Caitríona Redmond: Can we trust what we see when buying online?
How do you spot AI manipulation? Firstly, if it looks too good to be true, be suspicious
I can't say there was a time when, broadly speaking, you could always trust the photograph of what you were buying online.
Anyone who has ever ordered a burger in a restaurant knows the picture on the menu isn’t necessarily what arrives in the little paper wrapper. Food photography has been stylised for years. Those perfect burgers, glossy sauces, and suspiciously symmetrical chips are designed to make the food look as appealing as possible.
Advertising, more generally, has always been glossy, polished, and sometimes detached from reality.
The difference now is the sheer scale and accessibility of the technology being used to create those images.
Before the advent of AI, companies and individuals paid photographers and designers large sums to edit and curate photographs. AI has, technically speaking, democratised that process. Suddenly almost anyone can create or alter an image at very little cost.
That doesn’t necessarily make the system better.
The bigger issue is when the image isn’t just being used to entertain us, but to sell us something, as I found out to my detriment recently. I ordered a dress online and assumed it would look like the photograph. The dress in the image was a beautiful emerald green.
What arrived was frankly more army fatigue in colour. The annoying thing is I had suspected the original photograph might have been AI-generated, but I gave the retailer the benefit of the doubt, because it was a reputable company.
I should have known better, and it was an expensive mistake to make. Of course, I’ll be returning the dress and looking for a refund, for fear that if I wear it outside, I might be asked questions about camouflage and army rations.
It’s not just the colour that can be misleading. There have been numerous examples of people ordering what looks like a beautifully knitted garment online, only to discover they’ve actually bought a piece of ordinary cotton fabric with a picture of knitting printed on it.
At that point, we’re no longer talking about a photograph making something look slightly better than it really is. We’re talking about if the photograph is fundamentally representing the product.
The EU is moving towards greater transparency around AI-generated and manipulated content, including obligations to identify certain AI-generated or manipulated material.
Social media platforms are doing something similar and have introduced labels identifying content generated or altered using AI.
Some platforms have put more effective measures in place than others. LinkedIn, for example, has taken an interesting approach by providing additional information about the tools used to create or modify an image, rather than simply slapping an AI label on it and leaving you to work out what it means. Instagram recently announced it would label AI creators as well, which is an increasing concern for consumers.
I think the public aren’t paying too much attention to this yet. Think of the ChatGPT in the ’80s trend that occurred in the last several weeks. It went viral and trended over several days; several million images were generated using the tool and shared on social media. Very few of these images had the watermark, but that should change into the future.
So how do you spot AI manipulation? Firstly, if it looks too good to be true, be suspicious. Cheap AI-generated imagery can have some fairly recognisable characteristics. Colours may be unnaturally intense. There can be a strange halo of light around the subject. Skin can almost look impossibly perfect.
When looking at the details, hands and feet are still worth checking. Does everything make sense anatomically? Is the writing in the background properly spelt? Does it actually say anything, or is it a strange collection of letters that look vaguely like language?
People can also have a slightly cartoonish quality, while backgrounds can range from hyper-realistic to something that looks as though it belongs in a computer game.
Context is the most useful test of all. Recently, a number of businesses and organisations shared apparently miraculous photographs of themselves with Donald Trump during his visit to Ireland. But would that actually happen? If the answer is probably not, don’t automatically believe the photograph.
The same applies to product photographs. If a retailer is selling something that looks far better than the price would suggest, then it’s likely that something is wrong.
Online literacy is so important for consumers nowadays, and I hope some of this advice helps you detect whether a beautiful dress or a bargain is the real deal.
Remember, when it comes to AI material online, if it walks like a duck and quacks like a duck, it may also be a giraffe.
Netflix recently announced a price increase, which depending on the service, will cost between €1-2/month more. YouTube Premium has also increased by €2/month for individual users, and €4/month on a family plan. Disney+ announced changes to its subscription model, which will increase advertisements, something that may annoy its streaming customers.
If you’re not using a streaming service regularly, cancel it for now to save money. You can always sign back up later. I recommend stacking subscriptions depending on what you want to watch, and waiting until a full series is available, to watch in a few weeks rather than paying to stretch over several months.
CONNECT WITH US TODAY
Be the first to know the latest news and updates
