Fake food created by AI looks tastier than real dishes, scientists say - so, can you tell which of these are actual meals?

6 months ago 16

Some ultra-processed foods are packed with artificial colours and flavourings. 

But scientists have found that food which is 100 per cent artificial looks even tastier than the real deal.

Researchers from the University of Oxford and the University of Naples Federico II discovered that people rate AI-generated food images are more appetising.

And, as AI images become more realistic, the experts warn that they could even promote unhealthy eating habits.

So, can you separate the real photos from their AI copies? Check your answers at the bottom of the article to see how well you did. 

One of these images has been generated using OpenAI's DALL-E 3 - but which one could it be

The researchers arranged the images according to the basic food type and showed them in three stages of processing. For example: a raw carrot, carrot juice, and a carrot cake 

Experts say that the AI-generated versions might make use of details like symmetry, colour, and shape to produce more appealing designs 

The researchers showed 297 participants a series of images which were either real photos of food or AI-generated copies.

The AI images were created by putting the original photos into OpenAI's DALL-E 3 and asking the AI to 'replicate this photo'.

When shown the images side by side, the researchers found that participants were generally very good at spotting which was made by AI and which was real.

The rate of success ranged from 62 per cent for the image of carrot juice up to 91 per cent for images of chocolate-coated peanuts.

However, when participants were shown the images separately they found the task much more difficult.

When the basic ingredient is not eaten raw the researchers used a photo of it cooked simply as here with these boiled potatoes 

The researchers found that people are generally very good at spotting AI images when they are next to photographs  

AI images tend to make food appear more calorific, according to the scientists 

For separately presented images, the rate at which correctly identified whether they were real or not dipped to only 26 per cent in some cases.

In their paper, published in Food Quality and Preference, the researchers note that while the rate of success remained generally high there were some exceptions.

They suggest that this could be due to the 'heightened complexity of the task'. 

The AI images were generated by asking the AI to recreate the original photo. This means they have not been modified by reconstructed from scratch 

The researchers say that the AI appears to enhance the colour and shape of the foods it pictures 

The scientists say that AI-generated food images could be used to promote unhealthy eating habits 

However, the most surprising finding is that people appear to find AI-generated images more appealing, especially if they don't know they are made by AI.

The researchers asked participants to rate the images on a scale from 'Not appetising at all' to 'Extremely appetising'.

They discovered that when the participants were told which image was AI, they generally rated both images about the same. 

But when they were not told which image was real, participants found that the AI-generated images were significantly more appetising.

They suggest that this could be due to the AI making use of subtle changes to colour, symmetry, shape, and even the position of the food. 

The study found that these images of apples contained the most convincing AI fake, can you tell which is which?

The researchers say that AI-generated food images could lead to unrealistic expectations for food 

The position of food impacts how appealing we find it, which of these looks best based on its position

Lead author Giovanbattista Califano, from the University of Naples Federico II, said: 'As humans, we tend to feel uneasy with objects pointing towards us, interpreting them as threats, even when it's just food. 

'When tasked with replicating food photos featuring items pointing at the viewer, such as a bunch of carrots or a piece of cake, the AI often positions the food so that it doesn't directly point at the viewer.'

It might be the case that the AI makes these kinds of subtle tweaks to the images in order to make them more visually appealing. 

Lighting also makes a significant difference in how tasty food looks to the observer

AI image generators are now able to recreate even small details which can make spotting the difference difficult 

The study found that these two images were the easiest to recognise as real or fake, can you tell which is which?

However, the researchers also warn that this could lead to issues in the future.

Since using AI to generate images is cheap, quick, and easy, study supervisor Professor Charles Spence from the University of Oxford thinks it might become more common. 

He said: 'AI-generated visuals may offer cost-saving opportunities for marketers and the industry by reducing the cost of commissioning food photoshoots.'

Professor Spence also points out that these images pose a risk due to a phenomenon called 'visual hunger' which is where viewing images of food triggers appetite and craving. 

The study found that AI-generated images not only look better than the real thing but tend to show foods as more calorific. 

The AI tended to show more food in each image, for example showing more chips or bigger servings of whipped cream.

If these images proliferate widely, Professor Spence suggests that this could lead people to make less healthy choices about their food.

Professor Spence says: 'This could potentially influence unhealthy eating behaviours or foster unrealistic expectations about food among consumers.'

Could you tell which images were made by AI? Check your answers here: 

  1. Real
  2. Fake
  3. Real 
  4. Fake 
  5. Fake 
  6. Real
  7. Real 
  8. Fake 
  9. Real 
  10. Fake
  11. Fake
  12. Real 
  13. Real 
  14. Fake 
  15. Fake 
  16. Real 
  17. Fake 
  18. Real 
  19. Real 
  20. Fake 
  21. Fake 
  22. Real 
  23. Fake 
  24. Real 
  25. Real 
  26. Fake 
  27. Real 
  28. Fake 
  29. Fake 
  30. Real  
Read Entire Article
Progleton News @2023