\u2022 This article includes a range of photographs of people. As you read, try to guess which ones are real and which are AI-generated.<\/p>\n
The expert behind a new training program designed to help people debunk AI deepfakes has admitted we find ourselves in a \u201ccat and mouse game\u201d with the advancing technology.<\/p>\n
Australian National University has launched a new scheme that it says makes people about 40 per cent better at rooting out AI-generated faces from shots of human beings.<\/p>\n
But the lead researcher behind the program has admitted that it is only a matter of time before the technology figures out new ways of not getting caught out.<\/p>\n
The cutting-edge program begins with participants being given an identity parade of mugshots, some AI and some real, and tasks them with rooting out the real from the fake.<\/p>\n
Participants are then given a short course on how to spot the differences, then asked to repeat the task at the end.<\/p>\n
Lead researcher Amy Dawel said the end result was about a 40 per cent improvement in people\u2019s recognition skills.<\/p>\n
\u201cWe have found that even people who are super recognisers do not do as well as they expect in the first instance,\u201d Professor Dawel said. \u201cWe generally find people are surprised by quite how terrible they are at it and how much they improve. <\/p>\n
\u201cThe more confident they are, the more mistakes they tend to make.\u201d<\/p>\n
Professor Dawel said the task had become increasingly difficult as the technology advanced, with AI-generated images getting tougher and tougher to spot.<\/p>\n
She said gone were the days of simple telltale signs such as extra fingers on hands or non-matching earrings.<\/p>\n
\u201cAI-generated faces are generally very average looking and do not stand out from the crowd, which can also make them look more attractive,\u201d she said. <\/p>\n
\u201cThis has meant real people have actually become terrible at detecting AI gens.\u201d<\/p>\n
The research project behind the training has uncovered six tricks to rooting out the fakes: distinctiveness, memorability, proportionality, symmetry, attractiveness and expressiveness.<\/p>\n
Professor Dawel said symmetry in particular was a strong sign that an image was AI.<\/p>\n
\u201cAI generated faces tend to be far more symmetrical, while real people tend to be more quirky and have individual characteristics,\u201d she added.<\/p>\n
Cat and mouse game<\/p>\n
Professor Dawel said it is getting more and more important for people to be able to root out AI deep fakes due to the ways the technology is being taken advantage of in darks corners of the web.<\/p>\n
She said it is estimated that AI fraud is predicted to cost the global economy 40 billion US dollars by the end of next year.<\/p>\n
\u201cThere are so many ways AI generated faces are being used and abused, whether that is cat-fishing on dating sites, accessing bank accounts or counterfitting passports,\u201d she said.<\/p>\n
\u201cPeople are also in a position now where they can hide behind it and claim that things that are actually real are deep fakes.<\/p>\n
\u201cBeing able to tell the difference is so important.\u201d<\/p>\n
However, the associate Professor also admitted that as we get better at detecting deep fakes, AI technology will inevitably also get better at not getting caught out.<\/p>\n
\u201cI think it is definitely a cat-and-mouse game and there is no real getting away from that,\u201d she added.<\/p>\n
\u201cBut equally, while AI itself is getting better, so too are AI detectors, but human psychology is also an important thing to keep relying on.\u201d<\/p>\n
\u2018Getting harder and harder\u2019<\/p>\n
Professor Toby Walsh, from the University of NSW, said the rapidly advancing technology was forcing people to regularly question what is in front of them.<\/p>\n
\u201cThe real difficulty is that we cannot really believe our own eyes anymore,\u201d he said. \u201cI think we are going to keep on seeing things that are not real but also questioning things that are real at the same time.<\/p>\n
\u201cThe way to combat this is with education but also with digital watermarks.\u201d<\/p>\n
Digital watermarks are invisible, machine-readable identifiers attached to files such as a photograph that reveal its origin.<\/p>\n
Professor Walsh said these could potentially become the only way of telling apart the real from the deepfake as the technology continues to advance.<\/p>\n
\u201cIt really is getting harder and harder to tell real and fake apart, the technology is advancing so quickly,\u201d he said.<\/p>\n
\u201cAnother problem is that it is giving people a way to escape accountability. <\/p>\n
\u201cThere are politicians who have said and done awful things only to dismiss them as deepfakes \u2013 which is all their supporters need to hear to believe them.\u201d<\/p>\n
\u2018Arms race\u2019<\/p>\n
Adelaide University associate professor in computer sciences Wolfgang Mayer said the advancement in AI technology had been brought about through competitiveness among some of the world\u2019s largest tech firms.<\/p>\n
\u201cIt has come such a long way in the past few years and I think that is down to an arms race between big tech companies who are all determined to be the best and producing the most content,\u201d he said.<\/p>\n
\u201cClearly, they\u2019re not setting out to create deepfakes, but it is an obvious by-product of the technology being available.<\/p>\n
\u201cThe technology is there and easy to use, for both the good and the bad.\u201d<\/p>\n
He added that there was a greater need for people to learn to question sources of what they see online.<\/p>\n
\u201cTeaching people how to spot deepfakes is a useful thing to do in the short term, but in the long run it\u2019s far more important to raise awareness of the fact that not everything you see online is real,\u201d Professor Mayer said.<\/p>\n
He added that the technology had advanced to a point where rooting it out was about looking at what was not there rather than what was there.<\/p>\n
\u201cDeepfakes are almost becoming too perfect,\u201d he said. <\/p>\n
\u201cWe used to easily be able to tell from things like eyes, mouths or teeth, for example. Now you almost need to look for what is not there, blemishes that exist in real people.\u201d<\/p>\n
What does AI say?<\/p>\n
While the university experts have shared tips for spotting AI-generated faces, would the technology also be prepared to show its hand?<\/p>\n
NewsWire turned to Google\u2019s Gemini AI chatbot to request some tips that provided several telltale signs.<\/p>\n
\u2022 Hands and feet \u2013 look for extra fingers, missing knuckles, unnaturally long digits or hands that morph into objects being held<\/p>\n
\u2022 \u2018Plastic\u2019 or airbrushed skin \u2013 look for natural imperfections such as press, fine lines, moles and subtle blemishes to find a real person. AI images can often look like heavy beauty filters<\/p>\n
\u2022 \u2018Spaghetti\u2019 hair \u2013 look closely at where the hairline meets the forehead. AI often struggles with realistic hair roots<\/p>\n
The chatbot added that the most effective way to spot a fake was to make sure you looked closely. \u2022 Photograph 3 is Professor Dawel, photograph 5 is Professor Mayer and photograph 7 is
\u201cWhen trying to spot an AI image, always zoom in,\u201d it said. \u201cAI images look incredibly convincing as thumbnails on a smartphone screen, but zooming in by 200 per cent to 300 per cent on the eyes, ears and hands almost always reveal the digital seams.\u201d
To register an interest in the training, visit https:\/\/tinyurl.com\/ai-face-study-register<\/a>\u00a0<\/p>\n
Professor Walsh. All the rest were generated using StyleGAN3 and are not real people. <\/p>\n","protected":false},"excerpt":{"rendered":"\u2022 This article includes a range of photographs of people. As you read, try to guess which ones…\n","protected":false},"author":2,"featured_media":795297,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[256,254,255,64,63,187436,105],"class_list":["post-795296","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-au","tag-australia","tag-newswire-news","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/795296","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/comments?post=795296"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/795296\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media\/795297"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media?parent=795296"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/categories?post=795296"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/tags?post=795296"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}