Exam practice

Any four, for example:

• Infinite scroll: new posts load whenever you reach the bottom, so there is never an end, and no natural moment to stop.
• Autoplay: the next video starts by itself; carrying on happens if you do nothing, and stopping needs a decision.
• Variable rewards: pulling to refresh, or the next short video, sometimes gives something great and sometimes not; unpredictable rewards keep you checking.
• Likes, views and followers: approval turned into a number that changes, so you come back to see it grow and compare it with others.
• Notifications and red badges: the app calls you back when you weren’t thinking of it, often with messages the app itself made.
• Streaks: a count of days in a row that resets if you miss one, so you open the app daily for fear of losing it.

1. You watch, skip, replay, like, share or comment on a video.
2. The app records what you did, with metadata such as the time, how long you watched, and your phone and language.
3. A model uses this, and what people who behaved like you did next, to guess for each possible video how likely it is to keep you watching (or to be liked or shared).
4. The video with the best guess plays next, and your reaction to it becomes more data: the loop repeats with every video.

The goal is decided by the app’s designers and the company: for example “keep people watching as long as possible”, or one that mixes in variety. The model only does what it is set up to achieve.

A filter bubble is when a person sees mostly things that match what they already watch or believe, and little else, because a recommendation system keeps showing more of what held them.
Ways to stop it (any two): use Not interested on videos you don’t want more of; search on purpose for things you want to learn and follow people who teach you something; check every few days what your feed is showing and whether you chose it; reset or refresh the feed in the app’s settings; get information from places you choose yourself (a search, a book, a website) rather than the feed.

Benefits (any two): suggestions and search results that suit you; spotting a stolen account by a strange login; ads paying for the app so it is free.
Risks (any three): the data can be used against your interests (to keep you on longer, or show ads when you are most likely to buy); it can be stolen in a break-in and sold; it is kept for years and may be used in new ways; it is stored on servers in other countries, under their laws, out of your control; put together, it can reveal private things you never chose to tell (your mood, your school, when you sleep).

It can reveal the date and time the photo was taken, the phone used, and often the exact place from the phone’s GPS.
That is dangerous because a photo taken at home, at school or somewhere you go often, if shared with its metadata (for example sent as a file), can tell a stranger where you live or where to find you.
You can turn off saving the location in the camera’s settings, check a photo’s details before sending it, and share photos in ways that remove the location.

• i.From midnight to 6 am is 6 hours (less, if it takes him time to fall asleep). The AAP recommends 8 to 10 hours for ages 13 to 18, so he is at least 2 hours short every night.
• ii.Any three, for example:

• Charge the phone outside his bedroom at night: nothing to scroll in bed, no notifications waking him.
• Screens off an hour before bed: the endless feed can’t keep him up, and he avoids bright light late at night that makes it harder to fall asleep.
• Bedtime mode from, say, 9 pm: the phone goes silent and gray.
• A daily limit on the app he scrolls most, so “one more video” meets a decision.
• Go to bed at a fixed time, about 9:30 pm, to get at least 8 hours before 6 am.

The 20-20-20 rule: every 20 minutes, look at something at least 20 feet (about 6 meters) away for at least 20 seconds.
Eyes get tired (eye strain) because when we stare at a screen we blink much less, only about 5 to 7 times a minute, so the eyes dry out; and focusing on something close for a long time without a break tires them. It doesn’t cause lasting damage, and rest makes it go away.

A bystander is someone who sees cyberbullying happen but is neither the bully nor the target: for example, a member of the group chat where it is posted.
Four things to help: don’t like, laugh at or forward it; message the target privately to say it wasn’t okay and ask if they are all right; report the post or account in the app and save screenshots; tell a trusted adult or a teacher who can stop it.
A like is a decision because it tells the bully, and everyone else, that people enjoy it, and the app may show a post with many likes to even more people. Doing nothing is a decision too: it tells the target that nobody cares.

1. Would they agree? (If you can’t ask, don’t share.)
2. Could it embarrass or hurt them, now or later?
3. Does it show where they are, such as a uniform, their house or a place they go every day?
4. Is it a child? (Then ask a parent too.)
5. Was it meant only for you?

Never share (any two): a sexual or naked picture of anyone under 18; a photo of someone injured, ill or dead; a picture edited or made to mock someone.

A digital footprint is the trail of data a person leaves online: everything they post and everything recorded about what they do.

• Active footprint, what you put there yourself: for example posts, photos, comments, reviews, your profile picture.
• Passive footprint, what is recorded about you without your posting it: for example the videos you watched, your searches, the places your phone has been, the ads you tapped.

A student should care because it lasts for years (copies, screenshots, archives, company records), and schools, scholarship committees and employers often search people’s names before deciding. A cruel post can count against them; good things, like projects and helpful answers, can count for them.

Gains: the app company (people watch longer, so it shows more ads) and its advertisers; users on fast, unlimited Wi-Fi get smooth videos at no extra cost.

Pays: users on small mobile data packs, common in Nepal, whose data goes on videos they never chose to watch (high quality can use a gigabyte in well under an hour); people with weak signals, whose videos stall; anyone who loses time or sleep because there is no moment to stop.

A fairer design: autoplay off on mobile data, or low quality until the user taps; a clear switch for autoplay and quality, asked about when a new user joins; a data-saving mode on by default where data is expensive. The decision isn’t neutral: designers should look at who it costs, not only at who it earns for.

Sita is right that a recommendation system set only to “keep people watching” can keep people on far longer than they meant, narrow what they see into a filter bubble, and collect a lot of personal data. Rohan is right that recommendations can be useful: with millions of videos, no one could find good ones without help, and many people discover new interests this way.
The problem isn’t recommending; it is the goal and the control. A good recommendation system would aim at what users say they want, not only at the minutes they spend; mix in variety; let users see and change why they are shown things (Not interested, reset, a feed they choose); collect only the data it needs; and make it easy to stop, with ends, limits and quiet hours. (Any well-argued answer that takes something from both sides and names features of a better design gets full marks.)

Any three, with evidence, for example:

• A spam filter. Helpful: it keeps out almost all junk and scam emails, so fewer people are tricked by phishing. Harmful: it sometimes sends a real email, such as a scholarship reply, to Spam, where it isn’t read. Unfair to: small senders it hasn’t seen much of, like a school with a new email address.
• A map’s route. Helpful: it finds the quickest way and a way round traffic jams, using the speed of other phones. Harmful: an out-of-date map sent a car onto a broken bridge in Bareilly, India, in 2024, and three men died. Unfair to: villages and new roads that are mapped badly, and people on quiet lanes it uses as shortcuts.
• Surge pricing. Helpful: a higher price is meant to bring more drivers out when many people need rides (though the evidence is argued over). Harmful: during the Sydney hostage siege of 2014, Uber’s prices rose automatically to about four times the usual as people fled. Unfair to: people with less money, priced out when they most need a ride.
• Autocorrect. Helpful: it fixes spelling mistakes as you type. Harmful: it changes names and words it doesn’t know. Unfair to: people whose names aren’t in its list; in 2024 a campaign found a spell checker marked about 41% of babies’ names in England and Wales as mistakes, many of them African, Asian, Irish and Welsh.
• A feed’s recommendations. Helpful: it finds videos you enjoy among millions. Harmful: it can keep people on far longer than they meant, and narrow into a filter bubble. Unfair to: students on small data packs, when videos autoplay in high quality.

Full marks need real evidence (an event, a study, a fact from a reliable source, or something you saw, with when), not just “it is good” or “it is bad”.

How: teachers gave each student an expected grade and ranked them. For big classes, the algorithm mostly set the teacher’s grades aside: it took the grades the school’s students had got in the three years before and gave this year’s students the same pattern of grades, in the teacher’s order. Small classes kept more of their teachers’ grades.

Why unfair: a student was judged partly by how earlier students at her school had done, not only by her own work, so a very good student at a school with poor past results could not get a top grade. About 39 in 100 teachers’ grades were lowered. Private schools, which have more small classes, gained most: their share of A and A* grades rose by 4.7 points, against 2 points at ordinary state schools.

In the end: students protested, and four days after results day the government gave every student their teacher’s grade instead (or the algorithm’s, if higher). The algorithm did what it was designed for, keeping results like other years’, but nobody had asked whether that was fair to each student.