First published in 2019 on the Hayley is a Ghost blog
“Pics or it didn’t happen.”
For years, some variation of this has been the almost automatic response when somebody claims to have seen a ghost, Bigfoot, a UFO, or pretty much anything else they weren’t supposed to see.
It makes intuitive sense. Most of us now carry cameras everywhere we go, so if somebody really did see a large hairy humanoid walking through the woods, surely they ought to be able to whip out their phone and finally give us the clear Bigfoot photograph the world has been waiting for.
But suppose somebody did.
Suppose they handed you their phone tomorrow and showed you a beautifully lit, pin-sharp photograph of Bigfoot staring back at the camera.
Would you believe it?
Probably not.
One of the things I learned fairly early while investigating paranormal claims was that demanding photographic proof wasn’t actually getting me very far. A photograph can be useful evidence during an investigation, certainly, but even a completely genuine and unedited photograph only shows us what the camera recorded.
It does not automatically tell us what the thing in the photograph is.
And now generative AI has made an old problem considerably more complicated.
We were faking ghosts long before AI
Fake paranormal photographs are nothing new.
Victorian spirit photographers were producing images of the living apparently accompanied by the dead long before anybody had heard the term “artificial intelligence”. Elsie Wright and Frances Griffiths managed to convince rather a lot of people that fairies lived at the bottom of a garden in Cottingley using paper cut-outs and a camera.
People have manipulated photographs, staged paranormal evidence and created hoaxes for as long as there have been photographs to manipulate, stages on which to put things, and other people willing to look at the results.
AI changes the scale of the problem rather than inventing it.
It is now possible to generate convincing images of things which never happened with relatively little effort. I can ask an image generator to show me Mark Zuckerberg taking a selfie with Bigfoot and receive something which, at first glance, resembles an ordinary photograph.
Which is ridiculous, but also demonstrates the problem rather nicely.
If somebody presents a photograph as evidence of something extraordinary, the fact that the photograph looks convincing can no longer carry very much weight on its own.
Perhaps it never should have.
The photograph was never the whole investigation
The photograph was never the whole investigation and this matters because photographs and videos have sometimes been given far too much authority in paranormal investigation.
Someone takes a photograph in a supposedly haunted building and later notices a pale shape in the background. The photograph gets labelled a “ghost photo”, shared online and discussed as paranormal evidence.
But there is a human decision sitting between the photograph and that conclusion. It is a human that noticed the shape, a human who decided it was unusual, and interpreted it as possibly paranormal.
It is a human who selected the image, cropped it or uploaded it, gave it a caption and told other people what they thought they had photographed.
Even when nothing has been faked, there is still interpretation involved.
At the core of every paranormal claim is a human experience. A photograph or video may document part of that experience, but it is not the experience itself and it certainly isn’t automatically proof of the explanation somebody has placed upon it.
That distinction has always mattered to me when investigating strange claims, but generative AI makes it much harder to ignore.
There is always a human somewhere in the process.
AI creates another problem too
The obvious concern is that people will believe fake photographs.
Research into misinformation1 suggests that people do not always carefully analyse everything they encounter online, particularly when material fits something they already believe or produces a strong emotional response. Wider concerns about AI-generated media have therefore extended well beyond paranormal claims into journalism, politics, public health and national security.
But there is another side to this which interests me just as much - what happens when somebody presents a genuine photograph of something they sincerely could not explain?
“It’s AI” is an extremely convenient answer, and it can be just as intellectually lazy as accepting a photograph because it looks spooky.
If we reach the point where every unusually good paranormal image can simply be dismissed as generated, edited or manipulated without actually investigating it, we have not become better sceptics, we’ve just acquired a new assumption.
The existence of AI does not tell you that a particular photograph was generated by AI any more than the existence of Photoshop proved that every strange photograph produced after 1990 was photoshopped.
You still have to do the work.
So what would I actually do with a ghost photo now?
If somebody sent me an extraordinary photograph tomorrow, I wouldn’t begin by asking whether the hands looked funny or whether the background had that slightly uncanny quality associated with some generated images.
Those things might be worth noticing, but they’re not much of an investigation.
Instead, I’d start with the person.
Who took it? When? Where? Why?
What were they actually photographing at the time?
Did they see the strange feature with their own eyes, or only notice it afterwards?
Was anybody else there?
What happened immediately before and after the photograph was taken?
Can I see the original file rather than a screenshot or a version downloaded from social media?
Has it been edited, cropped or passed through an app?
Has the same image appeared elsewhere?
Those questions were useful before generative AI existed and they remain useful now.
Depending on the material, I might reverse-image search it, examine a video frame-by-frame, compare it with other photographs from the same location, or look for obvious signs of manipulation. But none of those things replaces talking to the person making the claim and establishing the provenance and context of what they’re showing me.
And sometimes the photograph will turn out to be deliberately fake, sometimes it will be an ordinary object photographed in confusing circumstances. Sometimes the person will have genuinely seen something they couldn’t identify and the photograph won’t help very much.
And sometimes we may simply not be able to establish what happened.
That’s allowed too.
Evidence has limits
There is a strange paradox here.
For years, sceptics have demanded photographs and videos of extraordinary things. Now we’re entering a period in which even a spectacular photograph may not persuade us that anything spectacular happened.
But perhaps that’s useful, because paranormal investigation should never have depended on finding the one photograph which finally proves everything.
The Cottingley Fairies didn’t need generative AI, and neither did Victorian spirit photographers. The fundamental problem has always been that images can mislead us, and that the meaning attached to an image depends heavily on the circumstances in which it was produced and interpreted.
AI has just made fabricating the image easier.
It also means that investigators need to become better at asking where evidence came from, what happened around it, what can independently be established, and what the person presenting it actually experienced.
There is another ethical reason to care about this - people really do have experiences they cannot explain. Flooding paranormal spaces with manufactured ghosts, monsters and UFOs makes it harder to take those people seriously and harder to distinguish sincere reports from somebody generating something impressive for attention.
But automatically shouting “AI!” at anything unusual doesn’t help them either.
The photograph was never the whole story, AI just makes that much harder to forget.
The Ghost Geek is an award-winning science blog about spooky things, written by Hayley Stevens, a former ghost hunter who is now studying for a Master’s in Forensic Psychology and uses scientific scepticism to investigate strange claims.
Pennycook, G. et al. (2021) The Psychology of Fake News. Trends in Cognitive Sciences, Vol. 25, No. 5, pp. 388 - 402 Available at https://doi.org/10.1016/j.tics.2021.02.007