The AI issue: how long does the truth have?

This topic has not gone undiscussed at any point in history. Much of philosophy is dedicated to the nature of reality, with pretentious academics claiming that nothing is knowable unless it is perceived by you, personally, because everything else could theoretically be an elaborate ruse. They argue that you can’t know with 100% certainty, for instance, that a tree falling in a forest will make a sound, since you can only extrapolate from personal experience. For thousands of years they were wrong, except in an abstract philosophical sense, but recently it seems it’s becoming literally true. The ease of creation and prevalence of fake information combined with

AI video is rapidly becoming more realistic. OpenAI’s latest AI video model Sora 2 and Google’s latest model VEO 3 were released just over a month ago, are leagues ahead of previous AI videos. The tells for older AI video like Sora 1 and were easy, since the primitive videos changed so rapidly between each frame, but the tells for Sora 2 are much subtler. AI video ‘slop’ is similarly becoming more realistic. This has already borne fruit on short-form video websites such as TikTok, Instagram Reels and especially Google Giggles. All of these platforms are inundated with AI-generated ‘funny’ videos, often through the veneer of a surveillance camera, which AI videos can recreate better than higher resolution cameras. The subjects of these videos are typically animals interacting, like a cat chasing off a bear or a dog picking up a baby racoon. These are so relatively mundane and close to reality on first glance that I didn’t realize they were AI at all until I saw other viral videos exposing them as AI. Since video editing is much more difficult than writing a sentence-long prompt into an AI video generator, the volume of content can increase. Not every video can be individually examined for accuracy, especially when the videos are short and cheap.

This lack of confidence in whether something is AI or not has even seeped directly into campus. Near the end of last month CSB+SJU released an advertisement on their social media titled “Built for what’s next”. If you want to watch it, it is currently at the top of the CSB+SJU Facebook page. It features high-resolution people who are meant to be college students saying platitudes about how picking a college is difficult, before cutting to Bryan Bruess and a montage of CSB+SJU footage. The actors uncannily stare into the camera and awkwardly pause before and after each clip. The advertisement’s comments are filled with people decrying it as AI generated. While the ad is certainly baffling, I don’t think it’s AI generated, since the two actors are consistent across the clips they are used in, something that current AI video struggles to consistently do; but the fact remains that people are unsure of whether it’s real or created by a machine, just because it’s awkwardly edited. Alternatively, it is AI, and I have been completely fooled. As AI video improves and no longer has clear tells as to whether or not it’s real, this will become an even more common debate.

I’m sure that this growth will have diminishing returns at some point, but in the worst-case scenario, AI video will become indistinguishable from actual video. Credentials and proof will become more difficult. Videos could no longer be trusted without clear metadata, maybe even physical film. Even in their current limited capacity, AI video is perfect for scams. YouTube is rife with streams of political figures and prominent businessmen promising free money if you click the link in the description. I am more familiar with this than I would prefer. A few months ago, my dad called everyone in the family into the living room to see the TV. On it was a livestream where Elon Musk was promising to give everyone who follows a link in the description. He only gave it a second thought when the link prompted him to leave personal information such as his address.

The lowering barrier of entry is leading to an erosion of trust and confusion. This is only going to get worse. Eventually, the amount of effort required to create passable content for video, music and possibly even writing will be low enough that it can be automated, destroying social media. To trust anything, then, people will need to see credentials and consistency. Only what is personally experienced, what acquaintances personally experience, and familiar institutions will be able to serve as refuge. Unfortunately, the public is also losing faith in institutions and government, and news coverage is becoming rarer, more centralized and more sensationalized. This is why institutions must be apprehensive about AI use. If they embrace it too much, they risk becoming indistinguishable from the sea of confusion and ‘slop’. In anticipation of this future, we (yes, we) need to prepare. Find information sources you can trust, learn to identify the signs of AI, and most importantly make friends. You’ll need them to hear trees fall in forests.