Name the forms AI-made misinformation takes and explain why it spreads.
Hafiz is in eleven WhatsApp groups: family, old classmates, his condo, his badminton kakis, and the parents from his son's school (he and his son are an invented example). On a typical evening at least one of them shares something alarming, such as a flood video said to be in Bedok, a screenshot of a "new MOH advisory", or a long message about a supermarket apparently closing all its branches next month. Most of it arrives with the same line: "Not sure if true, but just sharing." That line is how a lot of false information travels.
A convincing false story used to take effort. Someone had to write it, find or fake a photo, and build a website that looked real. That effort limited how much false content existed.
AI tools have removed most of that effort. A language model can write a plausible news article in any style and any language in seconds. An image generator can produce a photo of an event that never happened. A whole fake news site, with dozens of articles, author names and a professional design, can be put together quickly and cheaply. The same message can be rewritten a hundred ways so it does not look copied, then translated into Malay, Chinese and Tamil for different audiences.
AI does not make each fake better than a human could. It makes fakes plentiful. When false content is cheap there is more of it, and more of it ends up in your feeds and chat groups, including the eleven Hafiz reads.
Say Hafiz asks an AI assistant whether the supermarket is really closing, and it replies with a quote from "a company spokesperson speaking to The Straits Times". He still knows nothing until he finds that article himself, and often it does not exist.
AI assistants sometimes invent things: quotes nobody said, studies that were never published, statistics with no source, and links that go nowhere or to unrelated pages. They state these in the same fluent, confident tone they use for true statements, and this catches careful people too. A language model predicts plausible text, and a plausible citation looks like a real one. Lesson 6.1 of AI fundamentals: what it is, how it works, where it fails, "Why a fluent answer can still be false", explains this in more detail.
Treat any citation from an assistant as a lead to follow up. Find the original source yourself, and only rely on it once you have, expecting that it may not exist. The same habit applies at work as well as at home. A report that cites a study an assistant invented can embarrass you in front of a client.
False content also spreads because of how it makes people feel, and not only because there is a lot of it. Posts that make people angry, frightened or outraged get shared more than calm ones. The flood video stirs worry for relatives in Bedok. A fake health advisory makes parents fear for their children. The feeling arrives before any thought about whether the post is true, and the share button is right there.
People who create false content know this. With AI they can test many versions of a post to see which one provokes the strongest reaction.
A strong reaction, whether anger or fear, is your cue to slow down and check before forwarding. Hafiz now has a simple habit for this:
Notice the reaction: his stomach tightens, or he feels angry or afraid. Wait. Check the claim before sharing it. If he cannot check it, he does not share it.
Singapore's law on online falsehoods is the Protection from Online Falsehoods and Manipulation Act, usually called POFMA. Under it, the government can direct that a correction be placed alongside a false statement of fact, or in some cases that the statement be taken down, where doing so is in the public interest. You may have seen one of these correction notices attached to a post or article.
POFMA is a tool for specific cases the government decides to act on, and most false posts never reach that point. The supermarket rumour in Hafiz's condo group will almost certainly never get a correction notice, and neither will the misleading flood video or an invented quote in a family chat. Checking those falls to the people reading them.
Checking needs no special tools or training. Lessons 6.2 and 6.3 each give you a method that takes a few minutes.
Before then, go back to your own feeds or chat groups. Somewhere in the last week there's probably a claim that spread widely, and it's worth asking where it actually came from.
Find one claim that went viral recently in your feeds or chat groups. Note who posted it first and what source they gave.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).