License to Scroll

In plain English

For your whole life, "I saw it with my own eyes" settled an argument. A photo was proof. A video was better proof. That era ended somewhere around 2022, and most people haven't been told.

Computers can now make a face, a voice, or a whole video of a person who never existed, or of a real person doing something they never did. Not clumsy, not blurry. Good enough that trained researchers get fooled.

In plain English

When you watched the evening news, a room full of editors decided what went on. You could argue with their choices, but a person made them. On YouTube, Facebook and Instagram, no person picks what you see. A program does, and it has one job: keep you on the app as long as possible, because that's how the company sells ads.

The program doesn't know what's true. It knows what you stopped scrolling for, what you watched to the end, what you commented on. It learned, from billions of people, that fear, outrage and "can you believe this" are what make people stop. So that's what it brings. It's a waiter who noticed you reached for dessert once and now brings dessert every night, and three desserts if you ate the first one.

Here's the part the documentaries leave out, and it matters because it's the truth: in a 2024 study, researchers built robot accounts that only followed YouTube's recommendations, and compared them with real people. The robots ended up watching less extreme stuff than the humans chose for themselves. The feed isn't brainwashing you. It's amplifying whatever you already lean toward, and it never gets tired.

In plain English

Everything else in this course is a lesson, and lessons fade. In the Cambridge studies, the effect of media training wore off within about two months unless people got a refresher. A setting doesn't fade. Autoplay stays off until somebody turns it back on. So this is the session that does the most work for the least effort, and it's the one to do with somebody younger on the phone if the menus look different from what's below.

None of this stops you watching anything. It stops the app from choosing for you. You'll still find everything you go looking for.

In plain English

A used-car salesman has the same moves whether he's selling a Ford or a Chevy. The moves don't depend on the car. Same here: the tricks that get you worked up online are the same whether the post is left, right, or about a celebrity. Once you can name the move, it loses most of its power, the way a magic trick stops working once you've seen where the coin goes.

Aristotle named the three ways to persuade a person 2,300 years ago, and nobody has found a fourth:

• Ethos, "trust me." A man in a white coat. A badge. "As a mother of four." Borrowed authority. The question: is this person actually an expert on this, and would other experts agree?
• Pathos, "feel this." The sick child. The crying veteran. The Black grandmother being mistreated. Pictures chosen because they hit you in particular. The question: is the feeling doing the work the facts should be doing?
• Logos, "here's the proof." A number, a chart, a study. This is the one liars fake best, because a true number with the context removed is the most convincing lie there is. "Crime up 300%" (from one incident to four). The question: compared to what, and out of how many?

The Cambridge researchers add five everyday moves: emotional language (words picked to make your pulse jump), incoherence (claims that contradict each other, hoping you won't notice), false dichotomy ("you're either with us or against us"), scapegoating (one group blamed for everything), and ad hominem (attacking the person so you never examine the point).

In plain English

The fastest way to spot a con is to run one. Researchers built a game where you play the bad guy: you run a fake-news operation in a made-up town and try to get people fighting. It's silly on purpose. It also works; in the study, people who played it got measurably better at spotting the real thing.

Three true stories, in order

1918 to 1945: the posters. Governments on every side learned that a frightened crowd is an obedient one. Fear plus "they are not like us" won wars and started them. The playbook was written in paint before it was written in code.

1969: "Doubt is our product." That line comes from an internal tobacco-company memo. They didn't need you to believe cigarettes were safe. They only needed you to feel the science wasn't settled. Manufactured doubt is cheaper than a lie and harder to catch. You will see it every day: "experts disagree," "nobody really knows," "do your own research."

2016: Houston, Travis Street. On a May afternoon, a "Heart of Texas" Facebook group organized a protest outside the Islamic Da'wah Center downtown. A "United Muslims of America" group organized a counter-protest at the same spot. Real Houstonians showed up on both sides and shouted at each other across the street. Both pages were run from the same office in St. Petersburg, Russia. The same operation ran pages aimed at Black Americans, posing as activists, with one goal: get people angry at each other and discourage them from voting. They didn't care who won the argument. They wanted the argument.

In plain English

Stanford researchers watched three groups try to judge websites: historians, college students, and professional fact-checkers. The historians and students read the page carefully, looked at the "About Us," judged the design. They got fooled. The fact-checkers did something different: they left the page almost immediately, opened a new tab, and looked up who was behind it. They were faster and more accurate.

You don't buy a used car by reading the sign the seller taped to the windshield. You call around. Reading sideways is calling around. The method has four steps and a name, SIFT:

1. Stop. The moment you feel your pulse jump, that's the signal. Don't share, don't comment, don't let it in yet.
2. Investigate the source. New tab. Search the name of the site, the channel, the person. Thirty seconds. What do others say about them?
3. Find better coverage. If it's big news, is anyone you already trust reporting it? If only one source has the "bombshell," that's the tell.
4. Trace to the original. A quote, a clip, a study: find the whole thing, not the piece somebody cut out. The missing context is usually the whole story.

And for pictures: on the computer, right-click any image and choose "Search image with Google." You'll see everywhere it has appeared, and often that it's from a different year, a different country, or a movie.

In plain English

Three things to know, in order of how much they can hurt you.

Voices first, because this one costs people money. A computer needs only a few seconds of someone's voice, from any video they've ever posted, to make that voice say anything. The call comes at night. It sounds exactly like a grandchild: "I'm in trouble, I need money, please don't tell anybody." The fix is old-fashioned. Pick a family safe word today, something only the family knows, and agree that any emergency call has to say it. If the caller can't, hang up and call that person directly.

Second, the channels nobody runs. A growing share of YouTube is made by programs, not people: a robotic narrator reading a script a computer wrote, over stock footage a computer picked, with a thumbnail that says SHOCKING in red. Nobody fact-checked it because nobody wrote it. Signs: the voice never breathes or stumbles; the footage doesn't quite match the words; the channel is a few months old and posts dozens of videos a week; no real person's name appears anywhere; every title is about a celebrity or politician "finally exposed."

Third, the honest truth about spotting fakes. Every tip for spotting fake pictures (six fingers, strange teeth, blurry backgrounds) was true for about a year and then stopped being true. Detection is a race, and the fakes are winning. The thing that lasts is provenance: who made this, and can they prove it? Cameras and software are starting to attach "content credentials," a kind of nutrition label that records who made an image and whether a computer generated it. You can check any image for one at the link below. One warning: most social media strips those labels off when you upload. So "no label" does not mean fake. It means "unknown," which sends you back to Session 6.

In plain English

A number feels like a fact. That's why the best lies are true numbers with the context removed. Four tricks cover almost all of it:

• The percentage with no count. "Cases doubled!" From two to four. Always ask: out of how many?
• The chopped-off chart. A graph whose bottom starts at 95 instead of 0 makes a one-point change look like a cliff. Look at where the axis starts.
• The convenient window. "Up 40% since 2020" sounds different when you learn 2020 was the lowest year in a decade. Ask: compared to when?
• Two things that rise together. Ice cream sales and drownings both peak in July. Ice cream doesn't cause drowning; summer causes both. "Linked to" is not "caused by."

None of this needs math. It needs two questions, asked every time: out of how many? and compared to what?

In plain English

Four cases. For each one, the story you probably heard, and then what the evidence actually showed, because the honest version is more useful than the scary one.

The scammers (the one that matters most)

Older adults lose the most money per scam of any age group, and the biggest category of reported fraud is impersonation: someone pretending to be Medicare, the IRS, Amazon, a bank, or a grandchild. The call creates panic and a deadline. Every one of them wants the same thing: gift cards, a wire, or a "verification" of your account number. The rule that defeats all of them: hang up, and call the real organization yourself at a number you look up. Not a number they give you.

Cambridge Analytica

The story: a company stole 87 million Facebook profiles, built mind-control ads, and swung the 2016 election. The evidence: the theft of the data was real and the company was sleazy. But when the British data regulator finished its investigation in 2020, it found the firm's own staff doubted their targeting worked, and independent reviewers found no good evidence the ads changed anybody's vote. The real scandal is what Facebook let happen to your data. The mind-control part is marketing, mostly the company's own.

The Russian troll farm

The story: Russian posts reached everyone and tipped the election. The evidence: the operation was real (you met it in Houston in Session 5). But a 2023 study of Twitter found that 1% of users received 70% of the Russian posts, almost all of them people who were already strongly partisan, and found no measurable change in their opinions or votes. It was a loud operation with a small audience. The lesson isn't "it didn't matter." It's that the goal was to make people angry at each other, and on that score it got help from all of us.

Facebook in Myanmar

The case where the feed caused real deaths. In 2017, hate posts against the Rohingya minority spread on Facebook, where it was most people's only internet. The chairman of the UN fact-finding mission said in 2018 that social media had played a "determining role" in spreading the hatred that preceded mass killings and the flight of hundreds of thousands. The feed amplified what got reactions. Hate got reactions. Facebook later admitted it hadn't done enough.

In plain English

This session is about you, not the feed. Three findings.

The mean world. Starting in the 1960s, George Gerbner studied people who watched a lot of television. They believed the world was far more dangerous than it was: more crime, more strangers out to get them, less reason to trust a neighbor. He called it "mean world syndrome." The news wasn't lying to them about any single event. The pile of events, with no sense of how rare each one was, did the damage.

Fourteen minutes. In 1997, British researchers showed people a news bulletin edited to be negative. Fourteen minutes. Afterward, those viewers weren't just sadder and more anxious; they worried more about their own problems, problems that had nothing to do with the news. Fourteen minutes. Now think about three hours of YouTube.

Why the drive to a long drive feels dangerous. The brain judges how likely something is by how easily it can picture it. That's called the availability heuristic. After a week of crash videos, you can picture a crash in vivid detail, so your gut says the road is deadly, even though the road didn't change and the odds didn't change. The videos changed what's easy to picture. The feed found out you'd watch, and it kept the pictures coming.

Here's what doesn't fade: nobody can pour that much fear into you if you're only holding the cup out twice a day.

Four habits that work

• News at set times (say, after breakfast and before dinner), not all day. Twice a day is enough to know everything that matters.
• Nothing from the feed in the hour before bed.
• Following feeds, not For You feeds (Session 3).
• One question at the end of anything that upset you: "Was that useful, or just upsetting?" If it's the second one, that's the kind to prune.