The DeepFake Video Problem

 

Buzzfeed has created a video that shows a more troubling side of this technology. The video shows former President Barack Obama saying things he never said, and it looks surprisingly believable.

In the video above, Obama is voiced by Jordan Peele, who does a passable impersonation. Having Peele do the voice gets the video more attention, but there are probably voice actors who could do an even better job. Buzzfeed started by pasting Peele’s mouth over top of Obama’s, and then replaced Obama’s jawline to match the mouth movements. Rendering took 56 hours for a minute-long video.

The tool is known as FakeApp, but the videos are usually called “Deepfakes” because that’s the handle used by the original developer on Reddit. You can download the code freely all over the internet, but it’s not easy to set up — you need to configure Nvidia’s CUDA framework to run the FakeApp TensorFlow code, so the app requires a GeForce GPU. The video you want to alter has to be split into individual frames, and you need a large number of high-resolution photos of the face you want to insert. In the case of Obama, there are plenty of photos online that can be used to generate a model.

Currently, new technology on the internet lets anyone make videos of real people appearing to say things they’ve never said. Republicans and Democrats say this deceitful technology  will become the latest weapon in disinformation wars against the United States and other Western democracies. This technology uses facial mapping and artificial intelligence to produce videos that appear so genuine it’s hard to spot the phonies. Lawmakers and intelligence officials worry that the bogus videos — called deepfakes that could be used to threaten national security or interfere in elections.

When an average person can create a realistic fake video of the president saying anything they want, and the reverse is a concern, too. People may dismiss as fake genuine footage, say of a real atrocity, to score political points.

Realizing the implications of the technology, the U.S. Defense Advanced Research Projects Agency is already two years into a four-year program to develop technologies that can detect fake images and videos. Right now, it takes extensive analysis to identify phony videos. It’s unclear if new ways to authenticate images or detect fakes will keep pace with deepfake technology.

Deepfakes are so named because they utilize deep learning, a form of artificial intelligence. They are made by feeding a computer an algorithm, or set of instructions, lots of images and audio of a certain person. The computer program learns how to mimic the person’s facial expressions, mannerisms, voice and inflections. If you have enough video and audio of someone, you can combine a fake video of the person with a fake audio and get them to say anything you want.

Deepfake technology still has a few flaws. For instance, people’s blinking in fake videos may appear unnatural. But the technology is improving.

1. Don’t jump to conclusions

2. Consider the source

3. Check where else it is (and isn’t) online

4. Inspect the mouth

5. Slow it down