Home » Deepfake alarm, so artificial intelligence will help us (perhaps) to fight the fakes created by artificial intelligence

Deepfake alarm, so artificial intelligence will help us (perhaps) to fight the fakes created by artificial intelligence

by admin

The last example is the most harmless one, but one that has created the most stir: a fake Tom Cruise, practically identical to the original, joking and laughing in a series of widely shared videos on TikTok. So shared that they spark a wide debate online and that their creator temporarily remove them from the social network, a practical demonstration of the level reached by the so-called deepfakes, those fakes (images, videos, audio) realized thanks to the use of algorithms based on artificial intelligence, which starting from one face is able to simulate another, also recreating facial expressions and expressions.

Those clips were (are) made for fun and in fact on TikTok they have returned and remained, but the fear of many is that they are an appetizer of what awaits us in the future, when deepfakes will be used to imitate in an incredible (indeed, very credible) way a politician, a public figure, a neighbor, our daughter’s teacher. And make her say anything. How will people understand what is true and what is not? How will the reporters do it? With the help of technology, of course.

The IA used against the IA
Some researchers from the University of Buffalo, in the United States, they found a way to distinguish human faces from computer-generated ones by analyzing the reflection in the eyes. In document compiled by scientists (which is this, in pdf) it is remembered that the cornea works a bit like a mirror and reflects the light that is in front of it: in the case of humans, what is seen reflected in the two eyes it is almost the same, because they have the same objects and the same light sources in front of them; in the case of deepfakes this doesn’t happen, or in any case it almost never happens, because (simplifying) artificial faces (like this one) they are created from databases of faces that are combined together to achieve the desired result and the eyes can even come from two different faces.

See also  "Neil" team's new work "Voice of Cards Dragon Island" first revealed | 4Gamers

To find the fakes, the researchers used a software that has learned to recognize humans after studying tens of thousands of eyes and their reflections, that is, an artificial intelligence to counter another artificial intelligence. And the results are quite satisfactory, as they are close to 95% reliability.

One of the fake Tom Cruise videos on TikTok

deeptomcruise

My 🐢 impression…

♬ original sound – Tom

A battle that has just begun
With some contraindications, highlighted by the researchers themselves: the system works (very) well if in front of the face there is a light source clear and evident enough to generate a reflection on the corneas and above all if both eyes are visible, so that the AI ​​can compare them; moreover, a subsequent post-production work on the “fake” could also intervene at the level of these details, so as to harmonize the reflections on the right eye and the left eye.

Over time, however, these countermeasures are likely to become even more effective (in helping us), which they will do even deepfakes (in deceiving us). In short, it is only the beginning of the umpteenth battle between good and bad … only this time it concerns the machines.

You may also like

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More

Privacy & Cookies Policy