IB Digital Society HL — All Flashcards

Information age, computer age, post-industrial society, network society, fourth industrial revolution.

Access, use and skills are unevenly distributed, so people under the same infrastructure live very different digital lives.

That services and knowledge work have replaced manufacturing as the main form of employment.

The ITU estimated in 2023 that around 2.6 billion people had never used the internet.

Access (can you connect at all), use (how, how often, how well), and skills and outcomes (what you can achieve online).

Around 2.6 billion, on the ITU's 2023 estimate — roughly a third of the world.

Impact: some people cannot get online. Implication: services designed for the majority make the excluded minority poorer and less visible.

Rural against urban coverage, income and shared devices, age, disability, language, and in several regions gender.

Integrated circuit (1958), microprocessor (1971), personal computer (1981), the internet (ARPANET 1969), online social networks (2000s), mobile and cloud (2007 onwards).

The internet is the network of networks. The web is ONE service on it — linked documents fetched by a browser. Proposed 1989, public 1991.

A whole processor on one chip, so computing could be embedded in other products instead of occupying a room.

Smaller and cheaper, and more connected. Almost everything else in the course follows from those two forces.

A single binary digit — 0 or 1. The smallest unit of data.

Eight bits, giving 256 possible values.

A circuit can tell on from off reliably even as voltage drifts; ten levels would make errors far likelier.

Exact copies, compression, encryption, and storing text, sound and images in one system.

Analogue is continuous and may take any value in its range; digital is discrete and takes one of a finite set of values.

Measuring a continuous signal at intervals and recording each measurement as a number, so it can be stored digitally.

Sample rate (how often you measure) and bit depth (how finely each measurement is recorded).

It is more robust — copies do not degrade. It is not automatically more faithful: a digital recording is always an approximation.

The conversion of analogue data or information into a digital form.

Media decay, format obsolescence, hardware obsolescence, and link rot.

Formats and media keep dying, so the material must be migrated indefinitely or the copy becomes unreadable.

What gets scanned first is what later gets found, cited and taught; the unscanned quietly becomes invisible.

The use of digital systems to change the structure or operation of an organization.

Digitization converts the MATERIAL and leaves the process alone. Digitalization changes the ORGANIZATION — roles, structure, model.

Cost, speed, reach, and the data every digitalized process produces.

Exclusion of users, fragility when one outage stops everything, deskilling of staff, and surveillance of workers.

The evolution, transformation, adaptation or movement from one form, state or value to another.

History, science and technology studies (STS), and future studies.

Evolution is small steps — a change of degree. Transformation is a change of kind: the thing becomes something else.

A change deep enough that the arrangements built around the old way stop working.

Who gains and loses? Who was asked? When are you asking? Compared with what?

The gains and the costs fall on different people, so both can be accurate and reach opposite conclusions.

Not against the past, but against what could have been built instead.

Benefits are immediate and measurable; costs fall slowly on fewer people and are harder to count.

Technological determinism, social shaping, and co-production.

The view that a technology arrives and society rearranges around it — as though nobody designed, funded or deployed it.

The same technology produces very different outcomes in different societies — Estonia's internet voting against internet shutdowns.

Uncovering the assumptions and interrelationships behind a claim, not only agreeing or disagreeing.

Continuity: a new method for a need that already existed, which old arrangements can adapt to. Discontinuity: a break that needs new rules.

Has a judgement that used to be made by a person moved into a system?

It is the half most answers omit, and it is usually where the surprising point is.

Continuous in purpose; discontinuous in scale and reviewability — and it is the second pair that makes it a new problem.

The act, process or instance of representing an idea, feeling or identity in a form others can receive — words, images, sound, video, design or code.

A platform that decides how far an expression travels, between expressing it and being heard.

Publishing became free; reach did not. Reach is granted by a system.

Online, every audience receives the same message at once, so something written for one group lands badly with another.

Storytelling, persuasion, protest, art, and showing identity.

Only the targeted person sees the message, so there is no shared record for a journalist, rival or regulator to challenge.

Organising unseen — the same tools that make it fast also record who took part.

Access to speak improved dramatically; the conditions for a shared, checkable public argument got worse. Both are real.

It brings people together AND divides them, using the same property: people finding others who share something.

People who are rare where they live — a rare condition, a scattered diaspora, an identity not accepted locally.

Agreement within groups and conflict between groups, because both hold attention.

Platforms unite and divide by the same mechanism, so the balance depends on what the ranking rewards.

What defines a person, group, social entity or organization and distinguishes it from others.

Declared (you wrote it), recorded (observed as you use the service), inferred (estimated from the recorded).

The inferred one. It decides what you are shown, offered and charged — and you cannot see or correct it.

Whoever verifies identity decides who can act. A failed fingerprint check under Aadhaar can block food someone is entitled to.

It changes over time and according to context — a person at 14 and at 24, or at work and among friends.

Online, every audience receives the same message at once, so one presentation has to serve all of them.

It was accurate at the time. The problem is relevance rather than truth — which is the ground the right to be forgotten stands on.

Removes a search result under a person's name. The original page stays published.

A person holds several characteristics at once and they overlap — the combination is its own position, not the sum of its parts.

Age, nationality, religion, culture, gender, sexuality, race, ethnicity, and social and economic class.

Each test averages over everyone with that characteristic, so a small group at the intersection disappears into the average.

The fewest training examples exist for those combinations — and that is knowable before training begins.

The ability to influence or control people, events or resources.

Setting defaults, controlling access, deciding visibility, holding information.

They need no enforcement and give nobody anything to object to, yet almost nobody changes them.

A ranking drop removes a seller's customers without removing the seller — so the platform can truthfully say it banned nobody.

Power produced by how a system is arranged, which continues without anybody deciding to use it.

Network effects, switching costs, infrastructure position.

The position produces the outcome whoever occupies it — a kinder owner of the same position holds the same power.

Portability attacks switching costs; interoperability attacks network effects. Only together do they let you leave without losing people.

Who decides? Who gains? Who bears the cost of an error?

In almost every system the decision and the gain sit in one place and the cost of error sits with somebody else.

For the organisation it is an acceptable percentage; for the person it is their whole outcome.

A real route to human review — which makes it a design question rather than a verdict on automation.

An environment created by digital systems where people meet and act, whose boundary and rules are set by its owner.

Who may enter, what may be said, what is shown first.

Two people in the same space are shown different things, so there is no shared view to point at.

Short-term rentals made a flat worth more to visitors than to residents, so housing left the rental market in Barcelona.

Gathering, working, trading, learning.

Four functions share one online space, so one rule set has to serve all of them and nobody knows which function others are in.

The part nobody had written down, because the room used to do it — noticing a lost student, the unplanned conversation.

Channels, close-friends lists and second accounts are people rebuilding by hand the separations a single space removed.

A device and the skill, a connection, an unblocked route, being shown.

Not having a device or the skill — it affects most people and looks like nobody's decision.

The task crosses the border instead of the worker, so labour rules stay where the worker is.

They stop the reporting of what happens during them, removing the evidence anyone would need to hold the decision to account.

Personal, local, national, global.

Scale — how many. Time — how long. Reach — how far.

Decided at company scale, harm lands at personal scale, and only national scale can act.

Market size, not sovereignty — so formally equal countries have very unequal power over the same platform.

A set of connected parts where what one does changes what the others do, so the whole behaves in ways no part does alone.

The other parts respond, changing their own behaviour until the system returns to where it started.

A change to the connections — what is measured, rewarded or paid for — rather than to quantity.

“What will push back is…” — naming the response is what turns a wish into an answer.

Elements, connections, boundary, purpose, feedback.

Everything outside it is treated as fixed, so it decides which solutions the analysis can even consider.

Read it from what the system reliably produces, not from what its owner says it is for.

Output returning as input, so the system learns from behaviour it caused — which is why systems drift.

Steady (returns to a level), growing (each round bigger), tipping (flips and stays).

No single step is unusual, so nothing triggers attention until the total is already large.

Removing the pressure does not undo the change — the way back is not the way you came.

Steady: change the rule. Growing: act early. Tipping: keep a margin.

A simplified version of something, built to answer one question — useful because of what it leaves out.

What did it leave out, and does that matter for the question being asked of it?

A model simplifies; a simulation runs a model forward in time; a visualization shows data so a person can read it.

They are read in about a second, and axis, scale and colour have already made an argument by then.

Something a person or group holds to be important — privacy, safety, fairness, freedom.

The reasoning about what to do when two values require opposite actions.

Everyone is in favour of all of them, so the list contains none of the disagreement.

Its reasons are stated, so somebody who disagrees can examine and answer them.

Personal values, group norms, written codes, law.

Written by the organisation and enforced by nobody — and afterwards they can be shown as evidence of care.

A legislature cannot write a rule about something nobody has seen yet, so the other three carry the years in between.

A legal duty to achieve an outcome, with the industry deciding the method — roughly the EU AI Act's approach.

Personal preference, community expectation, national rule, global agreement.

At the national rung — nothing above the state carries a penalty of its own.

They bind only the countries already willing, because declining costs nothing.

They aim the same value at different threats — commercial collection in one place, state access in another.

Consequences, duties, rights, character.

More good than harm? Is the act itself wrong? Does it take something that may not be taken? What does it make normal?

They depend on the numbers, so new evidence can reverse the verdict.

Run two on the case, show they disagree, then say which you follow here and why.

Who is in the room, what the default is, what gets measured.

Nothing reports it, so it cannot influence anything the team does next — measurement decides what improves.

Neutral about purposes, not neutral about outcomes: it does not care what you want, but it makes some results far easier.

Privacy by default, accessibility requirements and the right to repair are values deliberately designed in.

Data (raw facts), information (data with context), knowledge (a pattern you can act on), wisdom (knowing whether you should).

Raw facts, figures or symbols with no context to give them meaning.

Data that has been given context, so that it means something.

Wisdom. They can find patterns, but deciding whether acting on a pattern is right is a choice about values.

Quantitative is numbers you can count or measure. Qualitative is words, images and sounds — descriptions.

Data about other data — the time, place, device or author attached to a file rather than its content.

It shows who, where, when and how often, without anyone reading a word — which is enough to map someone's life.

Cultural, financial, geographical, medical, meteorological, transport, scientific and statistical.

Finding trends, patterns, connections and links; and collecting measurable facts about people and communities.

A link means two things move together. A cause means changing one changes the other — which needs a controlled test, not a chart.

The pattern is real, so the system looks right, while the reason behind it may not exist at all.

Amazon’s scrapped CV tool learned from a decade of mostly male applications and downgraded CVs containing “women’s” (reported 2018).

Create/collect/extract, store, process, analyse, access, preserve, reuse.

Reuse — data collected for one purpose used for another the person never agreed to.

Cleaning can drop the awkward records, so the result looks tidier and more certain than the truth.

Because reuse is the stage that changes what the data does to a person, so each purpose needs its own lawful basis (in force 2018).

Primary: collected first-hand for your own question. Secondary: collected by someone else for theirs.

Organises and structures data so it is accessible, manageable and capable of being updated.

Who is NOT in it? Data from app users leaves out everyone without the app.

The categories decide what can be counted. An answer the form does not offer can never appear in the results.

Charts, tables, reports, infographics and visualizations.

Does the axis start at zero? What time range? Are the areas honest? Count or rate? What is missing?

A truncated axis, a chosen start date or an area scaled in two dimensions changes what the reader sees while every figure stays true.

Global Fishing Watch (2016) collected nothing new — publishing existing signals as a map moved power from fleets to the public.

Encryption (lock it with a key), masking (replace it with realistic fakes), erasure (destroy it so it cannot be recovered).

Whether a record has been changed since it was written. NOT whether it was true when written, and not privacy.

Deleting usually removes only the pointer; the contents stay until overwritten. Erasure overwrites the space.

Ordinary controls not kept up — a reused password (Colonial Pipeline, 2021), an unapplied patch (WannaCry, 2017).

Volume, variety, velocity, veracity.

Veracity — whether the data is right. Volume does not fix errors, it hides them.

Predicting behaviour, modelling what would happen under other conditions, and understanding past, present and future behaviour.

A group never collected from is still absent, so the blind spot stays while the model looks more reliable.

Bias, reliability and integrity; control, ownership and access; privacy, anonymity, surveillance and personally identifiable information.

A few details together — postcode, date of birth, sex — identify most individuals, and joining datasets makes the link easy.

The harm and the benefit come from the same cause, so removing one removes the other.

Quality: COMPAS. Ownership: the right to be forgotten (CJEU 2014). Surveillance: San Francisco's facial recognition ban (2019).

A set of defined steps that solves a problem or carries out a task.

Unambiguous, finite, well-defined inputs and outputs, feasible.

Often by efficiency — how much work it does as the size of the job grows.

No. A recipe is an algorithm. A program is one way of expressing an algorithm so a computer can run it.

Instructions, variables, conditionals, loops.

Runs a step only if something is true — which is where decisions, and unfairness, enter an algorithm.

Repeats steps until a condition is met, applying one rule to every item — which turns a rule into a policy.

Wherever the line falls, someone sits just the wrong side of it, and the rest of the system cannot see them.

Natural language, flow chart, pseudocode, and code in a programming language.

A rectangle is a step. A diamond is a decision with two ways out.

It decides who can read and therefore check the rule. Code excludes almost everyone; a flow chart excludes almost nobody.

It is ambiguous — the one thing an algorithm must not be.

Sorting (reorder), searching (find one), filtering (remove), classifying (label).

Prioritising, associating, counting, software development, and machine learning.

Sorting keeps everyone and changes the order. Filtering removes, and the removed are invisible to everything downstream.

What happens to the person at the bottom of the ranking?

Bias and fairness; accountability; transparency and black boxes; erosion or loss of human judgement.

The training data, who is missing from it, stand-in measures, and the target chosen — not from a biased rule.

Other fields stand in for it — postcode, school, a gap in work — so the model rebuilds it.

Knowing why you were rejected is useless if nobody is obliged to reconsider. Accountability is a separate requirement.

Mainframe, server, personal computer, tablet or smart phone, embedded system.

A computer built inside another product to do one job — a washing machine, a card reader, a traffic light.

Nobody thinks of them as computers, so nobody asks who updates them, who can reach them, or what they record.

Roles, not machine types. The same computer can act as either.

Input, process, store, output.

Hardware is the physical parts. Software is the instructions telling them what to do.

Memory (RAM) holds what is in use and empties when power goes. Storage keeps things when it does.

A CPU does one complicated thing at a time, very fast. A GPU does thousands of simple calculations at once — which suits AI training.

Writing instructions in a language with a specific syntax so that a machine can carry them out.

Open: anyone may read, change and share the code. Closed: only the owner may — but there is someone accountable.

It moves a judgement from a person who could make an exception into a rule that cannot.

Who is able to check it? That turns a software question into a digital society answer.

Vacuum tubes (1940s), transistors (1950s), integrated circuits (1960s), microprocessors (1970s on), AI and cloud (now).

The switching technology inside — each one smaller, cheaper and more reliable than the last.

Components on a chip roughly doubled every two years. A prediction the industry organised around, not a law — and it has slowed.

Computing had been moving closer to people; the cloud moved the processing back into a data centre and left a screen in your hand.

Wired, wireless and cloud — three layers of one connection rather than three choices.

Around 99% of international traffic runs through undersea cables. Only the last hop is usually wireless.

Computing rented from a provider's data centre rather than owned.

Where the data physically sits, which country's law applies, and whether the service exists tomorrow.

Client, server, modem, router, switch, hub.

A switch connects devices within one network. A router passes traffic between networks and chooses the path.

A switch sends each message only to its destination. A hub sends everything to everyone — slower and less private.

Roles, not machine types. Whatever asks is a client; whatever answers is a server.

Bandwidth is how much data passes per second. Latency is the delay before the first data arrives.

An agreed set of rules for how machines communicate — how to start, how to send, how to report an error.

The rules are public, so anyone can build equipment that joins the network — which is why no company owns the internet.

Routing around damage makes a network hard to break by accident and hard to censor on purpose. The same property.

Email, hosting, cloud storage and backup, phone service, domain names, filtering.

Which sites you reach and when — the pattern — even where the contents are encrypted.

Everything passes through them, so courts, governments and rights-holders all enforce through them.

The power is unavoidable — someone runs the wires — so the question is whether its use is visible and appealable.

The internet is the global network of networks. The web is ONE service on it — linked pages fetched by a browser using HTTP.

Internet: ARPANET 1969, TCP/IP 1983. Web: proposed 1989, public 1991.

If a browser and a page are not involved, it is not the web.

Blocking a website touches one service. Shutting off the internet touches everything. They are different measures.

1.0: a few publish, everyone reads. 2.0: users make the content, platforms own the place. 3.0: contested — semantic, or decentralised.

ARPANET 1969, TCP/IP standardised 1983, the web proposed 1989 and public 1991.

Moderation. What may be said became a company policy rather than a law.

Power to speak widened enormously; power to decide what is heard narrowed to a few companies.

Privacy, anonymity and surveillance; the right to be forgotten; the digital divide; freedom of expression and censorship.

The harm and the benefit come from the same property, so removing one removes the other.

Google Spain v AEPD, CJEU 2014 — search results may be de-listed where inadequate, irrelevant or excessive. The page itself stays.

A network built to route around damage treats blocking as damage, so material reappears by another path.

Text, images, audio, animations, video and web pages.

Content generated rather than recorded — synthetic text, voice, images or video.

It is not a new form but a new origin for every existing form, so any rule about it touches all of them.

It demands the most attention and therefore holds it longest, and engagement-based ranking promotes what holds attention.

Rapid sharing, efficient storage, interactivity, linear and non-linear paths, perfect copying.

Linear runs in one order set by the maker. Non-linear lets the audience choose a path — from options someone else decided.

Responses can be recorded, which turned an audience into a dataset.

Non-linear media means nobody sees the same sequence, and a duller correction is shared less and ranked lower.

AR adds a layer over the real world. VR replaces it entirely. MR lets virtual and real objects interact.

Surgical rehearsal, aircraft maintenance, emergency response training, empathy exercises.

Where the user looks and how they move — more intimate than a click.

Ask what the technology is used INSTEAD of, not whether it works.

Psychological concerns; truth and synthetic media; ownership and plagiarism; representation and moderation.

Misinformation is false and shared without intent to mislead. Disinformation is false and spread deliberately.

It is duller, so it is shared less and ranked lower — the same mechanism that spread the claim suppresses its correction.

Responsible for what they AMPLIFY rather than for everything posted, because ranking is the part they choose.

Narrow (weak, domain-specified) — one task, and all that exists. General (strong, full) — any human task, does not exist. Super-intelligent — hypothetical.

No. Weak describes the RANGE of tasks, not the quality. A weak AI can beat every human at its one task.

If a person cannot tell a machine from a human in conversation, it passes. It tests convincing, not understanding.

Arguing from general AI. Real questions are about narrow systems doing exactly what they were built for.

Supervised (labelled examples), unsupervised (no labels, finds groups), reinforcement (a goal and a score).

No — it describes many layers in the network, and can be supervised, unsupervised or reinforcement.

Pattern recognition, facial and speech recognition, image analysis, natural language processing.

The labels, who is missing from the data, stand-in measures, and the target chosen.

Layers of simple units joined by weighted connections. Numbers in, numbers out; training adjusts the weights.

Modelling complex, non-linear relationships, and generalising from their training to inputs they have not seen.

It holds millions of weights, not rules. There is no line to point at and no sentence to read out.

“Thinks”. Say it models a relationship — the guide's own wording, and it keeps the answer defensible.

A period when the promises outran the results and funding was withdrawn. There have been two: the 1970s and the late 1980s.

They captured what an expert SAID, not what an expert noticed, so they broke on the first case outside the rules.

Learning from data removes the cause of both winters — nobody has to write the exceptions down.

The method is genuinely different; the expectations are not — and it is expectations that have collapsed before.

Fairness and bias; accountability; transparency; uneven and underdeveloped regulation; automation and displacement.

Competing definitions — equal accuracy across groups and equal error types across groups — cannot both hold except in special cases.

Mostly about USE: who may deploy it, for what, with what duties — as in the EU AI Act's high-risk categories.

“Prompt, not decision” for automated decisions about people; “responsible for what it amplifies” for platforms.

Industrial (repeats a path, caged), service (works among people), social (interacts with people as its purpose).

Internet of things, autonomous vehicles, drones, virtual assistants.

The da Vinci system is teleoperated — a surgeon moves the instruments through a console. It augments rather than replaces.

How much the machine decides for itself, from a caged arm that decides nothing to a vehicle deciding continuously.

Sensory inputs for spatial, environmental and operational awareness; reasoning with machine vision and learned models; acting physically; some degree of autonomy.

Spatial (where am I), environmental (what is around me), operational (how am I doing).

A machine cannot react to what it has no sensor for.

A recommendation can be ignored. A movement has already happened and cannot be un-made.

Mechanical automata → caged industrial arms (1960s) → sensing robots → service robots among people → autonomous systems deciding.

By separation — keeping people out of reach — not by the robot noticing them.

Sensing. A machine that can detect a person can slow down or stop for them.

The pedestrians and cyclists sharing the risk never consented, where a factory worker did.

Anthropomorphism and the uncanny valley; responsibility for harm; safety; employment and displacement; surveillance.

Comfort rises as a machine becomes more human-like, drops sharply when it is almost-but-not-quite, then rises again.

Treating a machine as though it has feelings and intentions — which people do readily and designers encourage.

What is the machine funded INSTEAD of? It turns a technology answer into a social one.

Publishing became free; being heard became a ranking decision rather than a person's.

The platform sets the payment per play, so a large audience no longer produces a large income.

The work that was never made because it would not have suited the system.

Readers now decide what sells — but the publisher still prints it and still takes most of the money.

Routine rather than content — when you are in, what you watch, how you sleep.

It predicts behaviour, and it does not feel private, so almost nobody objects to it being collected.

The journey home — the boundary that ended the working day and that nobody had to maintain.

Watches that detect an irregular heart rhythm a yearly check would miss entirely.

Appearance, text and sound — and the record survives fire, flood and war.

Who may see it, when it may be performed, what presence means — rules held by a community, not properties of the object.

Not whether to digitize, but who holds the copy and whether the source community can still say no.

Once the copy is published, the community that set the rules about access no longer decides.

People too rare to meet in one town are one search apart once distance stops limiting who finds whom.

The ranking brings in people who do not know the group's rules, faster than it can absorb them.

Easy discovery builds support groups and harmful groups alike — it is one property producing both.

Limit amplification rather than existence: stop recommending, do not stop allowing.

Sell it once, subscribe, advertising, commission.

The advertiser is the customer, so the product is built to hold attention rather than to be useful.

A business that provides the place where two other groups trade, takes a share of each transaction and sets the terms.

Each side is there because the other is, so a competitor has to attract both at once.

Automation takes tasks out of a job rather than removing the job — the job that remains is a different job.

Predictability, volume, how cheap an error is, and whether the data already exists.

Skilled tasks are often predictable and richly recorded — reading a scan is easier to automate than clearing a table.

Entry-level tasks were how people learned the job, so removing them removes the route into the field.

Digital things copy; money must move. Spending has to remove it from the sender.

One organisation holds the record, or thousands of copies of one shared record are held.

Control by an organisation you can appeal to, against control by a rule that cannot be appealed to.

It used phones people already owned, so it needed no new branches or infrastructure.

Digital coordination became nearly free, payment became immediate, and demand became visible everywhere at once.

At design and at the sale — the two steps involving the fewest people.

At mining, assembly and disposal — the steps furthest from where the profit is recorded.

The government with the power to act has no harm in front of it, and the one with the harm has no hold over the company.

Digital systems reveal harm at enormous scale and consume electricity, water and mined metals to do it.

Scale, continuity over time, and observation where no inspector would be allowed.

Evidence only works where an authority has both the power and the will to act on it.

Did anybody decide differently because of what it showed?

Mined, made, used, discarded — owners see only one of them.

Making it. For a phone, manufacturing produces far more emissions than using it ever will.

Recycling acts after the emissions have happened; a longer life means they happen less often.

A guaranteed minimum period of security updates alongside the spare parts.

Sense, decide, adjust — continuously, without waiting for a budget cycle.

People walking, people without phones, and areas that were never instrumented.

There is no meeting and no decision to point at, so residents have nothing to appeal against.

What makes infrastructure efficient makes it reachable — a pipeline shutdown stopped fuel for a region with nobody touching a pipe.

Measures variation within a field so seed, water and fertiliser go only where they are needed.

It reduces an input while holding the output, rather than moving a harm somewhere else.

Field data and repair rights both move to the machine's manufacturer.

Rules giving the farmer the data and the right to repair — separating the environmental gain from the loss of control.

Symptom, access, triage, diagnosis, treatment — digital systems have entered all five.

Does the system propose to a person, or does it decide? An error in the second has no owner.

An error is a person waiting, and nobody sorted downwards is told they were sorted.

It hides much worse performance for groups barely present in the training data.

Carried, worn, implanted, and reading the nervous system.

Refusing costs more at each level — at the bottom it costs your health, so agreement is formal rather than real.

It cannot be changed, is collected continuously, and reveals illness, pregnancy and mood.

Limits on what may be collected at all, because a limit does not depend on the person's bargaining position.

The cause may run either way, or a third factor may move both.

A large effect on a small group disappears into a small average — and that group is the concern.

Comparison, disturbed sleep, harassment that follows a person home, and targeted content.

It names something a platform could change, so refusing to change it becomes the thing to judge.

Delivering, practising, noticing who is lost, making somebody want to continue.

Delivering and practising move well; noticing and motivating barely move.

Both depend on a relationship with a particular person, which no system supplies.

What surrounds the tool — a device, a quiet hour, somebody to ask — decides who finishes.

Ask, gather, analyse, check, publish.

Gathering and analysing — and publication became immediate.

It takes the same work it always did and nobody is rewarded for doing it, so results outran verification.

Publishing the data and the code, and registering the method before the data is examined.

A shared message that opponents and journalists could check — the accountability came with the inefficiency.

Each message is seen only by its group, so promises cannot be placed side by side.

Visibility — anybody can see what was said to whom, without restoring the shared audience.

Participation tools are optional and small; scrutiny losses reach every voter, including those who use none of these tools.

Available at any hour, identical for everybody, much cheaper, and traceable.

The official who could see that a case was unusual and do something about it.

It becomes the key to everything, so a single exclusion removes access to many services at once.

A real, reachable route to a person who can decide differently.

Infrastructure attacks, information attacks, autonomous weapons, and dependence on private infrastructure.

Every rule about conflict assumes you know who acted, and digital conflict is designed to be deniable.

A wrong selection has no person whose decision it was, so no one can be held responsible.

A company's commercial decision about service coverage can become a military decision — an arrangement no rule anticipated.

A rule can only be written after a harm appears and is measured, and each step takes years.

Name the technology, state principles, or sort by the risk of the use.

They govern by use rather than by name, so a system invented tomorrow is covered the day it is used for a risky purpose.

When the market is worth more to the company than the cost of complying — which makes reach a function of market size.

Find each other, get shown more, define against another group, stop meeting them.

What the system rewards — hostility holds attention, so the ranking favours the version with an enemy in it.

Platforms amplify division rather than cause it — narrower, better evidenced, and it points at a specific fix.

People whose group is rare where they live gain both practical knowledge and recognition.

Access (do you have it?), use (what can you do with it?), outcome (what do you get from it?).

The gap is produced by what surrounds the tool — a device of your own, quiet time, somebody to ask.

They close the first divide, which is also the one that gets measured, and leave the other two.

Having somebody to ask — being stuck ends one student's session and costs another three minutes.

Words, faces, news and arrangements — enough to stay informed and be present for big events.

Ordinary unplanned time, which is where close relationships are actually maintained.

Being trusted — which is how a young person learns to make a decision nobody will check.

Agreed rather than hidden, with a stated end — covert monitoring loses the safety benefit the moment it is found.

Converting a tool into money, work or study takes time, skills and contacts — exactly what the group with less lacks.

Give access, build the ability to convert, change what the technology rewards.

It is the one that can be counted and reported, and it leaves the conversion problem untouched.

Recommend, justify, state what it does not reach, conclude. The limits card separates the bands.

Ageing needs care delivered with fewer people; moving needs identity, money and language across borders.

The group with the highest need has the lowest device ownership and least confidence — so the service is aimed at the least equipped.

Identity that survives a border, cheap transfers home, multiple languages, and no assumption of a fixed address.

Publish waiting times for the digital and non-digital routes side by side — a number rather than a promise.

Retrain, redistribute, reduce hours, redesign what gets automated.

Nearby work, time to train, and an income while training.

A worker whose tasks could be automated negotiates against that possibility whether or not it happens — employment figures miss it.

Relocation support helps individuals and lets a region keep declining; they are different objectives.

Every measure buys real security and spends privacy, a new way in, or a power that outlasts the emergency.

Removing a power means arguing publicly that the threat has gone, so measures are extended rather than repealed.

By what it will be used for in ten years, not by this emergency — and attach a review date.

It collects from organisations rather than from people, so it corrects an incentive instead of removing a right.

Told, asked, heard, deciding.

In setting the question. A consultation on somebody else's options hands over much less than it appears to.

A published response saying what changed as a result and what did not.

Time, confidence, the language of the process, and what people happen to be shown.

Who is in the data, what it was trained to predict, how fairness was tested, and how the output is used.

It reaches only the first entry point; the label still encodes the past decision the system learned from.

Each averages over everybody with that characteristic, so a small intersection disappears into the average.

When the person reading it has thirty seconds and has to justify disagreeing with it.

Adds: data centres, making devices, cooling water. Saves: journeys, heating, wasted generation.

Variable output cannot be scheduled, so balancing supply and demand needs continuous measurement and control.

Cheaper per use produces more uses, so a saving per use is not a saving overall — judge on the total.

Device lifetime and grid balancing. Individual behaviour is talked about most and measures least.

Take, make, use, discard — the default every device follows unless something intervenes.

Rank by how close each gets to the making stage, because that is where the footprint is.

Support ends, batteries are glued in, parts are unavailable, and repair is tied to the manufacturer.

It acts after the emissions have already happened — it reduces the next build rather than preventing this one.

It is labelled as working second-hand goods, and a share of it genuinely is working.

People who strip and burn devices by hand — they never owned them and gained nothing from them.

It is the one party the exporting country can reach, because it sells into that market and wants to continue.

The record follows the device, so relabelling waste as working goods no longer breaks the chain.