The central tension of this section is that computing's benefits and its costs often land on different people. It takes up that tension in four steps: who can get access at all, whose languages and knowledge are represented once they do, who bears the material costs of the whole system, and what individuals give up, sometimes without much choice, in exchange for the services they use. It ends by showing how a single technology can pull in both directions, which is the reason equity has to be evaluated and can't be assumed.

Who has access#

Access is easy to treat as one question, whether someone owns a device, but it works more like a stack of layers. The digital divide is the gap between those who have and those who lack devices, reliable connectivity, and the skills and support to use them, and it often follows income, geography, race, age, and disability. Thinking in layers explains why single-layer remedies disappoint. Handing out devices doesn't address skills. Availability doesn't equal adoption either, since price matters as well: service that reaches a street may still cost too much relative to a household's income. It also explains why moving services online without alternatives can exclude people, including those who are missing only one layer.

For teachers, these gaps surface most visibly in assignments that students take home. The homework gap describes students who can't complete Internet-dependent assignments at home, sometimes because a household shares one device or has limited data. A laptop doesn't provide connectivity, so a one-to-one device program can leave this gap open. Schools can respond with class time, loaned hotspots or devices, and offline alternatives, and a teacher who sees missing work should consider limited access before assuming low motivation. It helps to plan assignments that don't assume home access in the first place. Evaluate programs by changes in students' work, such as completion rates before and after borrowing a hotspot, and not by the number of devices bought.

Before a district or a state can respond at larger scale, someone has to measure the problem, and measurement can hide it. Maps that count an area as served if any one address can get service overstate coverage, which can hide unserved households from funding programs. Providers also report their own coverage and may have incentives that affect those reports, and advertised availability differs from delivered speed and affordability. Address-level data and speed tests, some of it collected from residents, give a truer picture.

Who is represented#

Getting online doesn't guarantee finding much that's useful in your own language. Online content and language tools are concentrated in a few languages, so speakers of others find fewer resources and less accurate tools. The two problems are connected, because models learn from available text, so a language with little online content also gets weaker tools, and the knowledge of its speakers is less represented in both. The practical response for a teacher is to test tools in the languages your students actually use, and comparing search results and translation quality across languages can make a good classroom activity.

Who bears the material costs#

Representation is one uneven distribution, and physical harm is another. "Paperless" services still rely on physical infrastructure, and each stage of that infrastructure has costs: mineral extraction, manufacturing, data center energy and water use, and electronic waste (e-waste), which is discarded devices that often contain hazardous materials. These costs are unevenly distributed, and harms often fall on communities that benefit least. People living near mines and informal e-waste processing sites may face health harms without the protections available where the devices were sold, and a town hosting a data center may bear local water and energy costs for services used far beyond it.

What individuals trade away#

The costs so far fall on people mostly without their involvement. A different kind of cost is the one people appear to accept themselves. Personalized services offer convenience and lower prices in exchange for data, at the cost of surveillance, profiling, and loss of control, costs that are often hard to see at the moment of agreeing. The exchange is also less voluntary than it looks. When refusal is costly, as with insurance discounts for those who share data from a tracker, the choice may not be fully free, especially for someone who can't afford the penalty. It helps to separate two things that get bundled together: data needed for a feature differs from data collected beyond that need. And the common reply of "nothing to hide" overlooks that data can be combined and reused in ways nobody can foresee, so a person can't judge the risk by looking only at today's use.

Schools make a version of this trade on students' behalf. Monitoring software on school devices supports safety and policy enforcement, but it reduces privacy, especially outside school hours and for students who rely on school devices, since those with personal devices can step outside it and others can't. A tool meant to protect students can also harm them: it can expose sensitive information and create a chilling effect, in which people avoid legitimate activity because they feel watched. Narrow scope, such as limiting monitoring to school hours, and clear notice to families improve the balance.

Why one technology can go both ways#

Because of all these layers and trade-offs, it's rarely accurate to say that a technology is simply good or bad for equity. One technology can reduce and reinforce inequality at once. Online courses widen access yet favor those with connectivity and adult support at home, and telehealth helps rural patients with broadband and leaves out others. Equal availability isn't equal benefit, so the question to ask is who actually benefits. Evaluate equity by comparing participation and outcomes across groups, and direct added supports to those who lack them. That habit of comparing across groups is just as necessary when a computer system is making decisions about people, as 0001.d shows.

• digital divide: the gap between those who have and those who lack devices, reliable connectivity, and the skills to use them
• homework gap: the situation of students who can't complete Internet-dependent assignments at home
• electronic waste (e-waste): discarded devices, which often contain hazardous materials
• profiling: building a picture of a person from collected data to predict or influence their behavior
• chilling effect: avoiding legitimate activity because of a feeling of being watched

Practice: practice questions for 0001.c