The solution to deepfakes... is to make more deepfakes.

As we all know now, a certain generation tends to believe everything they see on the internet. I was thinking about this and the reasons behind it, and the most logical explanation I was able to find was that they grew up in a period where editorial content was the default. So, if they read a claim in a newspaper, a magazine, or a book, it was more likely to be correct than incorrect - as long as it was not written in bad faith, and while some topics like politics and so on were more open to ill intent, there were no incentives to consciously lie about most topics.

This was due to one main reason: Access to information was limited, but access to generating information was even more limited. So, the sources were more reliable. They had to be, since it was not possible for a high school student to publish a book. Obviously today, that’s not the case. The internet has democratized the knowledge, but it also democratized the production of knowledge, so come your internet trolls.

Internet literacy is a new skill

The generations that grew up with the internet developed a skill that the previous generations did not: assuming anything they read is fake until proven otherwise. This assumption was the opposite for the previous generations; real until proven to be fake.

OK, so we learned that online texts are rarely realiable, but there were always some cues that gave them away like bad grammar, poor writing skills, too-good-to-be-true stories, or too-crazy-to-be-true etc. The trained eyes were mostly able to tell one from the other. With the LLMs, though, that has changed, too. Now, any high school kid can generate convincing-looking text in the matter of seconds, and spread lies. So, the cues can’t be taken for granted anymore.

So, takeaway 1: We should adapt to that. We can’t expect fake text to be poor text anymore.

New formats

Well, no matter how seasoned you are on the internet, if you see a video of something, you believe it, right? Right? Not anymore.

Now it is quite easy to generate convincingly-real images, and it is getting easier and easier to generate videos. Probably we’re less than 2 years away from generating passing videos. Worst part is that it’s also possible to generate real people’s faces. Imagine the damage one can do. How do we stop it?

My suggestion, or rather my prediction, is that we don’t. So far, we haven’t really managed to stop any illegal content online as long as there’s demand for it. I don’t expect this to be different, either.

The solution is that people will lose their trust in what they see online. Similar to how we do now with what we read on Twitter. At one point, all images and videos on the internet will be fake until proven to be real. Until then, there’ll be some collateral damage.

Takeaway 2: Just like text, we can’t assume images and videos to be real.

I don’t know if you see it but there’s one big problem left here. Ok, let’s say, people faking other people’s videos, voices etc. is not a problem since everyone has learnt not to trust those, but about the real content? What about the videos that are real, and the videos that the society will benefit from knowing that it’s real?

Takeaway 3: This does not mean we will not be doomed in a post-truth hell.

Also, uhm, has the majority of people actually learned to distrust the text? Even in younger generations?

Takeaway 4: The first half of this article is too optimistic, we’re doomed already.

If you’re considering making deepfakes to speed up this process, don’t. It’s a figure of speech. 👀✍