The current level of development has reached the point that even a schoolchild can take a library with models, for example , train on data taken from public sources and apply it to your data with acceptable quality. Sometimes it can be funny when Jennifer Lawrence's performance is shown with her face . Or, for example, 11 options in a row with different personalities: But at the moment, the techno giants are already concerned about what is happening and there was already an article about it on , that Facebook and Microsoft are providing grants to detect DeepFakes. I don’t really believe in altruistic intentions; I would believe more in ordinary benefit. Most likely, this is a way to find even better strategies for training models to create even more believable DeepFakes. Most likely, the circle will close here, methods will be developed to combat DeepFake and they will also stimulate its development as a technology.
I think it’s no secret to anyone that such technology is an excellent tool for manipulation and provocation. As for politicians and famous personalities, there is a huge field for creating news in the yellow press and more. It is also worth remembering the sensational and quickly collapsed DeepNude project. In general, this was rather a harbinger of the creation of a new technological niche; most likely in some future (present?) it will be possible to see the emergence of products similar to the fact that in an adult video for a certain fee you can see anyone, even those who have distanced themselves as much as possible from this genre.
Let's assume no one needs this. I suggest checking this with an example. Let's check the interest in a well-promoted porn brand against a well-known but not officially featured actress. Take Sasha Gray and Jennifer Lawrence. Here's a comparison both in the last 12 months. A pure actress loses in interest in her person as such an activity, but we can definitely say that there is interest and it is sufficient. And if there is demand, there is a market, all that remains is to figure out how to monetize it.
It’s clear that we’re not just talking about adult content; if the problem is that you can pick up a double of a person, use DeepFake models and create content that no one can distinguish from a video created with a potential victim, then this can already create image and reputational losses for the victim’s personal brand. At the moment, both effort and money can be invested in a person’s image, and considerable ones at that, and this is valued. Therefore, I would not be surprised that regulatory authorities may appear to create content and it will not be possible to post photo/video content on public media platforms that may contain images of public figures that have not passed verification. Perhaps in the future not only public ones, but everyone in a row, because the state of non-publicity is a priori temporary. For example, some kind of “identity verification agency” may appear, and without the signature of such an authority it will be impossible, even technically, to publicly distribute content. This may sound like nonsense, but the social rating system is already , for example, here is a slide with , which illustrates this system, which is going to be implemented in 2020. Against, is already ready to provide information about users’ search history to banks for more accurate credit scoring. In general, it is not clear how this will work, since credit scoring usually requires that the model must be interpretable. Therefore, tree-based models are often used for this task. In the case of Yandex, judging from public information, it will be a black box. On the other hand, if a borrower has received a loan, most often he does not care why he was given a loan, and if he was refused, in this case banks are legally protected by the fact that they have the right not to disclose the reason for the refusal.
It seems to me that against all this background, the emergence of regulatory authorities is inevitable. Perhaps the Chinese will be the pioneers? Why is this inevitable? Because if it is easy to ruin anyone’s reputation, this will accordingly affect the social rating and it seems to me that citizens who have a high social rating and do not want to part with it will take measures to protect their social property.
In this regard, it will be interesting that most likely such systems will not have a statute of limitations and at the time of implementation they will remind everyone of all the skeletons found in the closet. The current situation with social ratings in China seems to be exactly this way; at the time of launch, it may suddenly turn out that you are now not allowed to do something.
I wonder how all this will come back to haunt me as a developer ? Time will tell.
Source: habr.com