Misinformation and deepfakes

Posts that shock, anger, or flatter people can spread quickly because they invite immediate reactions. Virality measures attention, not truth.

You do not need to prove every post false alone. For important claims, look for reliable confirmation and avoid amplifying uncertain material.

A deepfake is not a clever edit of an existing recording; it is media generated by a model that learned to imitate a specific person, and knowing the mechanism explains why detection by eye keeps failing. The underlying technology is generative modelling. For a face swap, a system studies many images of a target face until it can produce that face under new angles, expressions, and lighting, then renders it frame by frame onto source video. For voice cloning, a model trained on large amounts of general speech learns the components of how humans talk, and then needs remarkably little audio of a specific person — sometimes under a minute, easily harvested from voice notes or interviews — to capture their timbre and rhythm and make them say arbitrary sentences. Earlier generations of these tools left telltale artefacts: unnatural blinking, blurred jawlines, mismatched lighting, a metallic edge to the voice. Each artefact existed because the model had failed to learn some pattern, and each new generation of models learned it, which is why every published checklist of 'signs of a deepfake' has aged badly. Work through a realistic example of how such media does harm: a voice clip circulates on WhatsApp in which a well-known Zimbabwean business figure appears to announce that a mobile money service is collapsing and customers should withdraw everything. The voice is familiar, the audio slightly rough in the way voice notes are, and the message arrives forwarded by a relative. Every cue people traditionally use to judge authenticity — recognising the voice, trusting the sender — is satisfied, yet the clip may have been fabricated in minutes from publicly available speeches. The misconception to correct is that careful people can spot fakes by inspecting them closely. Inspection is exactly the game the technology is winning; the reliable questions are external to the media. Where did this first appear, and is the original account really the person's? Has any authoritative source — the company, the regulator, established newsrooms — confirmed it? Would this person plausibly announce this, in this way, through this channel? Provenance and confirmation, not visual detective work, are the durable defences, which is why the pause-and-verify routine in this module matters more as the fakes get better.

Misinformation does not spread because people are stupid; it spreads because it is engineered — sometimes deliberately, sometimes by accident — to fit the way all human minds work, and knowing the specific mechanisms is a better defence than feeling immune. The first mechanism is emotional arousal. Content that triggers anger, fear, or moral outrage gets shared more, a pattern confirmed in large studies of social platforms, because strong emotion narrows attention and creates urgency to act — and sharing is the easiest action available. The second is confirmation bias: claims that fit what we already believe receive a fraction of the scrutiny we apply to claims that challenge us, so a false story attacking a politician you dislike slides in almost unexamined. The third is the illusory truth effect, one of the most robust findings in psychology: mere repetition makes statements feel more true, because familiarity is easy to process and the mind misreads that ease as accuracy. A rumour seen in three different groups feels independently confirmed even if all three copies trace back to one post. The fourth is social proof — a message forwarded by a trusted aunt or a respected teacher borrows their credibility, though they spent no more effort checking it than anyone else. Platform recommendation systems amplify all four, since they are trained to maximise engagement and emotionally charged content engages. Work through an example: during a cholera scare, a message claims that a specific vegetable market is the source and names a vendor. It is frightening (arousal), it fits existing worry about the outbreak (confirmation), it arrives in four groups within a day (repetition), and it comes from known contacts (social proof). Every psychological system says share; the message may still be entirely false, and the named vendor's livelihood is destroyed either way. The misconception to correct is that falling for misinformation reflects low intelligence and therefore education makes one safe. Studies find educated people are sometimes better at constructing justifications for believing what suits them; the vulnerability is human, not intellectual. The practical defence follows from the mechanisms: treat strong emotion as an alarm rather than a verdict, notice when a claim flatters your existing views, remember that seeing something three times is not three sources, and before forwarding, ask the one question that interrupts the whole cascade — how do I actually know this is true?

A video claims that a school will close tomorrow, but it comes from an unfamiliar account. A responsible response is: Check the school's official communication before forwarding it. Correct. The school is the appropriate source for its own closure notice. Use this case to separate what the technology contributes from what people contribute. The team should compare the intended outcome with a baseline where applicable, record important assumptions, and keep a clear route to correct or stop the process.

Ask whether the examples, data, and assumptions fit your school, company, or community. Generic demos often miss local names, laws, connectivity, and languages.

Shocking claims spread in chats and feeds. Pause, verify with the school or trusted news, then decide whether to share.