
The outcome gap
People connect, but benefits remain unequal
Faster devices, quieter rooms, expert support, premium models and institutional authority let some users turn the same nominal service into more learning, income or influence.
These differences accumulate. Better-resourced users can convert faster devices, private space, expert help and premium services into credentials, saved time and greater influence. Excluded users often pay more through travel, failed applications, delays and privacy concessions. Across repeated decisions, the same nominal service can therefore widen differences in learning, income, health and civic power.
04 · Where gaps become consequences
Connection is an input, not an outcome
Broadband and devices can enable learning, work and care. They do not independently cause better grades, higher wages or better health. Instruction, occupation, housing, time, language, disability, institutional support and the design of the tool all mediate what happens next.
Education
A connected school is only the beginning
UNESCO reported that about 40% of primary, 50% of lower-secondary and 65% of upper-secondary schools were connected to the internet for pedagogical purposes. Its wider evidence review warns that technology’s value depends on access, governance and teacher preparation.910
- Home access may still be absent or shared.
- A phone may not support long documents, coding or specialist software.
- Teachers need time, training and authority to adapt a tool.
- Learning must be checked away from the system that supplied the answer.
Work & livelihood
Digital opportunity follows occupational power
Online recruitment, credentials, remote work and AI assistance can expand opportunity, yet the people with the strongest connectivity, recognised qualifications and discretion at work are often best placed to benefit. The World Bank identifies connectivity, compute, locally relevant data and skills as four foundations for inclusive AI participation; these remain concentrated unequally.11
- Application portals can exclude users with inaccessible forms or no stable address.
- Premium AI tiers and powerful hardware can change speed and capability.
- Workers may carry surveillance risk while employers retain productivity gains.
- Task exposure to AI is not the same as access, automation or job loss.
Health & public services
Digital-first must not become digital-only
A portal can save a journey, provide reminders or connect a specialist. It can also place an authentication puzzle, unreadable document or data charge between a person and care. Essential services should test completion with disabled and low-connectivity users, provide assisted support and preserve workable phone, in-person or paper routes.
- Measure completed appointments or claims—not account creation.
- Design for low bandwidth, older devices and interruption.
- Offer interpreters, proxy access and accessible identity checks.
- Never infer “refusal” from a failed interface.
The pandemic stress test
Participation lagged behind remote-learning provision
UNICEF estimated that at least 463 million schoolchildren—31% of those covered by its assessment—could not be reached by digital or broadcast remote-learning policies during school closures. Three quarters of the unreachable children lived in rural areas and 72% lived in the poorest households within their countries.12 Even this was a best-case estimate based on policy and household media ownership, not observed attendance or learning. The lesson is enduring: delivery channels, household conditions and human support must be measured separately.
Skills are distributed unequally
Skills and support turn access into capability
On average across OECD countries, 18% of adults in the 2023 Survey of Adult Skills did not reach even the most basic level in any of literacy, numeracy and adaptive problem solving.13 These results describe performance in a standardised assessment, not inherent ability. They signal a need for clearer services, adult learning and human assistance—not ever more complex gates.
AI can magnify the gap
The “intelligence layer” has its own infrastructure
AI participation depends on data centres and cloud access as well as a household connection. Models perform unevenly across languages and contexts; paid plans may provide stronger models, higher limits or different data-use and retention terms. A World Bank–ILO study estimated that digital-access and infrastructure gaps could hinder nearly half of the jobs it classified as having productivity-enhancing potential from generative AI in Latin America and the Caribbean—about 17 million jobs. That is a regional modelling estimate, not an observed productivity or employment outcome, but it shows why access belongs inside AI policy.14
The policy question is not “Did we distribute technology?” It is “Who could complete a valued task, under what conditions, with what lasting benefit—and how were the costs and risks distributed?”
05 · Disability, access & design
Accessibility is infrastructure
Accessibility is not a specialist polish added after launch. It determines whether education, communication, work and civic participation are available at all—and it often improves reliability and clarity for everyone.
The documented gap
Billions need assistive products; nearly a billion go without
WHO and UNICEF estimated in 2022 that more than 2.5 billion people needed at least one assistive product and nearly one billion lacked access. In some lower-income settings, access was reported as low as 3% of need.15 The category is broad—hearing aids, wheelchairs, spectacles and prostheses as well as apps for communication or cognition—so it should not be presented as a count of digital-device users.
The rights foundation
“Nothing about us without us” must shape the whole lifecycle
The UN Convention on the Rights of Persons with Disabilities addresses equal access to information and communications technologies and calls for accessibility at an early stage. It also requires States Parties to closely consult and actively involve disabled people, through their representative organisations, in relevant decisions.16 Co-design needs budget, decision-making authority and compensation—not a late usability session.
Beyond a compliance badge
Design for real tasks, intersecting needs and failure states
WCAG 2.2 supplies testable, technology-neutral criteria for web content and covers a wide range of visual, auditory, physical, speech, cognitive, language, learning and neurological needs. W3C also says no guideline can address every user need.17
Conformance is a baseline. Real services should test representative journeys with the people expected to use them, including combinations of disability, language, low bandwidth, older equipment and limited support.
The following checklist extends WCAG’s web-content baseline with service delivery, lifecycle support and user agency; it is not W3C’s formal taxonomy.
Perceive
Text alternatives, captions, transcripts, contrast, scalable text and content that does not depend on one sense.
Operate
Keyboard access, visible focus, adequate target sizes and timing, plus alternatives to multipoint or path-based gestures, dragging and device motion.
Understand
Plain language, predictable navigation, clear errors, flexible authentication and instructions that do not overload working memory.
Interoperate
Semantic structure and robust names, roles and states that work with assistive technologies across browsers and devices.
Maintain
Fund assessment, fitting, training, repair, batteries, software updates, replacement and human support across the product’s life.
Respect
Do not define disabled lives only through deficit or cure. Support communication and agency on the user’s terms, including refusal.
| Ask the vendor | Look for | Do not accept as proof |
|---|---|---|
| Which standard and version did you test? | Criterion-by-criterion findings, methods, dates, environments and known defects. | “WCAG compliant” with no scope, report or accountable owner. |
| Can representative users complete priority journeys? | Research with disabled users; completion, error, abandonment and support data by task. | A perfectly scored component library while the end-to-end service remains unusable. |
| What happens after an update? | Regression tests, release gates, remediation timelines and accessible change notices. | A one-time audit performed on an obsolete version. |
| How does the service fail? | Low-bandwidth modes, session recovery, human escalation and equivalent alternatives. | A help centre that requires the same inaccessible login as the failed service. |
In the European Union, Member-State laws transposing the European Accessibility Act began applying on 28 June 2025 to specified products placed on the market and specified consumer services, subject to scope, exclusions and transition rules.18 Covered categories include certain consumer computers and terminals, e-readers, specified electronic communications services, consumer banking, e-books, e-commerce and specified passenger-transport services. It is not accurate to say that the Act covers every product or every website. Compliance and enforcement must be assessed under the applicable national law.
06 · Lessons from real programmes
What changed—and what did not
Useful case studies do not turn implementation figures into proof of social impact. These examples show distinct parts of the access chain; each includes the limit that should travel with the headline.
Rwanda · coverage versus use
A strong network baseline did not remove adoption barriers
A 2021 World Bank appraisal described Rwanda’s 6,000-kilometre national fibre backbone and cited a 97% 4G network-coverage rate from the fourth quarter of 2020. It also cited a 2016–17 household survey in which only 17% of households had internet at home, and a 2021 Giga mapping exercise in which 43% of schools lacked internet access. Device affordability and limited digital literacy remained major barriers.19
Lesson: infrastructure can make use possible without making it affordable, suitable or supported.
Evidence limit: the figures come from different administrative and survey baselines. They are historical, not a 2026 snapshot or a causal evaluation, and they do not support claims about test-score gains.
Detroit · community governance
Residents helped build the network and the knowledge around it
In a 2024 update, the Detroit Community Technology Project reported that its Equitable Internet Initiative had grown from a 45-household pilot launched in 2016 to more than 500 households and small businesses across three neighbourhood networks. It also reported training hundreds of Digital Stewards who helped design and build the networks.20
Lesson: local skills, privacy, consent and control can be designed as infrastructure—not left for users to solve after deployment.
Evidence limit: these are implementer-reported reach and training figures, not an independent count of unique users or proof of effects on education, income or health.
Chhattisgarh · devices and durability
Hardware gains faded; a separate training trial proved more durable
India’s 2018 Sanchaar Kranti Yojana distributed more than two million smartphones to female household heads, built LTE towers in eligible communities and included 1 GB of free mobile data. An evaluation estimated a 56-percentage-point initial increase in women’s reported smartphone ownership and a 10.3-point increase in advanced use. About four years later, women in eligible communities were no more likely than women in comparison communities to own a smartphone or to have used one in the previous 30 days. In a separate randomised trial across 212 villages in one district—all eligible for the phone programme—women invited to digital-skills training were more than four percentage points more likely to report smartphone use three years later.21
Lesson: taken together, the two interventions suggest that one-time hardware can create a sharp short-run gain, while durable access also depends on skills, replacement, social permission and ongoing cost.
Evidence limit: the distribution evaluation and training trial used different samples and designs. The training result came from one district in villages eligible for free phones, so it does not show that training alone will generalise elsewhere. Neither study supports a fixed household-income gain.
United States · the subsidy cliff
Affordability support reached millions, then funding ended
More than 23 million households were enrolled in the Affordable Connectivity Program when new enrolments stopped in February 2024. The programme provided a monthly broadband discount of up to US$30—or up to US$75 on qualifying Tribal lands. Of 5,317 households responding to an FCC survey, 68% reported inconsistent or no connectivity before enrolment; among that group, 80% cited affordability. Seventy-seven per cent said losing the benefit would make them change their plan or drop internet service entirely.22 The programme ended on 1 June 2024 after Congress provided no additional funding.23
Lesson: a subsidy can offset a recurring-cost barrier at scale, but temporary funding does not provide households with durable affordability.
Evidence limit: this was a weighted email survey: 5,317 of 110,000 sampled households responded, so non-response bias remains possible. There was no comparison group, and the 77% combines plan changes with disconnection.
A different kind of public capacity
Map before contracting; verify after connection
UNICEF and ITU’s Giga initiative combines school mapping, connectivity measurement, financing models and procurement support. As accessed on 4 September 2026, its live overview reported more than 2.2 million schools mapped, more than 136,000 schools reporting real-time connectivity and work under way in 54 countries.24 Mapping a school is not connecting it, and a live signal does not prove educational benefit. The value is the operational sequence: establish the denominator, monitor connectivity, model infrastructure needs, mobilise finance and procure against measurable service requirements.
07 · The emerging divide
Four access debates that should not be collapsed
Today’s measurable inequalities concern mainstream digital services and assistive technologies. Restorative neurotechnology remains specialised and often investigational; implantable enhancement for healthy users remains speculative. Treating these as one market obscures very different evidence, risks and claims to access.
1 · AI tools
External software may support translation, planning, writing or learning. Its access problem begins with electricity, connectivity, devices, cost, language, accessibility, skills and institutional support. UNESCO warns that AI in education can make digital exclusion more consequential; access does not prove learning benefit.10
2 · Assistive technology
Products that support communication, cognition, hearing, vision, mobility or self-care meet documented needs now. The almost one-billion-person access gap estimated by WHO and UNICEF deserves attention before hypothetical consumer upgrades dominate the debate.15
3 · Restorative neurotechnology
Implanted BCIs are being studied as medical devices intended to restore or support functions affected by illness, injury or disability. FDA guidance addresses non-clinical testing and Investigational Device Exemption studies for people with paralysis or amputation; it is guidance, not approval of a consumer enhancement product.25
4 · Elective enhancement
Using neurotechnology to improve memory, attention or other capacities beyond medical need remains an anticipatory governance question. There is no robust evidence of a widespread market for safe, effective implantable cognitive enhancement, and no defensible annual price-decline curve predicts when one will arrive.
A scientific boundary
Restoring communication is not “uploading intelligence”
A 2024 study reported encouraging conversational communication from an implanted speech neuroprosthesis in one participant with amyotrophic lateral sclerosis.26 That achievement is profound. It does not show unrestricted thought reading, general cognitive enhancement or population-wide effectiveness. One participant also cannot establish long-term affordability, reliability or access.
A governance boundary
Normative guidance still requires product-specific approval
UNESCO’s Recommendation on the Ethics of Neurotechnology was adopted on 11 November 2025. It calls for equitable access, lower end-user cost, lifecycle continuity, participation by persons with disabilities, a right to refuse, safeguards for neural data and heightened scrutiny of uses beyond medical need.27 It is a global normative framework for Member States—not binding worldwide legislation or evidence that a device is safe.
Price is only the first gate
Assessment, surgery, travel, clinical time, training, connectivity, maintenance, software, rehabilitation, caregiver labour and eventual removal may determine who can use a complex device.
Data can become a condition of access
Subsidised tools may demand intimate behavioural or neural data in return. Purpose limits, minimisation, security and deletion must apply regardless of a user’s ability to pay.
“Optional” can become compulsory
If a school, insurer or employer rewards use—or punishes refusal—a nominally voluntary choice can become coercive. Non-users need equivalent routes and protection from discrimination.
| Decision | Fair-process requirement | Red flag |
|---|---|---|
| Who receives access first? | Prioritise evidence-based medical and assistive benefit, user goals and underserved need through published criteria and an accessible appeals process. | Prestige, publicity or ability to provide commercially valuable data quietly determines access. |
| What does coverage include? | The full pathway: assessment, hardware, clinical care, training, adaptations, maintenance, updates, cybersecurity and end-of-use support. | A funded implant or device with no provision for the services that make it usable. |
| How is consent protected? | Prior, informed and ongoing consent; communication access; independent support; easy withdrawal; no penalty for refusal. | Employment, education, insurance or essential care depends on accepting monitoring or intervention. |
| Who shapes the system? | Disabled people, patients, workers, children where relevant and affected communities hold paid, decision-making roles in design and oversight. | Developers define “benefit” and “normality” without intended users. |
| What happens if a trial or firm ends? | Prospective plans for continued care, data access, repair, explantation, migration and safe shutdown. | A user’s functioning depends on an unsupported server, subscription or insolvent supplier. |
Because access debates in this series also extend from digital systems to biology, the same boundary matters in genome editing. FDA’s 2023 approval of Casgevy demonstrated an approved somatic CRISPR-based treatment for sickle cell disease; it did not establish a route to cognitive enhancement.28 WHO’s governance framework addresses somatic, germline and heritable editing and tests its recommendations against hypothetical scenarios, including athletic-performance enhancement.29 Scenario analysis is oversight, not endorsement or evidence of availability.
08 · From principle to delivery
Build an equity stack, not a single intervention
The strongest programmes combine supply, affordability, user capability, accessible service design and accountability. They budget for recurring costs and measure whether priority tasks remain possible after the launch team leaves. For AI programmes, lifecycle frameworks such as NIST’s AI Risk Management Framework can help teams document context, map affected people and risks, measure performance and harm, and govern deployment—without treating a framework as certification.30
Map exclusion with people, not only infrastructure
Combine coverage and price data with task-based research. Pay affected users to identify language, disability, safety, identity, transport and household barriers that administrative datasets miss.
Define the minimum meaningful service
Specify reliability, latency, data allowance, suitable devices, accessibility, privacy and support for the intended task. “Any connection” is too weak a target.
Bundle infrastructure with affordability
Use transparent subsidies, public access points, shared institutional capacity or demand aggregation. Include devices, electricity, repair and subscriptions in the cost model.
Fund skills and trusted human support
Train users and frontline workers around real tasks, critical evaluation, privacy and troubleshooting. Create in-person, phone and remote support with clear escalation.
Procure accessibility and interoperability
Make standards, user testing, open interfaces, data portability, export, repair, security updates and exit assistance contractual deliverables with enforceable remediation dates.
Preserve equivalent non-digital routes
Keep phone, paper and in-person access where rights, care or livelihood are at stake. An alternative must be workable, timely and stigma-free—not theoretically available.
Measure completion, benefit and harm
Disaggregate by relevant, voluntarily supplied characteristics; protect small groups from re-identification. Track errors, abandonment, appeals, support demand and cost as well as uptake.
Plan for continuity before launch
Fund maintenance, replacement, training turnover, contract exit, data migration and disaster recovery. Publish what happens when a subsidy, vendor or research programme ends.
| Dimension | Useful local indicators | Misleading shortcut |
|---|---|---|
| Availability & quality | Uptime at the place and time of use; measured speed/latency; interruption recovery; power resilience. | Provider-advertised coverage alone. |
| Affordability | Total recurring cost as a share of disposable household resources; arrears; churn; data rationing; repair delay. | National-average tariff divided by national-average income. |
| Devices | Private availability when needed; capability for the task; age, storage and battery; sharing; assistive compatibility. | Counting any mobile phone as a suitable computer. |
| Accessibility | Priority-task completion with representative disabled users; severity and age of defects; equivalent alternatives. | A self-declared compliance badge. |
| Skills & support | Unaided and assisted completion; time to help; repeat contacts; confidence calibrated against demonstrated performance. | Training attendance or self-confidence alone. |
| Outcome & agency | Learning transfer, completed applications, resolved health tasks, user-defined benefit, complaints, appeals and refusal without penalty. | Log-ins, screen time or accounts created. |
A note on zero-rating
“Free” content can narrow both cost and choice
Exempting selected educational sites from data charges may lower an immediate price barrier, but it can privilege chosen platforms, distort competition and create a curated version of the internet. Selective application-specific zero-rating is generally inadmissible under EU open-internet guidance.31 Network-neutral, content-agnostic support—a general data allowance or household subsidy—is usually easier to defend. Elsewhere, regulators should test legality, transparency, switching, privacy and whether equivalent material outside the favoured platform remains affordable.
Before approving a digital or cognitive-technology programme, can you answer yes?
- We know which real-world task the programme is meant to improve.
- We have a baseline and denominator, not only a target number of devices.
- Affected communities have paid roles and decision power.
- The total cost includes data, power, repair, support and replacement.
- Priority journeys work at low bandwidth and on older equipment.
- Disabled users tested the complete service, not only components.
- Language and literacy needs are covered by content and human help.
- Privacy is not a luxury feature reserved for people who can pay.
- Automated decisions can be explained, challenged and corrected.
- An equivalent human or offline route remains genuinely usable.
- Outcome and harm data will be disaggregated without exposing people.
- Continuity and exit plans exist for funding or vendor failure.
09 · Four worked examples
Run the access chain before launch
The same checklist behaves differently in a classroom, telehealth service, assistive-technology programme or neurotechnology trial. These short scenarios show how to move from an attractive product claim to an equity decision.
School AI tutor
Do not count licences; test independent learning
Ask whether students have private time and a suitable device, whether the tutor works in their language and with assistive technology, and whether teachers can see and correct errors. Protect children’s data and provide a non-AI path. Compare what students can explain or solve unaided after use—not how many prompts they sent.
Telehealth portal
Do not turn authentication into triage
Test booking, identity checks, document upload, captioning, interpretation, proxy access and reconnection after a dropped call. Offer assisted phone and in-person routes. Measure completed consultations and clinically appropriate follow-up, not downloads.
Communication aid
Fund the service around the device
Start with the user’s communication goals. Include assessment, vocabulary customisation, mounting, training for communication partners, repair, backup, updates and a way to export settings. A device sitting unused in a cupboard is distribution, not access.
Implanted-BCI study
Plan beyond the impressive demonstration
Recruit fairly, make consent accessible, pay travel and caregiver costs, state what neural data will be retained and decide before enrolment who will support the hardware, software and security, fund explantation if necessary, and preserve the participant’s means of communication if the trial or company ends.
10 · Frequently asked questions
Short answers to difficult questions
The details vary by place and technology, but these distinctions prevent the most common overclaims.
What is the difference between the digital divide and digital inequality?
The digital divide often describes differences in connection, devices or use. Digital inequality follows the distribution of quality, skills, support, autonomy, risk and outcomes among people who may all appear “connected.” The first asks who gets online; the second asks what kind of participation and power they have once there. Both matter.
Is a smartphone enough to count someone as connected?
It depends on the task. A smartphone can be an excellent primary device for messaging, payments, navigation and media. It may be inadequate for long documents, coding, complex forms, simultaneous reference materials or specialist software—especially when storage is full, data are rationed or the phone is shared. Measure task capability and private availability, not device category alone.
Does giving students internet access improve learning?
Connectivity is often necessary for digital learning, but it is not sufficient and its isolated effect is difficult to estimate. Content quality, teaching, device suitability, time, household support and how a tool is used all matter. Some interventions improve specific outcomes; others improve digital proficiency without improving tested subjects. A credible programme names the learning goal, compares against a strong alternative and tests unaided transfer.
Does WCAG 2.2 make a site legally compliant everywhere?
No. WCAG 2.2 is a W3C Recommendation, not a global law. Jurisdictions incorporate or reference standards differently, sometimes using another WCAG version or a broader standard. Level AA requires satisfying every Level A and AA success criterion, subject to WCAG’s full-page, complete-process and conforming-alternate-version rules; it is not an alt-text-and-contrast checklist. Verify the legal rules for the service and still test real priority journeys with users.
Are free or zero-rated educational sites always good for equity?
They can lower a price barrier, but selecting which services do not count against data can also limit choice, favour powerful providers and fragment the open internet. Content-agnostic data support is generally safer. Any scheme needs regulatory review, transparency, privacy protection, switching and an assessment of what excluded content will cost.
How can schools stop AI tools from widening inequality?
Provide suitable devices and connectivity without assuming home resources; procure accessible, privacy-protective systems; give teachers time and training; support local languages; make strong non-AI options available; and test learning without AI. Watch whether premium accounts, private tutoring or faster hardware create an invisible second tier. Include students, families and teachers in governance.
Are brain–computer interfaces already cognitive enhancements?
Current implanted-BCI work is primarily research and medical technology intended to restore or support specific functions, such as communication or device control for people with severe motor impairment. That is fundamentally different from raising general intelligence in healthy users.
Does equity mean everyone must receive every new technology?
No. Equity means fair opportunity, reasonable accommodation, non-discrimination and accountable allocation. Experimental and scarce interventions require evidence, safety oversight and transparent priorities. Medical and assistive needs may deserve priority over elective uses. People also need the right to refuse without losing work, education, insurance or essential services.
What single metric best measures digital inclusion?
There is no adequate single metric. At minimum, combine availability, measured quality, total affordability, suitable-device access, accessibility, skills/support, safety and successful completion of a valued task. Publish definitions and denominators, separate people from subscriptions or households, and disaggregate carefully enough to reveal exclusion without creating privacy risk.
The social contract
Expand capability without making technology compulsory
The most equitable system is not the one with the most devices. It is the one in which people can turn reliable access into learning, communication, care and choice—and can recover when a connection, vendor or subsidy fails.
The current divide is measurable: billions remain offline; affordability, rurality, disability, gender and skills shape meaningful use; and nearly one billion people lack access to assistive products they need. A future divide in elective neural enhancement is plausible but not yet quantifiable. Policy should address the first urgently and the second honestly.
Build the floor
Universal, meaningful foundations
Reliable infrastructure, affordable service, suitable devices, accessible design, digital skills, human support and relevant content belong together.
Distribute power
Users help define benefit and risk
Communities need authority in design, procurement, evaluation and redress—especially disabled people and groups carrying the greatest cost of failure.
Protect the exit
Access includes the freedom to refuse
Equivalent routes, portability, continuity and limits on surveillance keep assistance from becoming dependence or coercion.
A fair cognitive-technology future does not merely place powerful tools within reach. It gives people the resources, rights and understanding to decide what those tools are for.
Evidence and official guidance
Sources
- International Telecommunication Union, Universal and Meaningful Connectivity framework (opens in a new tab).
- International Telecommunication Union, Facts and Figures 2025 (opens in a new tab), including internet use, mobile coverage, ownership and methodology.
- International Telecommunication Union, Affordability of ICT services, 2025 (opens in a new tab).
- International Telecommunication Union, ICT skills, 2025 (opens in a new tab).
- GSMA, Mobile Gender Gap Report 2026 (opens in a new tab).
- Eurostat, Digital skills statistics for the European Union (opens in a new tab).
- OECD, Closing Broadband Connectivity Divides for All (opens in a new tab), 2025.
- GSMA, State of Mobile Internet Connectivity 2025 (opens in a new tab).
- UNESCO, Global Education Monitoring Report 2023: Technology in Education (opens in a new tab).
- UNESCO, What you need to know about AI and the right to education (opens in a new tab).
- World Bank, Digital Progress and Trends Report 2025: Strengthening AI Foundations (opens in a new tab).
- UNICEF, Remote Learning Reachability (opens in a new tab), 2020 assessment and methodology.
- OECD, Do Adults Have the Skills They Need to Thrive in a Changing World? (opens in a new tab), Survey of Adult Skills 2023.
- World Bank and International Labour Organization, Generative AI and Jobs in Latin America and the Caribbean (opens in a new tab), 2024.
- WHO and UNICEF, Global Report on Assistive Technology (opens in a new tab), 2022.
- United Nations, Convention on the Rights of Persons with Disabilities (opens in a new tab), especially Articles 4 and 9.
- W3C, Web Content Accessibility Guidelines (WCAG) 2.2 (opens in a new tab).
- European Union, Directive (EU) 2019/882 on the accessibility requirements for products and services (opens in a new tab).
- World Bank, Rwanda Digital Acceleration Project appraisal (opens in a new tab), 2021.
- Detroit Community Technology Project, 2024 programme update (opens in a new tab) and Equitable Internet Initiative overview (opens in a new tab).
- Yale Inclusion Economics, Bridging the Digital Gender Divide in Chhattisgarh (opens in a new tab) and evidence brief (opens in a new tab).
- U.S. Federal Communications Commission, Affordable Connectivity Program Consumer Survey release and results (opens in a new tab), 2024.
- Universal Service Administrative Company, Affordable Connectivity Program status and benefit information (opens in a new tab).
- UNICEF and ITU, Giga: Connect every school (opens in a new tab).
- U.S. Food and Drug Administration, Guidance for implanted brain–computer interface devices for patients with paralysis or amputation (opens in a new tab).
- Card et al., “An Accurate and Rapidly Calibrating Speech Neuroprosthesis” (opens in a new tab), New England Journal of Medicine, 2024.
- UNESCO, Recommendation on the Ethics of Neurotechnology (opens in a new tab), adopted 2025.
- U.S. Food and Drug Administration, First approval of CRISPR/Cas9-based cell therapy, Casgevy (opens in a new tab), 2023.
- World Health Organization, Human Genome Editing: A Framework for Governance (opens in a new tab), 2021.
- NIST, AI Risk Management Framework (opens in a new tab) and Generative AI Profile (opens in a new tab).
- Body of European Regulators for Electronic Communications, Guidelines on the Implementation of the Open Internet Regulation (opens in a new tab), 2022 update.
Educational and editorial note: This article provides general information, not legal, medical, educational, procurement or financial advice. Laws, programmes, product capabilities and regulatory status change. Check the current rules and evidence for the relevant technology, population and jurisdiction. Global estimates describe populations; they do not determine an individual person’s needs or ability.
Continue the ethics and society series
Cognitive enhancement in context
Evidence reviewed through 4 September 2026. Global totals, programme reach and regulatory status are dated and scoped; estimates and case studies should not be read as universal causal effects.