Note: I’m not claiming that anything I’m writing here is novel. But I think I have to let it out somewhere.

I have been catching up on a work-internal podcast. It’s an AI adoption podcast, so obviously it is quite enthusiastic about AI uptake, which is fine. That is the genre and purpose of the show, and I am not suggesting they need to be critical in the way I might be. But there was a specific point I wanted to extract because it was included as a headline takeaway in their summary: the current scepticism towards AI-generated content.

In the episode, that scepticism was likened to historical attitudes towards photography. The speaker passed on a point that their partner had made, that photography is not objective reality because the photographer chooses the framing, the context, has control over the lighting, and can retouch afterwards.

I agree with the most charitable reading of this: when AI is used to accelerate or augment what human editing could already do (retouching, cropping, even to an extreme), the two are similar in kind. But the conclusion that they are fundamentally not that different because both can be untrustworthy is an argument worth pushing back against strongly, especially when AI-generated content is presented as depicting reality.

The human photographer certainly frames the shot: they choose the angle, the crop, the precise moment to capture, and they influence which photos make it to print. That process can mislead, and I am not claiming photos are neutral. But the event actually happened. Light reflected off physical objects that exist in the world. If those objects or that scene had been different, the photo itself would have been different, regardless of what the photographer wanted. That is where a photo’s value as evidence of reality comes from, and it is what makes the photographer’s choices meaningful. What would be the point of taking photos of your child’s first day of school if asking an AI to generate the scene were the same thing?

An AI image does not frame an event because there is no event. Compared to the real world, AI-generated scenes are fiction, or at best reconstructions; they are not factual records. There is a spectrum, of course: heavily retouched advertising material creeps closer to visual fiction, but even that retains a baseline level of trustworthiness (for instance, did someone model this clothing?). It still contains a kernel of truth. (There’s also a further argument to be made that genre expectations play a role here - we can know that advertising material is intended to serve a purpose, which might include deception.)

While it is technically true that neither AI nor photography represents pure objective truth, that is not a meaningful statement when paintings, novels, and ads also fall into that category. In fairness, the podcast speakers noted that we should not compare a perfect past to an imperfect AI-generated future, which is reasonable. But this discussion is not about whether historical photos were ever 100% reliable or objective; it is about whether we should treat AI-generated content as the same kind of thing as what it mimics. Photos can be edited to mislead, making them a flawed record of something that happened, but a substantially generated AI image is not a record of anything.

There was another argument that attempted to sidestep the issue by asking whether the comparison even matters, framing it as simply choosing whether to trust an AI image versus one produced by a person. I think that is a distraction. Untrustworthy images have always existed, but conjuring them out of whole cloth previously required substantial skill. While that barrier has been lowering for years, it used to be difficult; now, generating convincing imagery is in everyone’s pocket for the mere cost of internet access.

The comparison matters because widespread distrust does not just affect fake media; it degrades trust in things that should be trusted. Why believe news photos, holiday snaps, wedding pictures, or photos of your own children when any of them could easily be AI? It’s the liar’s dividend: today, anyone can argue that any photographic evidence of them has been fabricated. The issue is not just whether something was created by AI, but how effortlessly it can be made to look real. In that sense, images have caught up to text: provenance now matters above all else, and the person providing the image becomes the primary test of its credibility.

But even that is not very satisfying, because images carry their credibility with them in a way that text usually doesn’t, or at least in a way that text is less subject to - text is usually read as a claim by someone, and people want to know who. On the other hand, attaching any photo to a claim makes people more likely to believe it, even when the photo provides no evidence for the claim at all (Newman et al., 2012)

What do I mean by this? Images get copied and pasted, and they get forwarded in chats. Until now, if those images appeared credible, they have been assumed true.

We are going to need to grapple with and adapt to this new world where sources matter more and scrutiny matters more. Whether or not you can pay the mental tax of scrutinising every piece of media you receive will determine how likely you are to be duped by it.

Overall, I think this change makes the world a little bit worse, and I wish it didn’t.