Photographs were seen as less realistic than computer images but there was no difference with pictures of people of colour.
Imagine a situation right out of a movie – technology so advanced that it not only sounds more real than actual humans but looks even more convincing. Well, it seems that moment is already here.
A recent study discovered something quite surprising: people are more likely to mistake pictures of AI-generated white faces as real humans compared to actual photographs of people.
The researchers reported, “Remarkably, white AI faces can convincingly pass as more real than human faces – and people do not realize they are being fooled.”
This research, conducted by a team from Australia, the UK, and the Netherlands, has significant implications in the real world, especially in areas like identity theft. There’s a concern that individuals might be deceived by digital impostors.
However, the study found that these results didn’t apply to images of people of colour. This discrepancy might be because the AI faces were predominantly trained on images of white individuals.
Dr.Zak Witkower, one of the researchers from the University of Amsterdam, pointed out that this could have consequences for various fields, from online therapy to robotics. He mentioned, “It’s going to produce more realistic situations for white faces than other race faces.”
The team warns that this situation could lead to confusion between perceptions of race and perceptions of being “human.” Moreover, it might perpetuate social biases, impacting areas like finding missing children, as AI-generated faces play a role in such efforts.
The study, published in the journal Psychological Science, details two experiments. In one, white adults were shown AI-generated white faces and real human white faces. Participants had to determine whether each face was AI-generated or real and rate their confidence.
Results from 124 participants showed that 66% of AI images were considered human compared to 51% of real images. Interestingly, this trend did not apply to people of colour, where both AI and real faces were equally judged as human. The researchers also noted that participants’ own race did not affect the results.
In another study, subjects were requested to provide subjective evaluations of AI and human faces for 14 parameters (e.g., age or symmetry) in a blinded setup such that they had no knowledge which face was computer generated or not.
The team’s analysis was based on 610 participants whose responses implied that individuals are likely to mistake AI for a human if they had more proportionate face, familiarity with it, or low memorability levels.
However, in a somewhat ironic twist of fate, the research team managed to create an automated deep learning model capable of distinguishing between genuine and artificial facial images using 94 percent accuracy.
The leader author of the research from the University of Aberdeen, Dr Clare Sutherland stated that the research emphasized a need to address the biases in AI.
She says: as the world goes through massive change brought about by AI, we must ensure that at all times, no one is left out of the equation irrespective of their race, sex, age, or whatever protected characteristics they may have.
Second, a different set of participants were asked to rank AI and human faces on seventeen features (such as age or symmetry) without informing them that some of the face images were computer generated.
As a result, it can be assumed that people would mistake an AI-generated face as human because of its larger proportionality, familiarity, and less memorable appearance.
Interestingly enough, even though some humans cannot differentiate between naturally created images and those processed by AI, this team created a machine learning algorithm that does it with almost a hundred percent success rate.
Dr Clare Sutherland, one of the authors of this research work from the University of Aberdeen stated that such biases should be eliminated in Artificial Intelligence.
She explained saying “In this era when the earth adopts AI, everyone has the right not to be discriminated against by any trait such as race, gender, age, and so on”.
Leave a Reply