Psychology in the News | Can You Recognise Artificial Faces?

Artificial intelligence (AI) has developed at an incredible rate in recent years. As an individual, you may be confident in your ability to be able to distinguish the real from the artificially generated, but AI-generated faces are now so realistic that research shows that people often mistake them for being human.

Moreover, new research suggests that AI has now reached such a level of realism that AI-generated faces can appear more real than real faces and we can now be easily misled into believing that an AI-generated face is a real face. In the study, participants were shown images of different white faces and asked if the face was real, or AI-generated. AI faces were judged to be real two-thirds of the time, which was more often than for real faces. This result was not replicated on non-white images of faces.

Researchers believe that this is because AI algorithms tend to be trained disproportionately using white faces. One of the authors of the research explained why this finding is of such concern. Dr Amy Dawel said: “If white AI faces are consistently perceived as more realistic, this technology could have serious implications for people of colour by ultimately reinforcing racial biases online. This problem is already apparent in current AI technologies that are being used to create professional-looking headshots. When used for people of colour, the AI is altering their skin and eye colour to those of white people.”

A further finding from the study was that the people who believed that the AI faces were real were also the most confident in their judgement. Therefore, people are not aware that they are being tricked by AI faces which has implications for people encountering such images in their day-to-day lives.

Older AI face generation would create clues that a face was artificial. For example, a shadowing effect near the eyes may provide us with a clue that a face is not real. This is known as the ‘uncanny valley effect’ which describes the emotional feeling of unease or revulsion one might experience in response to, for example, images which appear to represent a human but which upon closer inspection are not quite life-like. However, this piece of research suggests that AI has now overcome this ‘uncanny valley’ for the creation of still images. AI faces are still different to human faces but in different ways, for example, white AI faces in this study tended to be more in proportion than real faces. However, with the rate of development of AI even these differences will likely disappear soon.

The researchers who published the paper argue that their findings have serious implications for identity theft and recommended that the public should be educated about AI hyper-realism so that any images they view online can be approached with a healthy level of scepticism.

ANSWER THE FOLLOWING QUESTIONS

1. In one part of this study, all 124 participants were shown 50 AI faces and 50 human faces, what experimental design is being used?

2. Why do you think this design was chosen, in this study?

3. What ethical implications are raised by this research study?

4. Why could the algorithms which are used to train AI be accused of ethnocentrism?

Extension: the public cannot tell the difference between AI-generated faces and real faces, how do
you think we could educate people about the issues raised by this research? Is there another way to
solve the issues raised?