Deepfakes

Photo: Yuezun Li and Siwei Lyu, "Exposing DeepFake Videos By Detecting Face Warping Artifacts." New York State University at Albany. 1 Nov. 2018.

Photo: Yuezun Li and Siwei Lyu, "Exposing DeepFake Videos By Detecting Face Warping Artifacts." New York State University at Albany. 1 Nov. 2018.

This graphic shows the process used when creating a deepfake. The facial movement and major features are detected and put into a transformation matrix. The face shape is then refined and replaces the original.

Photo: Yuezun Li and Siwei Lyu, "Exposing DeepFake Videos By Detecting Face Warping Artifacts." New York State University at Albany. 1 Nov. 2018.

Photo: Yuezun Li and Siwei Lyu, "Exposing DeepFake Videos By Detecting Face Warping Artifacts." New York State University at Albany. 1 Nov. 2018.

This diagram displays how Gaussian blur, a result of deepfake facial alignment, occurs.  Photo A is the original and photo B is the process of facial alignment. Photos C and D show the altered photo with Gaussian blur.  Gaussian blur is a tell-tale sign of an altered photo or video.

For more information about Deepfakes, visit our Deepfake Resources page.

What is a Deepfake?

A deepfake is a human video-audio manipulation that is based on artificial intelligence. Artificial intelligence is trained to detect specific speech patterns in a person and that data is used to make that person say words which they didn’t actually say in real life.

What do they look like?

Deepfakes come in all different shapes and sizes. A deepfake could simply be audio that was generated by an AI and not the person who it is mimicking. It could also be manipulated video footage where faceswap and audio generated by AI are used together.

How have they been used?

The most popular use for deepfakes has been for entertainment. As technology advances though, there is a concern that deepfakes will become more difficult to detect and that they will be used to interfere with elections.

How can you detect a deepfake?

Unfortunately, there is currently no public detection software for deepfakes. Professors at UC Berkeley are currently working on improving deepfake detection technology. Learn more about advances in deepfake detection. 

What's being done?

Fighting deepfakes is difficult because instead of there simply being a story to disprove, what most people consider evidence must be disproven. Normally, photos and videos are what is used to prove or disprove a story, but in this case, the photo, video or audio is what has to be disproven. Lawmakers are beginning to realize that this could pose a threat to democracy. Politically, malicious deepfakes and fake stories can be banned, but to what extent is what politicians are trying to determine. Socially, it is up to everyday citizens to be aware of the existence of deepfakes and educate themselves on signs of manipulated audio and video. While the world works to prevent deepfakes from spreading and causing chaos, tech companies like Adobe are developing deepfake detection technology that the average citizen can use. See more about Adobe’s collaborative work with UC Berkeley here.

 

Dilbert Deepfake Comic

Written by: Madison Latiolais