What is a "deepfake?"

A deepfake usually refers to a highly realistic but fake image, video, or audio of a person saying or doing something they never actually said or did.

Of course, deepfakes were created long before generative AI tools became popular —sometimes with Photoshop, and sometimes by changing the caption of an image to make it appear to be from a different context.

But now with generative AI tools making this easier, there is more concern about harms from deepfakes.

Deepfakes have several possible harmful uses.

• Showing a celebrity in a compromising situation.

  • This type of deepfake is often used for defamation or blackmail. You've probably heard about some instances non-consensual deepfake pornography, like the situation with Taylor Swift.

    You can learn more in this article: Celebrity Deepfake Porn Cases Will Be Investigated by Meta Oversight Board.

• Video or audio of politicians giving speeches they never actually made.

  • Manipulated media of politicians giving speeches they never actually made can spread disinformation and sway public opinion.

    Sadiq Khan, the mayor of London, said in February 2024 that faked audio that supposedly captured him making incendiary remarks about Armistice Day almost caused serious disorder after it was widely shared by the far right.

• Impersonating someone in a video or audio call.

  • This can be used to gain unauthorized access to sensitive information. 

    Hong Kong police reported that fraudsters used deepfake technology to trick a finance worker at a multinational company into transferring $25 million. In a video conference call, the worker thought he was speaking with the company's chief financial officer and colleagues, but they were all deepfake recreations. Believing it was a legitimate request, the worker authorized the $25 million transfer.

    Learn more: 

    Finance worker pays out $25 million after video call with deepfake ‘chief financial officer’

Deepfakes can also be entertaining or educational.

• Humor, satire, or cultural commentary

  • Will Smith eating spaghetti – An AI video from 2023, was followed by his good-humored response of a real video of him eating spaghetti). This was followed in 2024 by an AI video of spaghetti eating him.

    Watch the first video.	Watch the second video.

• Bringing a historical figure back to life

  • John F. Kennedy's "Resolution to End the Cold War" speech, (which was never delivered, due to his assassination), was recreated using a synthetic voice imitating his voice and speaking style. See "JFK Unsilenced."
    
    If you're curious about how it was made, watch this two-minute video.
     
    

• Bringing a famous artist back to life for a museum exhibit.

  • The Dali Museum in St. Petersburg, Florida created an exhibit called "Dali Lives." They used deepfake technology to recreate a life-size video of Salvador Dali.

    They trained an AI algorithm on over 6,000 frames of archival footage to capture Dali's facial expressions and synced it with a voice actor mimicking his accent. 

    The AI-generated Dali appears in a kiosk and tells visitors stories, commenting on things like the weather and news in an interactive experience. The goal was to help modern audiences empathize with Dali as a human being and bring his persona into a contemporary context, even letting visitors take selfies with him. 

    The museum got permission from the Dali Foundation to create this novel use of the technology. 

    If you're interested in how it works, watch this video.