Another day, another Instagram/Facebook reel that looks too good to be true. How can I tell if it is AI? asks a comment.

A good question. AI videos have only become popular in the last two or three years and are already nearly impossible to distinguish from real ones.

I had a few minutes so I thought I would contribute what I’ve learnt. And since I have a few minutes more, I thought I would add them here in case they were helpful for you.

1. Check the relevant account’s description, hashtags, and bio. Ethical users of creative AI videos mention there that the video is AI generated
2. Listen to the voices in the video. If they sound metallic, they are probably AI generated
3. Look at the texture of the video, especially the people and animals. I don’t know if there is any AI platform that can perfectly replicate human or animal bodies yet but there are definitely some that can’t. As a result they make them look too smooth
4. When you watch a video ask yourself if the action you are seeing is realistic. For example, do cats/dogs/etc behave in this way (or are they likely to)? If it seems unlikely that they would then the video may be AI
5. If the video is telling a news worthy story, google it. If you are unable to find it then that is an indication that the video is fake. For example, the video I saw today that inspired my comment and this post showed a woman running into the sea as a dolphin swam excitedly towards her. The headline was “A dolphin meets the woman saved him from fishing nets some time ago”. I googled “woman meets dolphin she rescued” and only found the same video elsewhere on Instagram and Facebook. A feel good story like this would surely attract attention from the media – if it were real
6. Finally, take a screenshot of the video and give it to one of the AI detector websites and see what it says. For the dolphin video, I used sightengine.com and it came back with a 92% chance of it being AI.

I’m not an AI expert. Indeed, my tech skills may generously be described as limited. The above is simply what I’ve learned over the last couple years. Hopefully, it is helpful.

In an ideal world (so it probably won’t happen) AI generation companies would be obliged to ensure that all their content is marked as such. Unmarked AI content is the most pernicious thing I can think of that has come to the internet. In its fakeness, it deceives and in doing so it becomes a tool to gain power and control over the user: a tool for bad actors, individuals, groups, and – here is the big problem – states. For them it is a weapon that can cause very great harm.

***

As a coda to my list, I will add that you really need to be on your guard with online videos now. It was only in writing this post that I realised that the description attached to the dolphin video mentioned above contains a paragraph on how replies to the video (from wherever it was sourced) contained scepticism over whether it was real or not. The paragraph didn’t say whether the video was real or fake, which it should have, but at least it referred to the possibility. When we scroll, we like to drift from one video to the next to the next without thinking about the veracity or otherwise of what we are seeing. Those days need to end.