🧠AI Foundations
Your progress0%
0 of 49 lessons
Reading10 min·Lesson 3 of 5

Misinformation and Deepfakes

Misinformation and deepfakes are not just problems in distant countries — they are actively shaping elections, business decisions, and community trust right here in Kenya, and AI is making them easier to produce than ever before.

Understanding the Spectrum: Mistakes vs. Manipulation

There is an important distinction to understand before we go further. Misinformation is false or inaccurate information spread without necessarily intending to deceive — like sharing a WhatsApp forward you genuinely believed was true. Disinformation is deliberate: someone creates and spreads false content knowing it is false, usually to manipulate opinion, harm a person or organisation, or cause confusion.

AI tools are now used in both. Sometimes AI produces misinformation accidentally (the hallucinations we discussed in Lesson 2). But AI is also increasingly used to create disinformation at scale — fake news articles, fabricated quotes from politicians, and synthetic media including deepfakes.

🇰🇪
Kenya context: During election seasons, WhatsApp groups, Facebook, and TikTok in Kenya have been flooded with fabricated screenshots of politicians' statements, manipulated videos of speeches, and fake 'government announcements' about subsidies or fees. These spread fast in communities where people trust messages that come from people they know. Knowing how to spot them protects you and your community.

What Is a Deepfake?

A deepfake is a video, audio recording, or image where AI has been used to make it appear that a real person said or did something they never said or did. The technology works by training an AI model on many real images or recordings of a person, then generating new synthetic content that mimics their appearance and voice convincingly.

Early deepfakes were easy to spot — blurry edges, unnatural blinking, distorted backgrounds. Modern deepfakes, especially video, are far more convincing. Audio deepfakes (sometimes called voice clones) are particularly dangerous because they can be convincing even to people who know the person well.

🎭
Real risk scenario: Imagine receiving a voice note — seemingly from your bank's customer service — telling you that your M-Pesa or KCB account has been flagged and asking you to confirm your PIN. A voice clone AI can replicate a person's voice from as little as 30 seconds of audio. This is already being used in financial fraud globally, and Kenyan consumers are not immune.

How to Spot Deepfakes and Fabricated Content

  • Check the context, not just the content. A video of a politician saying something shocking — does it appear on their official page, or only in a WhatsApp forward? Official accounts are harder to fake than viral clips.
  • Look for unnatural details in video. Watch the edges of hair, the movement of the neck, the blinking pattern, and how the mouth forms words. AI video generation still struggles with fine details, especially at the edges of faces.
  • Listen for audio artefacts. AI-cloned voices can sound slightly flat, with unnatural pauses or odd emphasis. Unfamiliar background noise cuts in and out unexpectedly.
  • Use detection tools. Tools like Hive Moderation, Sensity AI, and Google's About This Image can help identify AI-generated content — though none are perfect.
  • Verify through official channels. If a 'government announcement' or 'statement from Safaricom' is circulating, check safaricom.co.ke, the relevant ministry website, or the official verified social media account before believing or sharing it.

AI-Generated Text: Harder Than You Think

Fabricated written content is also a growing concern. AI can produce news articles, official-looking press releases, and social media posts that are entirely false. Tools exist to detect AI-generated text — GPTZero, Copyleaks, and others — but they are imperfect. The more reliable approach is the source-checking habit from Lesson 2: if a claim is extraordinary, the evidence needs to be proportionally strong.

🛡️
Your role in the information ecosystem: Every time you verify before sharing — whether it is a WhatsApp message about fuel prices, a viral quote from a government official, or a shocking video — you are protecting your family, your community, and your professional reputation. In Swahili: Thibitisha kabla ya kushiriki. Verify before you share.

What to Do When You Encounter Suspected Disinformation

  • Do not share it — even to 'warn' others, sharing spreads the content.
  • Report it — WhatsApp, Facebook, TikTok, and Twitter/X all have reporting functions for false information.
  • Refer to Africa Check — africacheck.org has a submission form for fact-check requests and publishes verified debunks of viral Kenyan misinformation.
  • Tell the person who sent it — politely, with the correct source. Most people share misinformation because they did not know it was false, not because they intended harm.

Now that you understand the theory and the risks, the next lesson is hands-on: you will practice fact-checking a real AI response using the methods from this unit.