Just Eat has apologized after its new AI service designed to auto-generate images of food items got confused and put an image of bacon and eggs swimming in a glass of cola on the menu for a Northern Ireland restaurant. The restaurant, which offers a range of soda breads, ended up with sausages, bacon and fried eggs floating in pints of fizzy drinks instead of the Irish breakfast sandwiches they were supposed to represent.
These AI-generated images were part of a Just Eat trial for a new AI-powered system. The bacon and egg cola wasn’t the only hilarious image that the trial produced. You can see many other examples on the uknewestfoods Instagram page:
https://www.instagram.com/p/DeteNVUiIVE/
Referring to the AI photos, one Just Eat user on X said, “I swear if my salt and pepper chips come like this I’m going to cause murder”
But while these AI-generated disasters are easy to laugh at, the real problem might be the images that look perfectly believable.
It’s the realistic images you need to worry about
Talking to The Independent, a Just Eat spokesperson said, “We’ve been trialling AI-enhanced images and as with any new technology, it isn’t always quite right the first time. We’re sorry for any inaccurate images that have been posted. We’re in the process of removing them and have paused rolling this out while we further develop the tool to ensure it best serves our partners and customers.”
Just Eat isn’t the only food delivery service trialling AI-generated images. Last week DoorDash opened its text-to-order AI agent for testing: you text what you want, it picks the dish and texts back a photo of it. Bloomberg News testers found several errors during testing, including “photos that didn’t match the recommendation”.
The wider question is, why are Just Eat and DoorDash trialling using AI to generate images of food that people are ordering anyway? While obvious mistakes like this are easy to spot, what about the smaller inaccuracies that are easily missed, like a wrong-sized dish or missing topping?
Michael Bortinger, Director of Marketing and Partnerships at Tastewise, explains: “An image model has no concept of a specific dish. It renders the average of everything labeled ‘burger’, and better tools just make that average more convincing. There are pages of brand rules for how a logo sits on a cup and nothing at all for what’s in the burger.”
Is this a good use of AI?
I can understand why using AI could save the time required to take and then upload individual photos of each menu item; however, food images would seem to be a case where accuracy of what you’re ordering is of supreme importance. Outsourcing this job to AI is not only fraught with problems, but surely wouldn’t produce a faithful representation of the food anyway.
There are many good uses for AI; I’m just not sure this is ever going to be one of them. AI-generated food photos might make life easier for delivery platforms and restaurants, but it’s hard to see what benefit they offer customers if they can’t be trusted to show what’s actually being sold.
After all, when I’m ordering a bacon sandwich, I want to see a picture of the item I’m actually going to get, not AI’s interpretation of what a bacon sandwich looks like.
