The creation of deepfakes is a growing problem intended to deceive people. Anyone who keeps up with current events tends to trust video and audio, but with Artificial Intelligence, those mediums get recreated with extreme accuracy. While most deep fakes get used for light-hearted purposes, they can be used with more malicious intent. Today, we want to dive deeper into deepfakes. Below is a report from Good Morning America with some information about the practice.
What Are Deepfakes?
The term ‘deepfake’ first appeared on Reddit back in 2017. The term describes the recreation of someone’s voice or appearance through AI. You might think the process is difficult, but someone with a low-end computer and a few hours can create incredible accurate recreations. The software is only a few years old, but its low entry barrier has made its popularity explode. Hollywood has even caught on and used it to create a younger version of its iconic characters. Similar technology created the young Princess Leia in Star War: Rogue One.
The Process To Create Deepfakes
For a deepfake to be convincing, the AI must learn how to complete its intended task. The AI learns through machine learning or a process of brute force trial and error. For example, if you task an AI to win in chess, it will repeatedly play until it has the best strategy. The person creating the AI will need to provide some information and parameters for when things don’t go as intended.
For the deepfake recreation work, there will need to two things:
- Destination Video: Clear, clean videos work best with the current deepfake software. This is why most you will find will be of politicians or interviews.
- Two Data Sets: For the movements to look accurate, you will need a dataset of the original person and a target dataset. If the original person looks left and nods their head, you will need a picture of the target doing the same movements.
After that, it is a waiting game while the AI creates, then recreates the deepfake. Each time it will learn what worked and what didn’t, and the process will get refined. To further deceive the audience, the deepfake may incorporate audio of the target. The AI also can include lip-syncing technology to ensure the facial movements match the words spoken.
The Dangers Of Deepfakes
As described above, deepfakes have the potential to be a funny joke but also cause some harm. The ability to make anyone say anything is a powerful tool. In India, Hindu nationalists deployed deepfakes to incite violence and discredit female journalists. Deepfakes were used in the US for political sabotage. As they become more popular, they will undoubtedly be used for more malicious purposes.
What To Do
The reality of the situation is as technology advances; the deepfakes will be harder to detect. While movements to stop these videos exist, even AI researchers aren’t convinced it will stop the practice. The critical thing to remember is checking your sources and verifying that a video is legitimate before believing or reacting to its contents.
If you found this article interesting or helpful, check out our other posts!