There is nothing that can persuade someone that something has happened quite like a video recording. After all, the camera never lies – right? Well, what if this was wrong and what if we couldn’t always trust our eyes and ears?
What are deepfakes?
Deepfakes are visual and/or audio content manipulated using artificial intelligence to change how someone, or something, is presented. Think about it as a much more powerful Photoshop applied to photos, videos or voice recordings.
While deepfake technology provides you and your friends with the ability to swap faces using apps such as Face Swap or Snapchat, it can also present more sinister challenges. For instance, the tech has the potential to produce a video of a politician announcing a new policy or supporting a controversial candidate, or even a recording of your boss telling you to “get on with the deal“. Will you be able to tell if it is real or fake?
Why should I care? Won’t I spot a deepfake anyway?
In 2019, a criminal group used deepfake technology to impersonate the voice of a German CEO who had asked another member of the management to urgently transfer funds to a Hungarian-based supplier. Once the money was transferred, it was quickly moved to a different country and distributed to other locations. The executive in question became suspicious only after they attackers made further payment requests.
With the tech getting better, it will also get harder to spot the somewhat robotic voice in calls, or slight blurring or unnatural blinking in videos. What if it’s just the poor internet connection we all sometimes experience when working from home? As our careers develop, we should keep in mind that what we see and hear may not be always true and that we might need to do a bit more work to check the accuracy of what we are seeing and hearing.
So, am I going to be deepfaked?
Unless you are a YouTuber or a young politician (hey, NextGen is open to all!), it is unlikely that the amount of publicly available audiovisual content (think your social media) will be sufficient to create a persuasive deepfake of you at this moment in time. But technology gets better and with it the ability to create convincing material with less resources.
As the deepfake technology improves, so does technology to detect deepfakes. For example, the FaceForensics++ team has developed a tool which can spot manipulated facial images considerably better than any human. However, any improvement in detection capabilities also means that someone will try to build a better deepfake software which is even harder to detect.
While politicians have been discussing introducing new laws to prevent the spread of audiovisual disinformation, the legislation has not caught up yet and it is currently up to companies and individuals to be vigilant and train themselves to recognise deepfakes.
Want to know more?
- Test yourself if you can tell which face is real.
- Read this paper on deepfakes and audiovisual disinformation.
- Take this free course on the basics of AI.
And remember, next time you watch the video of your favourite celebrity or get a call from the CEO of your organisation, have a think about whether you are 100% sure that it is really them or if you need to go one step further.
Written by Ondrej Hajda, Reece Randall and Valerie Vanryckeghem, Intellectual Property Associates at Mayer Brown for NextGen