Can YOU spot the fake faces? Take the test to see if you can distinguish between real and AI-generated people

A new study from Australian National University finds that most people can only distinguish AI-generated faces from real ones at random levels, but training focused on six key characteristics—facial distinctiveness, memorability, proportionality, symmetry, attractiveness, and expressiveness—can improve accuracy. The researchers warn that AI-generated faces are becoming harder to spot, fueling a projected $40 billion in fraud losses in the US by 2027.

Can YOU spot the fake faces? Take the test to see if you can distinguish between real and AI-generated people READ MORE: Kylie Jenner collaborates with Meta on £359 AI glasses /sciencetech/article-15922373/Kylie-Jenner-collaborates-Meta-359-AI-glasses.html See more Daily Mail on Google - save us as a Preferred Source https://google.com/preferences/source?q=dailymail.com Can you tell the difference between a real person and an image generated by artificial intelligence /sciencetech/ai/index.html AI ? According to a new study, it might be a lot harder than you think. Researchers from Australian National University ANU warn that the average person is no worse off guessing at random when it comes to spotting AI-generated faces. However, the experts say you can train yourself to spot the imposters by honing your natural intuitions. The researchers found that people can be taught to focus on six key characteristics which can help separate real humans from digital doppelgangers. Those are: Facial distinctiveness, memorability, proportionality, symmetry, attractiveness, and expressiveness. But lead author Amy Dawel, associate professor of psychology at ANU, says just knowing what to look for isn't enough - you have to learn by practising. So, how many of these AI-generated faces can you distinguish from real people? Take the quiz below to find out. In a new paper, published in the journal PNAS https://doi.org/10.1073/pnas.2602122123 , Dr Dawel and her co-authors warn that AI-generated faces are getting much harder to spot. Today some programs are able to create faces that are all-but indistinguishable from the real thing. This is driving a boom in AI-powered fraud, which is projected to lead to losses totalling $40 billion £30.2 bn in the United States alone by 2027. One of the big issues is that AI's ability to generate deepfakes has accelerated much faster than our ability to spot them /sciencetech/article-15751525/spot-fake-AI-generated-VOICES.html , as once-reliable advice becomes outdated. For example, telling people to look for 'AI artefacts' like sixth fingers, misaligned teeth, or wonky ears simply no longer works. Studies have shown that this advice doesn't improve people's ability to spot deepfakes /sciencetech/article-15283159/spot-AI-fake-face-test.html , and real-life fraudsters can easily edit out or avoid these errors. Instead, the researchers have developed a new training method that teaches people to pick up on 'global impressions' rather than specific features. Dr Dawel says: 'Our training approach has a deliberate twist: we do not tell participants what to look for. Researchers found that you can learn to spot AI-generated faces more reliably by rating each of these labelled examples from zero to seven according to six criteria: Facial distinctiveness, memorability, proportionality, symmetry, attractiveness, and expressiveness 'Instead, we expose them to AI-generated and genuine human faces while directing their attention to the qualities that distinguish the two. 'Over repeated exposure, participants build an intuitive sense for spotting AI faces, in the same way that expertise often develops through experience rather than explicit rules.' In their study, participants were shown pictures correctly labelled either as AI-generated or human and asked to rank them on the six key characteristics. This wasn't so that the participants could learn specific rules, such as 'high attractiveness is a sign of being an AI', but to help them hone their intuition. What was so striking is just how much this short, online intervention improved people's ability to distinguish real and fake pictures. Before training, people were able to find the AI imposter hidden alongside two real humans just 41 per cent of the time. Likewise, people correctly identified a single human face as real in only 52 per cent of cases and correctly labelled an AI-generated face with 47 per cent accuracy. But after practising on the labelled examples, the average accuracy doubled after a brief online training session, with some 'high performers' achieving near-perfect results. Rating labelled examples on these criteria helps you develop an intuitive ability to distinguish real and AI-generated faces Scientists found that a short online training session using this method doubled the average accuracy with which participants spotted AI fakes Remarkably, these test results were then replicated by a team led by Professor Jim Tanaka and Dr Eric Mah at the University of Victoria, Canada. Dr Mah says: 'The replication shows that the findings weren’t a fluke – when we trained a new set of people in a different country, we saw them improve just as much. 'Online training was effective, so our training program could easily be implemented at scale for little cost.' The researchers say this works because facial impressions are formed rapidly and intuitively, and are very sensitive to the sorts of systemic biases inherent in AI algorithms. That sense of when a face looks right is something we all have, but people generally fail to leverage those impressions without training. Directing people to pay attention to the broader, global characteristics trains them to hone their intuitive knack for spotting real faces. While algorithms for detecting deep fakes do exist, these tend to be incredibly opaque 'black boxes' with potential hidden flaws. Instead, the researchers argue that we 'urgently' need to improve our own AI-detection abilities to fight back against deepfake scams.