I’ve recently become enthralled by a Facebook group called “People who think AI-generated photos are real.” It’s exactly what it says on the tin: images of Facebook posts attempting to pass very obviously AI-generated images off as genuine photography. Sure, sometimes an AI-generated “photo” might seem plausibly real, but take a look at some whoppers the group has shown me:

Exactly none of these images pass even a first-glance check. Whether it’s due to anatomical screw-ups, unrealistic textures, or simply being too impossible to believe, these images are very clearly created by an AI. These casually-presented abominations are representative of “hallucinations” or AI mistakes. A hallucination in AI is something the AI claims to be true, but is quite self-evidently not. Many different factors go into this phenomenon, but I find it helpful as a way of understanding the general limits of AI. When it comes to researching and accessing data on the internet, AI is fantastic. It’s really good at searching through the data it has and presenting an image or text that summarizes what it found. The issue is that AI as it stands now is not capable of critical thinking and is unable to understand how different things relate to each other or any of the underlying connections between concepts. This is why if you ask an AI image generator to create a cow, you might get a cow with two heads, a tail that becomes a fifth leg, and extra nostrils; AI is able to understand all the traits that a cow can have, but is unable to understand why a cow with four legs is correct and a cow with five legs is not.

Given how obviously off these images are, you might think the denizens of Facebook would laugh them out, but no! The comments on many of these images would have you believe they’re quite popular. Take, for example, this image of an American veteran on his birthday:

There are several tells here: the seam between the leg and prosthetic, the scattered flags on the ground, the switchboards, the ear, they all give this away as AI. It’s hard to look at it without feeling there’s something uncanny happening. And yet the comments section obliges:

I felt stymied. How in the hell could anyone look at something like this and think it’s real? One possible reason I came up with is confirmation bias, or some demented cousin thereof. After looking through post after post, patterns began to emerge, and I could pretty reliably sort the images into different buckets together. There are art posts where we’re shown someone posing with something they made, very commonly a cake, a sculpture (usually of wood or sand), or clothing. These types are posted with a message to the tune of “they made this, like and comment to tell them how good it is!” In a way, these posts provide an illusion that commenting on the post is a net-good, but it’s all simulation. It might feel good to drop a “great job” in the comments section, but neither the cake nor the person is real.

Then there are the birthday posts in which typically one of two things are shown: a pitiable figure holding up a sign or an elder with a cake. The sign-holders can run the gamut from simply weird to outright problematic, but the clear goal is to instill a sense of sympathy. The cake elders are typically played for aww-cute value and there’s a weirdly high incidence of the cake having “peach cream and filling.” In both cases, viewers are asked to express sympathy or well-wishes to a figure implied to be underappreciated or unduly distressed. By providing the illusion that some downtrodden person could be lifted up by the nice comments someone left on Facebook, these posts flatter viewers with the idea that they’ve done a good thing for someone in need. Cheap and easy validation.

There’s also the “why don’t posts like these trend” posts and these are probably my favorite kind because they’re so overt in their intentions. These posts almost always involve some depiction of Jesus Christ or a giant Bible. Sometimes, these combine with birthday posts to show a skeletal Christ in a hospital bed clutching a sign. Especially when paired with a question like, “Why don’t posts like these trend,” the suggestion is clear: our godless world has strayed too far from the light, and if we want to proclaim or assert our goodness, we’ll like this post on Facebook. This type of post also has a deeply bizarre tendency to feature uniformed flight crews, planes, and sharks. The confirmation bias burns the strongest here because there’s no way someone takes these images seriously unless they want to on some level. In this explanation, these images confirm something for enough people that they get on board:

So confirmation bias is one possible reason actual humans are commenting on blatantly janky AI images as if they’re real, but what if they’re not humans? Bots are known to be a presence on our cursed internet after all. Many (many many many) comments seem very same-y like so:

Sure, some of those comments could be made by people using Facebook’s auto-comment function in which the platform populates a comment with pre-rendered text, but could it really be all of them? Confirmation bias alone can’t explain how this many people are consistently fooled by aggressively bad AI, so what gives? This is where the Dead Internet theory comes in. Petr wrote about this a while ago and it feels very relevant here. In short, the Dead Internet theory posits that at some point, the presence of AI/bots on the Internet outgrew the human presence and that the Internet that remains is largely devoid of human activity. Instead, vast oceans of bot-generated content and bot-generated comments surround pockets of human communities which means that in many cases, the traffic we see on the internet is mostly bots talking to other bots. So bots could be posting AI images so that other bots can comment on them and drive engagement.

The typical column of hashtags would support this. No matter what the image is, the hashtags seem chosen at random from a wide range of celebrity names. This would boost traffic to those posts and make them more likely to trend. Given how unrelated the hashtags are to the post, it’s plausible it’s all bot nonsense:

Given that critical thinking and media literacy are atrophied muscles these days, I’m inclined to believe that it’s some mix of a dead internet and confirmation bias that birthed this phenomenon. Bots are likely responsible for a lot of the comments these posts get, but there are seemingly very real people in the mix, too. It’s possible those people know on some level the image is false, but the meaning they projected onto it compelled them. I think it’s notable too that many of these images speak to preconceived notions, often vaguely conservative ones. The DI theory posits that the ultimate goal of all this bot activity is to manipulate people and their perceptions of reality, so perhaps this is meant to give the false impression of community. Or perhaps the aim is to flood as many AI images as possible onto the internet under the guise of “photography” to gradually degrade our ability to tell a real photo from a fake one. Or perhaps all this AI drek is simply the product of the world we created, where we automate art so that humans may labor. It could very well be all of the above.

To the question of whether we are dumb or the internet is dead, I say: yes.