Pros and Cons: Deepfake technology and AI avatars

While the negative effects of deepfakes are easy and frightening to imagine, there are also some advantages to using the technology for productive purposes.

Deepfake technology was originally developed out of the work of computer scientist Ian Goodfellow and is a by-product of Goodfellow’s work on generative adversarial networks (GANs). GANs generate convincing images using an algorithm in which two GANs try to ‘fool’ each other into thinking an image is ‘real’. Today, GANs can convincingly recreate a person using just a single video clip of that person (read more about GANs here).

If you want to be scared, or possibly astounded, watch this video from Ctrl Shift Face. It shows comedian Bill Hader on a talk show, telling a story about meeting Tom Cruise and Seth Rogen. Watch carefully – when Hader gives his impressions of Cruise and Rogen, those actors’ faces seamlessly melt into his own. 

At the moment, the technology can only stretch to superimposing one face onto another, but a future is fast approaching when GANs will be able to recreate an entire person – convincingly enough to fool anyone. Yet, while the cons of deepfakes are easy and frightening to imagine, there are also advantages to the technology.

So, what are the pros and cons of deepfake technology?

Deepfakes: Cons

Trust

Let’s start with the easy one. It is not hard to think of reasons to be terrified of a technology that can potentially make anyone appear to be doing or saying anything. Imagine you are watching the nightly news and you see a press conference of the Prime Minister inciting violence. But the whole thing was a deepfake. The PM could deny it, but how do you know that is not a deepfake, too. How can you know what to believe? 

In 2017, researchers at the University of Washington shined a light on the potential pitfalls of generative technology when they released a paper describing how they had created a fake video of President Barack Obama. Google’s chief executive, Mark Zuckerberg, has also been the target of a deepfake video that appeared to show him credit a secretive organisation for the success of the social network.

More recently, a website called Deepnudes was proposed, which would allow users to superimpose anyone’s head onto pornographic content. Although the site’s launch was cancelled, the technology is now out there. 

Scamming

Another area of concern is financial scams. Audio deepfakes have already been used to clone voices and convince people they are talking to someone trusted and defraud them. Earlier this year, scammers used a deepfake of a tech CEOs voice to try and convince an employee at the company to transfer money to the scammer’s account. And this is not the first time: last year, scammers using the exact same trick managed to defraud a company out of $240,000.

While everyone, from DARPA to the FBI on down, are working on ways to identify and defeat deepfakes, or legislate against them, it may already be too late. That’s because the biggest problem with deepfakes is that they corrode our trust in technology – our trust in the news on TV and video, our trust in clips on the Internet, our trust in live-streamed events, and even, our trust in our own eyes.

So, against this seemingly doom-laden scenario, what pros could there possibly be to this technology? Surprisingly, quite a few.

Deepfakes: Pros

Art

Star Wars fans are likely aware of the use of deepfake tech to bring the actor Peter Cushing back to “life” for 2016’s Rogue One: A Star Wars Story, but the technology could also allow a host of other useful arts applications. These include being able to go back and change the dialogue in a video or movie without the need for a reshoot and creating entire videos just by selecting from a menu of presenters and entering the script.

Last year, a UK-based charity used deepfake technology to create a video of David Beckham delivering an anti-malaria message in nine languages. And advertising company WPP created corporate training videos that used AI to create a presenter that could speak the recipient’s language and address them by name. 

Researchers in Moscow have used the technology to bring the Mona Lisa to life, creating a video where she moves her eyes, head and mouth. And although deepfake tech represents a serious challenge to the fight against fake news, it has also been used to create presenter-led news reports tailored to individual viewers.

Professional training

Deepfake technology can be used to create AI avatars for use in training videos. Startups like London-based Synthesia have been getting more attention from the corporate world during the COVID pandemic since lockdowns and health concerns have made video shoots involving real people much more difficult to pull off.

Research

Outside the fields of entertainment and training, deepfake tech can also be used to create personal avatars. These can then be used in apps that allow people to try on clothes or new hairstyles at home, or in training applications in a number of professions.

One of these is medicine, where generative technology is already being used to create “fake” brain scans based on actual patient data. These fake scans are then used to train algorithms to spot tumours in real images.

Identity protection

Perhaps more crucially, AI-generated avatars have been used to protect the identity of interviewees in news reports about the persecution of LGBTQ people in Russia. 

Financial saving

Others point out that generative technology could potentially democratise a number of industries. By allowing the cheap creation of everything from videos, to advertisement and games, generative technology could allow individuals and companies to enter these fields with less investment.

It may be that it is less the technology that is dangerous, than the potential use it is put to. If companies and individuals using deepfake technology can be held to high ethical standards and if harmful uses of the tech can be successfully prevented, then this technology could hold a lot of promise.