Can you tell the difference between a real face and one generated by artificial intelligence (AI)?
According to a new study, the answer is probably ‘no’.
Researchers from the University of Greenwich found that most people can only spot AI-generated faces a third of the time.
That means you’d actually be better off guessing with your eyes closed than trying to tell the difference between real and fake faces!
However, scientists say there are five ‘strange anomalies’ that can help you to separate fact from fiction.
Speaking to the Daily Mail, co-author Professor Josh Davis, advised keeping an eye out for strange warping of the nose, misaligned ears, wonky eyes, missing teeth, or an unusual hairline.
If you think you can spot any digital trickery, take the below quiz to see how well you distinguish real and AI-generated faces.
In each pair of faces, there is one AI-generated fake. How many can you spot?
Question 1: A new study shows that people can only spot an AI-generated face around 30 per cent of the time. So, which of these two faces is fake, A or B?
Question 2: Researchers say the key to spotting an AI fake is to look out for any strange anomalies. Which of these two do you think is the fake?
Question 3: Which of these is the fake? Pay close attention to facial features like the eyes and nose
Although AI-generated images are getting harder to spot as the technology improves, these artificial images still often contain ‘artefacts’.
These are small distortions or things that don’t look quite right, which are the best sign that a picture is generated by AI.
‘It is the things that aren’t aligned in the image, things that don’t look right, things that are out of place,’ says Professor Davis.
‘You might see a slight discontinuity, things like ears in the wrong place.
‘The nose wouldn’t be quite right; there would be some strange patterns around it that look sort of artificial.’
The eyes are another common source of AI artefacts, as they might be unnaturally asymmetrical or have reflections that don’t match up.
Similarly, if you suspect an image might be generated by AI, experts say that you should pay close attention to their smile.
AI-image generators often struggle to give faces the right number of teeth and to have those teeth appear in the right places.
Question 4: Which of these two men is the AI-generated image? Look for anything that might be asymmetrical or not quite right
Question 5: Which of these two is the fake? Look closely at the details of the hairline
Question 6: The teeth are often a good clue that a picture has been made by AI. Which of these do you think has been generated by a computer?
That makes counting someone’s incisors a fairly reliable method of spotting a potential deep fake.
Additionally, the researchers say that AI-generated faces often feature unusual hairlines that are blurred or run in a strange direction where they meet the forehead.
However, even with these five key signs, research shows that most people still really struggle to spot AI-generated faces.
Researchers from the universities of Reading, Greenwich and Leeds recruited 664 people to take part in an AI-spotting test.
In the study, published in Royal Society Open Science, most people spotted AI fakes just 31 per cent of the time.
A small group of people known as ‘super recognisers’, who have a unique talent for identifying faces, performed better but were still worse than random – only spotting fakes 41 per cent of the time.
This suggests that people are actually more likely to think AI-generated faces were real than a photo of an actual person.
Scientists call this the AI hyper-realism effect, and it means that people often deem AI images to be more real than the real thing.
Question 7: Which of these is the fake? According to scientists, only a small group of ‘super recognisers’ have a better than random chance of spotting it
Question 8: Can you tell which of these two photos has been made by AI?
Question 9: Most people are actually more likely to think an AI-generated image is real than a picture of an actual person. Can you tell which of these two women is AI-generated?
Previous studies have even shown that people find AI-generated faces more trustworthy than real people.
The participants were then given a five-minute training session on how to spot deep fakes, in which examples of artefacts were explained to them.
Following this brief training, people’s scores improved dramatically.
Normal people were now able to spot AI images 51 per cent of the time, while super recognisers could spot the fakes 64 per cent of the time.
However, although this is a promising improvement, most people were still only just better at finding AI-images than they would have been guessing with their eyes closed.
‘You could toss a coin and be just as accurate,’ says Professor Davis.
‘A person is unlikely to be able to make an accurate decision, even after training. I think that is the real risk.’
