Deepfakes
Deepfakes are video, images, or audio that look and sound real – but aren’t.
They’re made by artificial intelligence (AI), computer systems that learn patterns from huge amounts of data in order to create new content. Type the video you want to see and the system will generate it for you.
Deepfakes are made for lots of reasons. TikTok is full of harmless deepfakes that are fun to watch like a giant cat in a supermarket. OpenAI’s new app Sora is like TikTok but full of nothing but AI-generated content. There is nothing real in the app.
Deepfakes can entertain us but they can also be used to deceive and manipulate. During elections, fake videos of politicians have spread false information. Scammers use AI generated voices to trick people into sending money.
Humans have always fooled each other through fake paintings, photos, and lies. What’s new is how easy and convincing these fakes have become.
Big Ben on fire. Stephen Hawking towing a caravan. A cat playing a violin. A monkey delivering a parcel in a tornado. Queen Elizabeth rapping. None of these things happened but videos like this have been watched billions of times online.
How to make a deepfake
The secret behind deepfakes is a system called GAN (Generative Adversarial Network). Ben Garside from the Raspberry Pi Foundation told Science+Nature’s Mysteries of Science podcast that these networks involve two computers criticising each other’s work. One, the Generator, creates the deepfake and then sends it to be checked by a second computer called a Discriminator. The Discriminator decides whether it looks real. If it doesn’t, it sends it back to the Generator to try again. This happens over and over until we’re left with something that both computers believe looks real.
Your brain on autopilot
People are often fooled by deepfakes, and small AI mistakes often go unnoticed.
Why are humans so helpless against them? The human brain is a prediction machine that has been around for 1.7 million years. It loves patterns and is constantly guessing what might happen next, to save time and energy. You anticipate what should happen and fill in missing details. This is useful in everyday life. It helps you recognise people, hear clearly in noisy rooms and react quickly to danger.
Researcher Kimberly Mai told Science+Nature that people “rely on intuition to make decisions” but are often wrong. For most of human history, seeing was believing. Now, deepfakes trick you by looking just real enough.
Perhaps the biggest mystery is what happens when people think they can’t trust what they see. Some believe that when there are lots of fake stories and images, people may start distrusting even real things. Deepfakes might not only be used to spread lies but could also make us question real things.
Spotting a deepfake
Scientists are now using the same technology that creates deepfakes to detect them. Deepfake detectors can suss out what’s real and what might be computer-generated. Kimberly says they are trained to spot things you might miss, eventually learning “the quirks or differences that make fakes sound fake.”
Deepfakes might challenge reality but they also give us a chance to sharpen our critical thinking skills. The technology is neutral – whether people use it for good or bad is up to them – but understanding how fakes are created, why they fool us and how to spot them turns you into a detective.
Also
Deepfakes are convincing but you might be able to spot clues that give them away.
- Start by looking at faces. Strange blinking or eyes that look too glossy are common faults in deepfake videos.
- Listen to the sound. Voices might lack emotion, pauses or breaths. Does the movement of their lips match what they’re saying?
- Look for glitches. Does the lighting change suddenly? Do light and shadows move with objects?
- Question why. Ask yourself how the content makes you feel: most deepfakes are designed to make you angry or surprised.
You shouldn’t automatically trust every video you see but with these top tips, you could become a deepfake detective.
First published in The Week Junior Science+Nature, issue 97, February 2026.