Understanding AI: Online Content

Generative AI is a new and evolving technology. It has many productive uses like acting as a research aid for homework or an assistant when planning an event. But like most technology that came before, generative AI can also be used for harmful purposes; for instance, it can be used to create fake websites, voice recordings, videos, images, and social media posts that mislead, embarrass or hurt people. What used to require a lot of technical skill to create, can now be done easily: most people who make fake content use widely-available AI tools.

Large language models (LLMs), which are the AI behind many "chatbots " or “assistants”, are not programmed like regular computer code. Instead of being programmed with specific rules, they learn by finding patterns in enormous amounts of text. Data used in training is not stored in a model; it’s used to update model weights which then produce model responses. When you ask questions in the form of a “prompt,” they use what they've learned about how words and sentences fit together to come up with relevant answers. Because they learn from existing information, their answers will only be as good as the information they were trained on. If that information has a bias, the LLM's responses might reflect that bias.

Recommendation algorithms are what social networks and video sites use to decide what to show you. They usually are given several goals, then record how you respond to different content and show you more of what they think will meet those goals.

Half of teens say they use AI to find information online, but more than a third also say they have been misled by AI-generated content - and more than a quarter have been unsure about whether they were talking to a human or a chatbot. That’s why we need to teach our kids about different kinds of AI media and how to handle them. But just telling them to be skeptical, without teaching them how to tell synthetic media from real content, can actually make things worse. It can lead us to doubt everything we see, whether it's true or false. In particular, we are most likely to think that things we don’t want to be real are made with AI. 

Chatbots are good at giving answers that are very fluent and persuasive, even when they’re not true. If your question is based on a misconception (e.g. “What patterns are hardest for a chameleon to camouflage against?”) the response may treat that misconception as true instead of correcting it. 

What should we do about it?

We can’t always trust our eyes when trying to spot AI-generated images or videos. Older signs of fake images, like strange-looking hands or uneven eyes, are now much less common with newer technology and can no longer be relied on as clues that something was made with AI. Also, relying on these visual flaws can be misleading: many smartphones now automatically “enhance” photos and videos in ways that can give them the smooth-and-shiny look associated with AI-generated images, and as many as one in five hundred people really do have at least one extra finger! 

Instead, we need to teach kids to use information sorting techniques, like those in programs such as MediaSmarts’ "Break the Fake." These methods teach you to look for information from other, more reliable sources that are much harder to fake. Learning to identify and understand the context around an image, and questioning who is sharing it, is key to developing media literacy and personal fact-checking skills.

• Treat visual cues as “red flags” that tell us to be extra skeptical – but don’t assume that something is fake if it has them, or that it’s real if it doesn't.
• Watch out for things that provoke a strong emotional reaction or that a lot of people want to know about. AI scammers have said they often target fan communities because they feel so strongly about their subject. (We’re also more likely to wrongly think something real is AI if it’s on a topic we feel strongly about.)
• Try to find the original source of a story, photo or video. Is there good reason to think they are trustworthy or would have been able to take that photo or video?
• Look for other sources about a story or claim and look for other versions of an image or video. Can you see the same people and things from different angles?
• Use search engines and Wikipedia to make sure sources are real and have a good claim to expertise. Double-check the web address to make sure it’s the source’s real website.
• Teach kids a family password to use if they ever have to ask you for help. Don’t reply directly to a call (such as by pressing a number on your keypad or calling a number they give) or to a text or email. Find the organization’s real website, find their contact information there, and find out from them whether the message was legitimate.

Using AI as a source of information

For important information, go directly to a source you already know is reliable. Although some chatbots have “guardrails” around certain types of content, they’re not foolproof, and they can sometimes give inappropriate information, so make sure to provide kids with reliable sources of information on things they might want to turn to a chatbot for (like Kids Help Phone for mental health or Sex&U for sexual health) and make sure they know to talk to you if anything they see online upsets them.

If you’re using a chatbot to get information, give it a detailed prompt that is as clear as possible. What do you want the response to include, and what should it leave out? How should it approach the prompt – are you looking for different viewpoints on the topic or the expert consensus?

Ask more than once, with different prompts: changing your prompt to get a better response has as big an effect on accuracy as switching to a more advanced AI model. Make sure your prompt isn’t giving it clues about what answer you want, and ask it to tell you how confident it is about its answers.

Make sure search functions are turned on if possible. (Some chatbots are able to do retrieval-augmented generation, which means they can also search sources that they weren’t trained on in response to your prompt.) Treat it as a gateway to information, not an information source itself: ask for citations and links, and hover/skim to make sure they match.

Improving your feed

There are two ways to improve what social networks and video sites show you: curating your feed and training the AI.

Curating means finding sources that you know you like and can count on. To get reliable information, find and follow sources that have knowledge or expertise on the topic, a process for verifying and correcting information, and a motivation to be accurate, and that either aim to be objective or are transparent about their point of view.

Training your algorithm means sending signals that show what you want (and don’t want) to see. Some apps let you start fresh first: for instance, on Instagram you can reset your algorithm by doing this:

• Tap your profile icon (bottom right of the screen)
• Tap the menu icon (the three lines at the top right)
• Choose Content Preferences
• Choose Reset Suggested Content
• Choose Next
• Choose Reset Suggested Content (again)
• Click or tap Reset Suggested Content (again).

Once you’ve done that, make a point of closing or swiping away anything you don’t want to see as soon as possible, and engage with the content you do want to see by Liking it and watching it to the end. 

Critical thinking

Remember that we’re most likely to be fooled by things we want to believe are true – and most likely to doubt real content that we don’t want to believe. Remind yourself to be intellectually humble by asking these questions before you engage with something:

• What do I already think or believe about this?
• Why might I want to believe or disprove this?
• What evidence would make me change my mind?

Sponsored by 

Disclaimer: Meta provides financial support to MediaSmarts. This tip sheet has been developed in collaboration between Meta and MediaSmarts. MediaSmarts does not endorse any commercial entity, product or service. No endorsement is implied.