Deepfakes on the rise

9 de May de 2019 Off By morse

A combination of Big Data and Artificial Intelligence may change the audiovisual industry as we know it. Some say that it may even totally change our perception of what is or what is not real. Others say that it is the democratization of special effects. The fact is that there are businesses already making millions with it. So, what do you really know about deep-fakes and the so-called synthetic media?

WTF is this?

First, some explanations: deep-fake is a technique using deep learning to create extremely realistic media – mainly videos, but they can also be images and audio. The technology has been used to combine movies or to recreate videos from existing images (and movies). The results may range from a playful video that places Elon Musk in the lead role at Interstellar to one that shows a 100% fake speech by Barack Obama – which emulates facial expressions, stance, voice, and pitch of the former president of the United States.

Bad reputation

If you could create any video you wanted, with any actor, actress, or celebrity of the world, what kind of movie would you make? Some would carry out an interview series with Salvador Dali. Others would produce pornography. It is not a mere coincidence that the second group of people was the very one to introduce the technology to the world back in 2017, which helped provide deep-fakes with a bad reputation (a break here to say how much pornography has been disruptive, as we said at Tech Rouge back in 2016).This is so much so that the portion of the market using the technique “for the good” decided to rename it as “synthetic media.”

Reality check

In the “synthetic media” market, there are two sides: those producing it and those detecting it. From the side of those creating it, Synthesia, responsible for that video starred by David Beckham  we mentioned here last week stands out. Some days ago, the startup received an initial investment of US$ 3 million. According to the British company’s head, the solution can increase companies’ capacity to create videos in up to 10 times with just 10% of the cost for a “conventional production.” “A small company can create a new interview-like video on the screen and have it delivered within 48 hours.” From the other side, we have companies like Truepic that use the same algorithms for creating “fake” images to detect whether the published photos are real or not. And there is a lot of people betting on this model, in such a way that the company already has over US$ 10,5 million funding,

Strike a pose

Photos and sounds are also products of this “synthetic media.” A company took the idea from the site “This Person Does Not Exist” (which gathers photos of AI-generated faces) to the fashion world and created a “synthetic model agency.” That’s right, the models shown by DataGrid’s “book” were made by using machine learning. In turn, another startup, Modulate, recreates voices from existing data; in other words, it can make you say anything. Last month, the startup raised US$ 2 million in seed funding!  Still in this audio world, it is worth mentioning Project Revoice” by Lybebird, which is recording people with amyotrophic lateral sclerosis, so that they can continue speaking even after having lost their voice due to the disease.

Nothing is real, and nothing to get hung about…What the startups show is that, between fighting against and thoughtlessly reproducing the technology, there is a middle way: coexisting with deep-fakes. Maybe, this is the best way-out when it comes to innovation!