Ethical, Legal, Cultural and Environmental Impacts of Computing

A computing innovation depends on a computer program to do its main job. It might be a device, an app, or a way of offering a service, such as online shopping. Its impacts are the changes it causes, including changes outside the group using it.

Look through five lenses

The same technology can raise several different questions. These lenses organize them without forcing every effect into a single box.

Privacy can also be an ethical or legal issue. The lenses help you ask different questions about an effect; they are not separate worlds.

Follow the change to the person

A delivery-route app can reduce fuel use and get drivers home sooner. The same location records let a manager check every stop. Drivers may value the shorter route and dislike the close monitoring. A specific answer explains both changes and identifies whose life changes.

Compare "The app is bad" with "The manager can use the route record to question a driver about every pause." The second answer gives a mechanism: recorded locations create a way to monitor work. You can then judge whether that use is reasonable.

Purpose is only the start

An intended effect is part of the reason for building a product. An unintended consequence was not planned. In 2018, a fitness app's public activity map exposed outlines of military sites in remote places. Showing popular routes also revealed patterns that were sensitive when combined on a map.

Unplanned effects can help people too. Satellite navigation, developed for military use, now supports farming, emergency services, and everyday travel. New computing tools also enable progress in medicine, engineering, communication, and art.

Speed and scale matter. A sharing setting that exposes one person's route is a problem. The same default on millions of accounts spreads that risk widely. Developers should test with varied users, consider misuse, and monitor what happens after launch so they can respond to effects they did not predict.

Ask a question that fits the technology

Cloud storage raises questions about who can read files and where they are held. A medical implant raises questions about attacks and continued software support. An autonomous car raises questions about responsibility after a crash. Mobile networks make communication easier but can extend work into personal time. A useful analysis connects the technology's actual features to a change in people's lives.

A stakeholder is a person, group, or organization that affects a technology or is affected by it. Ownership and use are not required. A neighbor recorded by a smart doorbell has an interest in its privacy settings even without an account.

Find the people beyond the screen

Start with users, customers, and employees. Then look behind the product at developers, owners, suppliers, and investors. Look around it at families, local communities, schools, and competitors. Include governments and regulators. Follow the physical supply chain to miners and waste workers, and consider people who will live with its environmental effects in the future.

These groups have different interests. Users may want less data collected; a business may want a detailed profile to sell advertising. A supplier may gain work while local residents face extra waste. Naming groups precisely helps explain these conflicts.

Work through a case

Imagine a school considering face scanning for lunch payments. Its aim is a quicker line without payment cards.

1. Name the stakeholders. Students, families, cafeteria staff, the vendor, and the data regulator all have interests.
2. Connect benefits and harms. Students lose fewer cards and staff process payments faster. Students also face lasting harm if stored face data leaks. A password can be changed; a face cannot be replaced in the same way.
3. Apply the five lenses. Consider convenience, fairness, privacy, the legal duties around biometric data, and equipment costs and waste.
4. Check the alternatives. A card or PIN can make payments without storing face data.
5. Reach a judgment. The school should avoid compulsory face scanning because a less intrusive method meets the same need.

The conclusion follows from comparing options. Merely listing faster payments and privacy risk leaves the decision unfinished.

Weigh effects, not votes

A large number of small benefits may sit beside a severe harm to a small group. Examine seriousness, likelihood, whether harm can be reversed, and whether affected people have a real choice. Vulnerability matters: children may have little power to refuse a school system.

Suppose a bank closes a branch and offers an app instead. The bank saves money, and some customers gain access at any hour. Customers with poor connections or inaccessible devices may lose essential banking access. Branch workers may lose jobs. Keeping a shared service point and phone support could preserve some savings while reducing exclusion.

Build a strong discussion

A discussion needs specific points on both sides, linked to stakeholders, followed by a supported judgment. "It affects people" leaves both the group and the effect unclear. "Customers without reliable internet cannot use the replacement service" explains the barrier. Recommend a change that addresses that barrier, and explain what it costs or gives up. A fair answer can acknowledge a remaining disadvantage without pretending every stakeholder will benefit.

Law is a set of rules that a state enforces with penalties. Ethics is reasoned thinking about right and wrong. Asking "Is it allowed?" checks legality. Asking "Should we do it?" also requires ethical reasoning.

Give reasons for a judgment

Legal permission alone does not settle fairness. A company may hide an important data-sharing choice in long terms that users struggle to understand. Even where this meets a legal requirement, it can still exploit users' lack of understanding. New harms can also appear before lawmakers respond.

Three questions help organize a dilemma:

• Consequences: What benefits and harms follow from each choice?
• Duties and rights: What do we owe people, such as honesty and privacy?
• Fairness: Who receives the benefits and who carries the burden?

Suppose an app keeps recording location after its tracking switch is turned off. Fixing this reduces advertising income, but continued tracking exposes users to harm. The switch also makes a promise that the company must honor. Users carry the risk while the company gains income. Consequences, duties, and fairness each support fixing the fault and telling users.

Codes guide people with power

A code of ethics states values and standards for a profession or organization. The ACM's computing code emphasizes human well-being, avoiding harm, honesty, fairness, respect for others' creative work, privacy, and confidentiality. Confidentiality means protecting information entrusted to you.

Programmers can make choices that users cannot easily inspect. They could hide spyware, copy code without permission, or release a known unsafe system. A professional code helps them recognize these responsibilities and challenge an unethical request.

A business code can address customer honesty, data handling, company equipment, conflicts of interest, online conduct, and reporting concerns. A personal code gives an individual consistent commitments, such as asking before posting another person's photo.

Policies set everyday limits

An acceptable-use policy, or AUP, sets an organization's rules for devices, accounts, and networks. It commonly explains allowed uses, banned content, password rules, monitoring, and consequences. A school might prohibit installing games on shared computers even when the games are lawful.

The same act can breach more than one kind of rule. Misusing customer records can break a workplace code, an AUP, and data protection law. Employees risk warnings, dismissal, legal action, and reputational harm. Businesses can lose trust, customers, and income as well as face fines or compensation. "Everyone does it" supplies no reason that addresses these consequences or the rights of the people affected.

Privacy concerns who knows things about a person and how that information is used. Security protects data from unauthorized access or damage. A secure database can still invade privacy if its owner collects too much or uses records unfairly.

Recognize personal data

Personal data identifies a living person directly or makes them identifiable when combined with other information. It includes names, photos, email addresses, locations, and device identifiers. An IP address can also be personal data when linked to someone. A username that hides a real name may still connect to a recognizable pattern of activity.

Organizations obtain data in three ways:

An inference is a prediction, so it can be wrong. Treating it as a fact can lead to unfair advertising or decisions.

Collection can happen quietly

A cookie is a small browser file that can support logins or tracking. A tracking pixel can report that a page or email was opened. Apps use permissions and sensors; loyalty cards record purchases; smart speakers and wearables collect other activity. Metadata describes an activity or file, such as the time and participants in a call. Knowing who contacted a clinic and when can be sensitive even without hearing the conversation.

Organizations use these records to provide and secure services, improve products, support research, target advertising, or sell information. A data broker gathers and sells information about people. In an advertising-funded service, users' attention and data help support the business.

Combine the pieces

A weather app records one phone at the same address most nights, at a school on weekday mornings, and at a clinic on Thursdays. Each location seems like a small detail. Together they suggest home, routine, and possible health information. The person need not type any of those conclusions into a form.

Sharing data can support navigation, useful recommendations, and medical research. It can also enable manipulation, unfair treatment, identity theft, and exposure through a breach. Copies passed to other companies can outlast deletion of the original account.

The Cambridge Analytica case became public in 2018: information collected through a Facebook quiz also involved friends who never used it, with up to 87 million people potentially affected. In 2019, the US FTC imposed a 5 billion US dollar Facebook settlement over privacy failures. The case shows why collecting for one purpose and reusing for another matters.

A click can hide a poor choice

Dark patterns trick or pressure users or obstruct an informed choice, such as making acceptance easy while hiding rejection. Long terms and dependence on a service also weaken real choice. Private browsing mainly limits what the local browser saves; websites and networks can still observe activity. Privacy requires examining collection and use, not just the browser's history list.

Surveillance is organized, systematic monitoring of people's movements, behavior, or communications. Cameras, license-plate readers, workplace GPS, school network logs, and online tracking can all form part of a surveillance system.

Identify the purpose and the cost

Monitoring can help find missing people, investigate crime, protect lone workers, or detect network attacks. Employers may need to know where work vehicles are. Schools may need records of harmful network activity. These are real benefits, but the purpose does not justify every possible form of monitoring.

Monitoring creates records that staff might misuse or attackers might steal. Function creep occurs when records collected for one job become used for other jobs. A chilling effect occurs when people avoid lawful speech or meetings because they expect to be watched. Errors can also expose innocent people to repeated checks.

Privacy protects ordinary activities such as seeking medical help or talking with family. "Nothing to hide" ignores these needs as well as the possibility of mistakes or future misuse.

Apply a practical test

Ask these questions about a proposal:

1. Is the purpose clear and lawful?
2. Is monitoring necessary, or could a less intrusive method meet the need?
3. Is its extent proportionate to the problem?
4. Are people told what happens?
5. Are collection and retention limited?
6. Is there independent oversight?

Work-hour GPS in a delivery van can support routing and driver safety. Tracking the same driver's private trips on a day off supplies no work benefit that requires those records. Limiting collection to work hours changes the balance. Legal controls, including when warrants are needed, depend on the country.

Check what an alert means

A face-matching system scans 10,000 people, including one wanted person. It detects that person and falsely flags 1% of the other 9,999 people. That gives about 100 false alarms alongside one true alert.

Only about one of the 101 alerts is a true match. A low error rate across all scans can still produce mostly false alerts when the sought event is rare. An alert therefore needs careful checking before action is taken against a person.

Follow access to private communications

End-to-end encryption keeps message content readable at the communicating endpoints rather than by the service carrying it. Investigators argue that this can hide serious crime. Security specialists warn that a special access route could also be exploited by attackers and weaken users' security.

Family location sharing can help in emergencies, but demanded tracking can become control. Stalkerware secretly monitors someone. Whether a system is described as protective says little about the tracked person's freedom or the collector's actual power. Examine who can refuse, who can see records, and what limits exist.

A platform feed is a selection, not a complete view of everything available. A recommender system predicts what someone is likely to watch, click, or share, then ranks content using those predictions.

Break the process into steps

A simplified feed collects activity, builds a profile, scores available items, shows high-scoring items, and measures the response. That response becomes data for later recommendations. Watch time, likes, location, and similar users' behavior can all contribute.

Computational thinking helps explain this process. Decomposition separates the stages. Pattern recognition finds shared behavior. Abstraction reduces a complex person to selected features such as interests and age group. The algorithm applies rules or a learned model to those features.

This reduction loses information. A long watch might mean interest, confusion, or leaving a video running. A system treats signals as evidence for a prediction; it does not understand every reason behind them.

Calculate a ranking

Consider an illustrative score: twice the expected minutes watched, plus a sharing score from 0 to 10.

• Video A predicts 3 minutes watched and a sharing score of 2.
• Video B predicts 1 minute watched and a sharing score of 9.

Their scores are:

Video B ranks first. The rule rewards predicted engagement. It contains no check for truth or usefulness. Designers' choice of goal shapes what reaches the top even when every calculation is correct.

Benefits and influence

Personalization helps people find relevant information, music, and creators. It helps small businesses reach interested customers. Targeted ads use profiles to choose an audience. Personalized offers may use the same information, while dynamic pricing changes prices with demand.

An engagement goal can also reward outrage or false claims that trigger reactions. Autoplay, infinite scroll, notifications, and streaks encourage continued use. The resulting pattern can influence spending, attention, and political opinions.

A filter bubble describes a narrowed view produced by repeated recommendations matching earlier behavior. An echo chamber is a social group that mostly repeats similar views. They can overlap, but people's choice of friends and sources matters too; the strength of these effects varies.

Ask what you cannot see

Users usually cannot inspect the full ranking rules. A "Why am I seeing this?" explanation can help, but outsiders still need enough information to check harm and fairness. Two users may see different feeds or search results. To broaden a view, compare sources and deliberately follow different perspectives. Adjusting recommendations and attention settings can change the signals a platform receives.

An automated decision is made or strongly shaped by a computer with little or no human judgment. Systems can screen applicants, approve loans, price insurance, or flag people for checks. They act quickly and consistently, but consistency also repeats any flaw.

Decisions come from human choices

Some systems learn patterns from earlier cases and give each new case a score. A lender might approve scores of 650 or higher. A person scoring 649 is refused even though the one-point difference may be small. The chosen cutoff turns a continuous score into a life-changing decision.

Computing bias is a systematic unfair result for a group. It can enter when developers choose the goal, gather and label data, select features, test the system, or decide how to use its output. Past records may reflect unfair treatment. A group may be poorly represented. A postal code can act as a proxy, carrying information linked to income or ethnicity even after those columns are removed.

Reported in 2018, an Amazon hiring project learned from resumes that mostly came from men and marked down wording associated with women. The project was dropped. A computer can reproduce unequal history without anyone explicitly writing a rule to disadvantage a group.

Look below an overall average

The 2018 Gender Shades study examined commercial gender-classification systems using face images. Error rates reached 34.7% for darker-skinned women; the highest rate for lighter-skinned men was 0.8%. The ratio of those reported rates is about 43 to 1. These results concern gender classification, not a measured police identity-match rate. They show why testing must examine groups separately.

In an illustrative lending example, group A has 160 approvals from 200 applications. Group B has 45 from 90.

The gap is 30 percentage points. That is a reason to investigate the data and criteria. It does not establish the cause by itself: examine whether relevant differences or unfair design explain the outcomes. Comparing raw approval counts would miss the groups' different sizes.

Improve the system and its use

Use representative, accurate, complete data and protect its integrity. Check whether old records suit the new purpose: reusing posts to train a model can raise both quality and consent concerns. Test outcomes for different groups, examine proxies, involve varied users, and audit results after launch.

People affected need notice, understandable explanations, and routes to human review and appeal. A reviewer must have authority and time to question the output. Automation bias occurs when a person trusts a machine's result too readily. Adding a human who only confirms each result leaves the underlying harm in place. The people and organizations building and using a system remain responsible for its effects.

Data protection law gives people rights over their personal data and duties to organizations using it. It permits useful processing while setting limits on collection, storage, sharing, and deletion. Processing includes all of these activities.

Know the roles

The data subject is the person described. The controller decides why and how data is processed. A processor handles data on the controller's behalf, such as a cloud service storing a school's records. The regulator enforces data protection law.

The controller's decisions matter even if a different company runs the servers. Paying a processor does not remove the controller's responsibility for the purpose it chooses.

Apply seven principles

These are widely used principles, expressed especially clearly in the EU's GDPR. Countries organize their laws differently. The United States combines federal laws for particular areas, such as health and children's data, with state privacy laws. A foreign company's activities can still fall within the law of the people it serves.

Consent is one lawful basis

Processing needs a lawful reason. Alongside consent, common bases include a contract, a legal duty, protection of life, a public task, and legitimate interests balanced against people's rights. An online seller needs an address to fulfill a delivery contract. Keeping tax records can follow a legal duty.

Valid consent is freely given, specific, informed, and expressed by a clear action. A preselected box supplies no such action. Withdrawal must be as easy as agreement. Dependence on a service can undermine a person's freedom to choose.

Health, ethnicity, religion, political opinions, sexual orientation, and biometric identification can need extra protection because misuse can cause serious harm. Children's data also needs special care. Some online-consent rules use age thresholds between 13 and 16; the precise rule depends on the legal system and service.

Work through a club form

A chess club needs names and emails to organize meetings. Its form also asks for home addresses, religion, birth dates, and family income. Those extra fields have no role in that purpose, so they fail data minimization. Selling the emails to a chess retailer raises purpose limitation. Keeping former members' records for ten years without a need raises storage limitation.

A better form collects the needed fields, explains their use, limits access, and deletes records when the purpose ends. Small clubs can have data duties too, and publishing personal information online does not remove its protection.

People commonly have rights to information, access, correction, deletion, limits on use, portability, and challenges to important purely automated decisions. For example, a school may need to retain records that a law requires it to keep.

Make a right usable

To request access or correction, contact the organization, identify the record or use concerned, and verify your identity. Requests are usually free. EU and UK rules normally allow about one month for a reply. A refusal can be challenged through the regulator. A game player might request purchase and chat records, then ask for an incorrect birth date to be corrected.

Build protection into the service

Technical measures include access controls, encryption, backups, and updates. Organizational measures include staff training, clear handling rules, and secure rooms. Privacy by design means considering protection from the start. Privacy by default means protective settings are already active when someone first uses the service.

Passwords illustrate why design matters. Instead of saving the plain text "sunflower7", a system stores a password hash made with a suitable one-way function and a random salt. At login, it applies the same process to the typed password and compares results. Hashes are harder for a thief to exploit, but weak passwords still need protection against guessing.

Remove links, not just names

Pseudonymized data replaces names with codes while a person can still be identified, perhaps through a separate key. "Student 47" remains personal data if the school can link the number to a name. Anonymized data cannot reasonably be linked back to a person.

Removing a name alone may leave unique clues. An age, village, and rare hobby can identify someone. Grouping ages, removing rare details, and reporting totals for sufficiently large groups reduce this risk. The practical question is whether someone can still be singled out or linked using available information.

Respond to a breach promptly

A personal data breach can involve loss, destruction, alteration, or unauthorized disclosure. It can start with a hacked server, lost paper file, or misdirected email. Respond by containing it, assessing risk, recording events, notifying the regulator and affected people when required, and fixing the cause. Required reporting deadlines can be short: EU GDPR uses 72 hours after awareness for reportable regulator notifications. It also requires notifying affected people when the risk to them is high.

A teacher sends students' health notes to a class mailing list. The school needs to limit further sharing, assess the exposure, document it, and make required notifications. Deleting the teacher's sent copy does not retrieve recipients' copies.

Regulators can warn, order changes, and fine. Under the EU's higher maximum, the limit is 20 million euros or 4% of worldwide annual turnover, whichever is higher. For turnover of 2 billion euros, 4% is 80 million euros. For turnover of 100 million euros, 4% is 4 million euros, so the higher limit is 20 million euros. Trust and compensation costs can add to the loss.

Support staff should access only needed records, lock screens, keep information confidential, and report mistakes. Before equipment is transferred or discarded, its storage needs secure erasure.

Computer misuse law protects the confidentiality, integrity, and availability of data and systems. Confidentiality limits access; integrity protects against improper changes; availability keeps systems usable. Attacks can threaten one or several of these aims.

Recognize the main acts

Common categories include unauthorized access, access intended to support a further crime such as fraud, unauthorized modification or impairment, and supplying tools intended for computer crimes. Laws also commonly address interception of private communications. The Budapest Convention supports international cooperation around shared cybercrime concerns.

Reading another person's private messages without permission threatens confidentiality. Changing their records threatens integrity. Flooding a service so others cannot use it threatens availability. The harm can exist even if the original files are never copied.

Permission has boundaries

Authorization means permission from someone entitled to grant it. A weak password, unlocked computer, or exposed web page supplies an opportunity, not permission. An employee may have access to one system area while being excluded from another.

Consider these cases:

• A person guesses a friend's password and reads messages: unauthorized access.
• A person breaks into an account to carry out payment fraud: access with intent to commit a further crime.
• A person deletes records or launches a denial-of-service attack: unauthorized change or impairment.
• A person sells a kit intended for stealing credentials: supply of tools for misuse.

Unauthorized access can be an offense even when nothing is changed. "I only looked" does not supply missing permission.

Compare testing with intrusion

Good intent alone does not make testing lawful. Written scope matters: permission to test one server does not extend to every server owned by the organization. A bug-bounty program also has limits.

If you unexpectedly encounter a flaw, stop, avoid further access or testing, and report it to the owner. Trying extra records to collect proof can extend unauthorized access and expose other people's data.

Other online harms have laws too

Computer misuse law is not the only legal concern online. Fraud law can cover scams and phishing. Threats, stalking, harassment, and intimate-image abuse can be crimes without a break-in. Online safety and spam rules can place duties on platforms and senders.

Penalties depend on jurisdiction, intent, and harm. Attacks that endanger life or critical services can be especially serious. Schools and employers may also impose discipline under an AUP. Being young, using a VPN, or finding credentials left in view does not create authorization.

Intellectual property, or IP, consists of legal rights over creative work, inventions, brands, and valuable confidential knowledge. A single digital product can involve several kinds of IP.

Match the right to the creation

Patents usually last about 20 years. Copyright can last much longer. It protects how an idea is expressed, not the underlying idea or a fact. Two developers can both create coin-collecting games; copying the first game's artwork or code raises copyright issues.

Protection starts without a symbol

Copyright generally arises when original work is recorded, including in a saved digital file. The Berne Convention establishes protection without formalities across its members. A missing © symbol or public internet access does not make a work free to copy.

Owners control activities such as copying, distributing, adapting, and public performance or display, subject to legal exceptions. The creator is usually the first owner, but employment rules and contracts can give rights to an organization. Buying a copy of an app or poster gives rights to use that copy; it does not transfer the underlying copyright.

When copyright expires, a work enters the public domain. For individual authors, many systems use life plus 70 years; the Berne minimum is generally life plus 50. A Beethoven composition may be public domain while a recent recording of it remains protected. The composition and performance recording are different works.

Credit and permission solve different problems

Plagiarism presents another person's work as your own. Copyright infringement uses protected rights without permission or an applicable exception. Piracy commonly means unauthorized copying or sharing of software or other protected media.

A public-domain poem submitted under your name is plagiarism even though copyright has expired. Uploading a protected film with full credit can still infringe copyright. Naming the director does not grant permission to distribute the film. No general "change 10%" rule makes copied work yours.

Digital copying is cheap and exact, which makes large-scale sharing easy. Owners may seek removal and damages, and some piracy can lead to criminal penalties. Cracked software can also expose users to malware.

AI does not remove ownership questions

Follow school rules and acknowledge AI use in assessed work. Presenting AI output as your own can breach academic rules. In the United States, copyright requires human authorship: purely AI-generated material lacks that protection, while human creative contributions can qualify. Whether particular uses of protected works for training are lawful remains a matter of legal decisions and debate.

Finding an image, song, or font online is only the start of a reuse decision. First identify the work, its owner, and the activity you plan: copying, editing, publishing, or commercial use may have different conditions.

Follow a permission route

Ask whether the work is in the public domain. If it is protected, check for a license that permits your planned use. Next consider whether a legal exception applies. If neither route permits the use, ask the rights holder for permission. Otherwise, create your own content or choose something else. Keep the source and license information with the project.

US fair use balances four factors: purpose and character of use, nature of the work, amount and importance taken, and effect on the original's market. Criticism using a short, relevant clip differs from a commercial advertisement using a complete song. There is no safe fixed number of seconds or percentage.

Countries using fair dealing, including the UK, Canada, and Australia, define particular permitted purposes. The details differ. "For school" alone does not settle whether a use meets an exception.

Read Creative Commons conditions

Creative Commons licenses give public permission in advance while the owner keeps copyright.

The six combinations are BY, BY-SA, BY-NC, BY-NC-SA, BY-ND, and BY-NC-ND. CC0 is a separate tool for giving up rights as far as possible.

A business blog can use a CC BY photo with proper credit. A CC BY-NC photo does not fit the blog's promotional use. A CC BY-ND photo can be shared unchanged under its terms, but a creatively recolored adaptation cannot be shared using that permission. CC BY-SA allows editing, with the adaptation shared under the required license. Copyright remains with its owner throughout these permissions.

Give credit that can be followed

For CC work, remember TASL: title, author, source, license. For an invented example: "'Harbor at Dawn' by Ana Ruiz, Open Photo Library, CC BY 4.0; cropped." Link to the original and license when possible, and indicate changes. In school work, cite paraphrases as well as quotations using the required style. Put copied words in quotation marks.

Open access research is free to read, but reuse rights still depend on its terms. A stock image, music track, or font also comes with a license. "Royalty-free" means the license avoids a fee for each covered use; it can still cost money and impose limits. A print-font license may not cover a website.

Copyright permission for a photo does not settle the privacy or portrait rights of the person shown. Seek consent before publishing recognizable people, especially children, and respect community protocols for cultural knowledge, stories, songs, images, and voices. For your own creations, keep original files and state your chosen license clearly.

A software license says what a copyright owner permits users to do. It can limit installations, copying, modification, and distribution. An end-user license agreement, or EULA, commonly presents these terms during installation.

Compare rights before price

Source code is the human-readable program a developer writes. A compiled program is prepared for execution and is much harder for people to inspect. The key distinction between open source and proprietary software is the license's rights, not simply whether a download costs money.

Linux, Firefox, LibreOffice, and Python are open source examples. Microsoft Office and Adobe Photoshop are proprietary examples. An open source company may charge for software or support. A company may publish code for viewing while forbidding modification or redistribution; source visibility alone does not make it open source.

Freeware describes zero price, often with proprietary restrictions. Shareware and trials allow testing before payment. Freemium provides basic features free and charges for extras. Public-domain software has no copyright restrictions. These labels describe different features and should not be treated as synonyms.

Read open source conditions

Permissive licenses such as MIT and Apache allow broad reuse, including within proprietary products, while retaining required notices and other terms. Copyleft licenses such as the GNU GPL require distributed covered adaptations to preserve the license's freedoms and provide corresponding source to recipients.

Changing GPL software privately does not create a duty to publish the change. Distributing a combined program that incorporates GPL code can extend GPL duties to that combined work. Simply using a separate GPL tool is a different situation. The way code is combined and distributed matters.

Make a reasoned recommendation

Consider budget, technical skills, support, compatibility, customization, and distribution plans. Open source can reduce license fees, allow adaptation, and reduce dependence on one vendor. It may need skilled setup or paid support. Public code helps inspection only if someone actually reviews it.

Proprietary products may offer familiar workflows, training, and a support contract. They can also bring recurring fees, limited modification, and dependence on the supplier. Neither model guarantees maintenance: inspect the actual product and support commitment.

A charity with a skilled volunteer may benefit from open source office tools. A hospital needs reliable support and could obtain it through either proprietary software or an open source support contract. A startup planning a closed product may choose a permissive library instead of incorporating GPL-covered code into the product.

Check installations against permission

A school has 25 per-device licenses but runs the program on 30 computers. Five installations lack a license. It can buy five more, remove five installations, or switch products. Keeping a license inventory makes these checks possible. Paying once gives the agreed usage rights, not unlimited copying or ownership of the code.

A device's environmental footprint begins before anyone switches it on. Its life cycle includes raw-material extraction, manufacturing, transport, use, and end of life. A fair comparison looks across these stages.

Materials and manufacturing

Computers and phones use copper, gold, lithium, cobalt, and rare earth elements. Mining consumes energy and can damage land and water. Working conditions matter too. The US Geological Survey estimated that Congo (Kinshasa) supplied 73% of mined cobalt in 2025; dangerous conditions and child labor have been reported at some mines.

Manufacturing chips, screens, and batteries uses resources and releases greenhouse gases. For many phones and laptops, manufacturing causes more emissions than several years of use. Replacing a working laptop to save a little electricity may add a larger manufacturing burden. Servers that run continuously can have a much larger use-phase share.

Calculate operating energy

Power is the rate at which energy is used. Convert watts to kilowatts, then multiply by hours:

For an illustrative 50 W laptop used 6 hours a day for 200 days, the power is 0.05 kW. Its operating time is 1,200 hours.

It uses 60 kWh. A 150 W desktop used for the same time uses 180 kWh, three times as much. These totals describe electricity during use, not mining or manufacturing. Emissions depend on the electricity source too; equal energy use can have different carbon impacts on different grids.

The cloud is physical

Cloud services run on servers in real buildings. Data centers use energy for computation and cooling, and many use water as part of cooling. Networks also need equipment and electricity to carry data.

The IEA's 2025 Energy and AI report estimated data-center electricity use at about 415 TWh in 2024, roughly 1.5% of world electricity use. It projected around 945 TWh by 2030, about 2.3 times the 2024 figure. That 2030 number is a projection, not a measured total. AI is a major source of expected growth; streaming, gaming, and storage also contribute.

One small online action may have a small footprint, while billions of repeated actions create large demand. Calling files "in the cloud" can hide the buildings and equipment behind them.

Compare complete alternatives

Going paperless can save paper, ink, and delivery. A video meeting can avoid travel. Both still require devices, networks, and servers. Compare the activities actually replaced as well as the digital costs added. Printing only needed pages and using both sides can reduce paper use without pretending that every digital substitute has zero impact.

E-waste is discarded equipment with a plug or battery. Global totals include appliances and toys as well as computers and phones. It contains both useful resources and materials that can harm people and ecosystems.

Understand the growth

The Global E-waste Monitor 2024 reported 62 million metric tons of e-waste generated in 2022. Only 22.3% was documented as properly collected and recycled, about 13.8 million metric tons. The report projected 82 million metric tons in 2030, around 32% above 2022. A projection describes an expected future, not a measured outcome.

More ownership, fashion, cheap replacements, costly repairs, and ending software support all contribute. Planned obsolescence means designing products to become unusable or outdated earlier than necessary. Glued batteries and unavailable parts can make a repairable fault the end of a device's useful life.

Follow the harm and the resources

Lead, mercury, cadmium, and some flame retardants can cause harm when waste is dumped or burned. Mercury can damage the nervous system. Informal workers may burn cables or use acids without adequate protection to recover metal. Exporting broken devices can move this burden to people who gained little from using the original products.

The Basel Convention controls cross-border movements of waste, including e-waste, though illegal trade continues. Shipping a useful, working device for reuse is different from sending unusable equipment that becomes waste on arrival.

Old devices also contain copper, gold, and other valuable materials. Recovering them reduces the need for new extraction. Recycling still uses energy and cannot recover everything, so it comes after ways of avoiding waste.

Choose the next action

Use the order reduce, reuse, repair, recycle as a guide. Reduce unnecessary purchases and keep working equipment useful. Reuse a device through resale or donation when it suits another user. Repair faults where practical. Send unusable equipment to a certified electronics collector.

A four-year-old laptop that feels slow may need a replacement battery, more memory, or lighter software. Check these possibilities before buying a whole new machine. If it no longer meets your needs but works, someone else may still use it.

Keep electronics and damaged lithium batteries out of household trash; damaged batteries can start fires. Use the appropriate collection service.

Handle the data as well as the hardware

Before transfer, back up needed files, sign out, and securely erase or reset the device using an appropriate method. Deleting file names can leave recoverable data. Organizations may use verified wiping services or destroy storage that cannot be safely erased.

Manufacturers can offer replaceable parts, long support, and take-back schemes. Extended producer responsibility makes producers contribute to collection and recycling. Governments can set repair rules and recycling targets. Repair cafés, refurbishers, and community collection centers provide practical routes for keeping materials in use.

Reducing computing's footprint involves everyday settings, purchasing, design, and the electricity supply. Computing can also help other activities use fewer resources. The useful question is how much a change saves after its own costs are counted.

Start with many small savings

Lower screen brightness when practical, enable sleep and power-saving settings, and shut down equipment when finished. Standby still uses electricity. A small saving per machine becomes significant across many machines and many hours.

For an illustrative school, 200 computers remain idle at 60 W for 14 hours each night. Sleep uses 2 W. The saving per computer is 58 W, or 0.058 kW.

That is 0.812 kWh per computer each night. For 200 computers it is 162.4 kWh. Across 190 school nights:

The total is 30,856 kWh. With an illustrative factor of 0.4 kg of CO₂ per kWh, the saving is 12,342.4 kg of CO₂, about 12,300 kg. The factor is an input for this example; a different electricity supply changes it.

Change the system around the user

Organizations can choose efficient, repairable hardware with long update support. Virtualization lets several virtual systems share physical servers, improving equipment use. Renewable-powered data centers and scheduling heavy work for cleaner electricity periods can reduce emissions.

Software contributes too. Efficient algorithms reduce computation. Lightweight pages reduce transfers. Programs that still run on older devices delay replacement. These choices can matter more than asking each user to make a minor adjustment repeatedly.

Use computing to reduce other costs

Models help predict floods and climate change. Sensors and satellites track pollution, wildlife, and forest loss. Smart meters reveal energy patterns. Models let researchers test changes that would be slow, costly, or dangerous in the real world.

Smart heating reduces heat in empty rooms. Traffic control can reduce idling. Engine management adjusts fuel and air. Precision farming targets water and fertilizer where needed. Each system still has equipment and operating costs.

Calculate the net effect

A building uses 20,000 kWh for heating each year. A smart system saves 12%, or 2,400 kWh, while its sensors use 50 kWh.

The operating saving is 2,350 kWh per year. A whole-life comparison must also count manufacturing and disposal of the equipment. "Smart" alone says nothing about the total saving.

The rebound effect occurs when greater efficiency makes use cheaper or easier and people use more, reducing the expected saving. Cheap cloud storage can encourage keeping more files. Greenwashing uses misleading environmental claims, such as "eco-friendly" with no useful evidence. Ask for a comparison basis, measured results, and the costs included before accepting the claim.

The digital divide is the gap in access to devices, connections, and the skills needed to benefit from them. It exists between regions and within communities. A person can own a phone and still struggle to use an essential digital service.

Check three layers

Access asks whether there is a suitable device and connection. Quality asks whether these are reliable, affordable, and available when needed. Skills ask whether the person can use them safely and confidently. A household sharing one phone has different opportunities from one with several laptops and unlimited broadband.

Costs of devices and data matter, but so do electricity, cables, mobile coverage, and public access. Language barriers, inaccessible websites, low confidence, and restrictions imposed by governments can also exclude people.

In ITU's 2025 estimates, 6 billion people, about 74% of the world, used the internet; 2.2 billion remained offline. Use was 94% in high-income countries and 23% in low-income countries. The high-income rate was about 4.1 times the low-income rate. Urban use was 85%, compared with 58% in rural areas; use was 77% for men and 71% for women. The figures describe population patterns, not the skills of every person within a group.

Follow the effect into daily life

Low-income households may face unaffordable data. Rural communities may have poor coverage. Older people may lack support, while disabled people may face unsuitable interfaces. Being young guarantees neither access nor skill.

Digital-only schooling, banking, job applications, and government services can widen existing inequality. Missing access also limits a person's voice in public debate. The issue is not just whether people can entertain themselves online; essential opportunities may depend on participation.

Check a real task's demands

An illustrative school assigns eight 300 MB videos each month. Together they require 2,400 MB, or 2.4 GB using decimal data units. A student sharing a phone with a 2 GB monthly allowance cannot complete the downloads within that plan, even before any other use.

A device donation alone does not fix the data limit. The school could offer campus downloads, low-data materials, printed alternatives, and device loans. Each action addresses a different part of the barrier.

Match support to the barrier

Governments can improve broadband and public access. Libraries can offer connections and staff support. Companies can provide affordable plans and accessible services. Charities can refurbish devices and teach skills. Friends and families can help someone build confidence without taking over every task.

Check whether a proposed solution works at the point of use. A public computer is less useful if opening hours clash with work. A training course may fail if its instructions are not in a language people understand. Keeping an offline route protects access while digital support develops. Closing that route can widen the divide even when the online service is excellent for people who can use it.

Accessibility means designing technology that disabled people can perceive, understand, and use. Inclusive design considers a wider range of differences: language, culture, age, income, temporary injury, and the situation in which a product is used.

Understand the tools people use

Screen readers speak digital text aloud. Magnifiers enlarge content. Captions show speech and relevant sound in text. Voice control, switches, and adapted keyboards support people with limited movement. Clear language and consistent layouts help people understand what to do.

An image with no meaningful text alternative may be invisible to a screen reader user. A button that needs a mouse may block someone using a keyboard. These are design barriers, not a lack of effort from the user.

Use four principles

The W3C's Web Content Accessibility Guidelines, or WCAG, organize web access around four principles:

• Perceivable: people can sense the information.
• Operable: people can use the controls.
• Understandable: content and controls make sense.
• Robust: content works with different tools, including assistive technology.

Practical checks include useful alternative text for images, captions, clear headings and labels, resizing text, full keyboard operation, and adequate contrast. Under WCAG level AA, normal text generally needs a contrast ratio of at least 4.5:1. Color alone should not carry essential meaning.

Fix a sign-up form

A club form uses pale gray text, marks mistakes only in red, includes an uncaptioned video, and has a submit button that keyboard users cannot reach.

Start by linking each feature to its barrier. Low contrast makes text hard to read. Red alone may be missed by someone who cannot distinguish it. The video excludes someone who cannot hear its instructions. The unreachable button blocks completion without a mouse.

Use stronger contrast, error messages in words, captions, and controls that work from a keyboard. A text alternative such as "Bar chart showing recycling rising from 20% to 35%" communicates the chart's message; "chart.png" only names a file.

Build access from the start

Test with disabled users and assistive tools while design choices are still easy to change. Automated checks catch some problems, but cannot judge whether every label is meaningful or every task is easy to complete. Human testing adds that evidence.

Accessibility supports equal participation and is required by many laws, especially for public services. It also widens the audience. Captions help in a noisy room; keyboard control can help when a mouse fails. Inclusive design also asks whether a product works on old devices, limited bandwidth, and different languages. Adding these checks late can leave barriers built into the core workflow.

Automation uses machines or software for tasks people once did. It removes some roles, creates others, and changes tasks within many jobs. Digital change therefore affects both employment opportunities and working conditions.

Separate tasks from whole jobs

Routine data entry, checkout, and assembly tasks can be automated. Development, cybersecurity, analysis, and support create other work. AI also changes writing, coding, and design. A job often combines several tasks, so automating one does not necessarily remove the entire role.

People who lose work may not have the skills or location needed for the new jobs. Retraining, ongoing learning, and support during transitions help. The wider effects can reach town centers when shopping and banking move online and local premises close.

Remote work reduces commuting and can increase flexibility. It can also cause isolation or make work spill into home time. Smartphones make people reachable outside work hours. Gig platforms offer flexible delivery or other work, but income and hours can be unstable and software may assign, monitor, and rate tasks.

Identify health risks

Repeated movements can contribute to repetitive strain injury, or RSI: pain affecting muscles, nerves, or tendons, often in hands and arms. Awkward posture can strain the neck and back. Long screen sessions can bring eye strain and headaches. Extended sitting reduces movement, while late-night use can disturb sleep.

These risks affect young people too. A comfortable chair helps, but it cannot remove repeated movements or the effects of staying still for hours.

Fit the workstation to the person

Ergonomics adapts work and equipment to people. Adjust the chair and desk so the back is supported and feet rest on the floor or a footrest. Keep wrists straight, forearms roughly level, and the keyboard and mouse within easy reach.

Place the monitor directly ahead, about 50–100 cm away, with its top at or just below eye level. Adjust the position and text size to avoid leaning or squinting. Reduce glare from lights and windows. Arrange cables to avoid trips and avoid overloaded sockets.

Change the routine too

Short, frequent breaks or changes of task help relieve repeated strain. HSE guidance gives 5–10 minutes each hour as an example preferable to 20 minutes every two hours. During a break, move, change posture, and look away into the distance instead of moving straight to another screen.

A student with wrist pain after gaming can adjust the setup, reduce repeated strain, and take breaks. Persistent pain needs medical advice. Report workplace problems early. Employers commonly assess workstations, provide suitable equipment and training, and organize breaks; workers need to use equipment properly and raise concerns. Healthy work depends on both physical setup and how tasks are arranged.

Digital technology changes how people communicate, learn, shop, follow news, and spend time together. Earlier technologies, including print and telephones, also changed social life and raised hopes and fears. Comparing effects helps identify what is new about today's scale and speed.

Connection and cultural choice

Online groups link distant families and people with shared interests. Streaming and digitized archives widen access to creative work and preserve records. At the same time, large platforms and dominant languages can crowd smaller cultures out of view. Design can reflect its makers' assumptions, such as voice tools that work better for some accents.

Easy publishing supports citizen journalism, activism, and community organization. The same channels can carry bullying, misinformation, and deepfakes, realistic media created or altered with AI. Some governments restrict sites or monitor critics. Effects on expression depend on who controls access and how power is used.

Many people can contribute

Crowdsourcing gathers ideas, effort, or information from a large online group. Crowd-built maps and Wikipedia combine contributions. Citizen science lets volunteers collect research data, while crowdfunding gathers small financial contributions for a project.

These approaches draw on more people's knowledge and can reach places a small team cannot. Quality checking still matters, and unpaid contribution raises questions about who receives the value created.

Understand why switching is hard

A network effect occurs when a service becomes more useful as more people join. A messaging app's value grows when friends use it, encouraging more people to choose the same service. Lock-in means that switching is difficult or costly, perhaps because files, contacts, habits, or services are tied to one provider.

A school may find one cloud provider convenient for email, documents, and classes. If prices or terms change, moving everything can be difficult. Regular data exports, open file formats, and a switching plan reduce dependence. An outage also shows how many activities rely on the same company.

Judge regulation by its effects

Competition, privacy, safety, illegal-content, and transparency rules can protect people and challenge concentrated power. Poorly designed rules can also restrict lawful expression or make useful innovation harder. Citizens influence decisions through elections, consultations, debate, and purchasing choices.

Consider social-media age checks. They may reduce some harms to teenagers but also restrict contact and participation. Platforms bear costs, and users may need to prove age. Keeping copies of identity documents creates a different privacy risk from a method that supplies only an "over the age limit" result without identifying the person.

A reasoned judgment compares those methods and the people affected. Saying either "Regulation is always good" or "Online life is not real life" avoids the actual consequences. Rules, interface choices, and business power shape opportunities and harms that people experience beyond the screen.