SALT LAKE CITY – What is a deep fake? By manipulating images, video and audio of a real person through artificial intelligence (AI), a deep fake can show that person doing and saying things they never did nor said.
Deep fakes can also be entertaining.
This Tom Cruise impersonation is very convincing – down to the actor’s laugh.
“You can’t do it by just pressing a button,” said Belgian visual effects expert Chris Ume, who created the Cruise deep fakes. “That’s important, that’s a message I want to tell people.”
These hyper-realistic videos required months of preparation, Ume said, using the open-source DeepFaceLab algorithm and established video editing tools.
Deep fake nostalgia
MyHeritage, a genealogy website, is offering a tool to digitally animate photographs, creating a short, looping video in which people (sometimes famous or infamous) can be seen moving their heads and even smiling.
“Experience your family history like never before!” MyHeritage proclaims on its website.
How do I know if what I’m watching is a Deep Fake?
- Blurring evident in face but not elsewhere in the video.
- A change of skin tone near the edge of the face
- Double chins, double eyebrows or double edges on the face
In 2018, US researchers discovered that deep fake faces don’t blink normally. The algorithms never really learned about blinking so the majority of images show people with their eyes open.
People depicted in deep-fake videos often blink far less often than real humans do. According to Swei Lyu, a professor at the University of Albany, this method has a 95% detection rate.
The AI Foundation has developed Reality Defender, which is a program that runs alongside other online applications, identifying potentially fake media.
Why should I care?
Deep fakes are the most dangerous form of artificial intelligence (AI) crime, according to a report from University College London.
A deep fake algorithm can mimic the accent, intonation and tone of a specific person’s voice.
Criminals are using deep fakes to impersonate CEOs and steal millions from companies, Axios reported in 2019.
- Scammers were mimicking the CEOs’ voices with an AI program that had been trained on their speech from earnings calls, YouTube videos and TED talks.
- Millions of dollars were stolen from each company; names were not revealed. The attacks were first reported in the BBC.
“I don’t think corporate infrastructure is prepared for a world where you can’t trust the voice or video of your colleague anymore,” Henry Ajder of Deeptrace, a deep fakes-detection startup, told Axios.
At the beginning of 2020 and with the US presidential campaign underway, Facebook announced it would ban manipulated videos and photos intended to mislead its users.
The company said it would remove misleading media under the following criteria:
- It has been edited or synthesized – beyond adjustments for clarity or quality – in ways that aren’t apparent to an average person and would likely mislead someone into thinking that a subject of the video said words that they did not actually say.
- It is the product of artificial intelligence or machine learning that merges, replaces or superimposes content onto a video, making it appear to be authentic.
Today’s Top Stories
- Utah Utes let go of Coach K after 10 seasons
- New dress guidelines announced for male missionaries — in some areas
- State parks hope to have new ‘glamping’ campsites available for busy Memorial Day weekend
- Utah’s number of deaths from COVID-19 jumps to 8
- Second Amendment rally at state capitol to push back against new gun laws
- Sergeant Joe Cyr, Salt Lake City Police Department
- Police looking for man who “brandished gun” at Clinton Walmart
- What is coronavirus and Covid-19? An explainer
- “Blue Rally” gathers in downtown SLC to support law enforcement
- Boil water advisory still in effect for a portion of Pleasant Grove