Jargon:Deepfake
1 min read
Say this at the party
“A deepfake is a fake video, photo or voice made by AI to look and sound like a real person doing or saying something they never did.”
In plain English
AI can learn what someone's face and voice look and sound like, then make new clips of them. The result can be very convincing.
Some uses are harmless fun, like putting a friend's face in a film scene. Others are harmful: scams, fake political clips and fake images of people made without their consent.
Spotting them by eye is getting harder. That's why people are working on labels and checks that show where a piece of media came from.
Where it came from
The word comes from a Reddit user called "deepfakes", who shared face-swapped videos in late 2017. It blends "deep learning" (a type of AI) and "fake". It built on AI research such as generative adversarial networks, introduced in 2014.
See it in action
Dave gets a voice note that sounds exactly like his daughter, saying she's lost her phone and needs money urgently. He rings her real number first, and she's fine. The voice was a deepfake.
What people get wrong
- It isn't only video. Fake voices are often easier to make and are common in phone scams.
- A blurry or badly edited clip isn't automatically a deepfake. Plenty of fakes are made with ordinary editing, and some real clips just look odd.
Go deeper
- ArticleWikipediaDeepfakeWhere the word came from, how deepfakes are made, and the harms and laws around them.
- GuideOfcomDeepfake Defences: Mitigating the Harms of Deceptive DeepfakesThe UK regulator's look at how to detect and limit harmful deepfakes.
- PaperGoodfellow et al., 2014Generative Adversarial NetworksThe paper behind one of the early techniques used to make realistic fake images.
Related jargon
Drafted with AI. The AI Council (Claude, ChatGPT and Gemini) gave counsel. Phin signed it off.