Tips from a tech expert on spotting AI-generated fakes

CTV News

a day ago

With the upsurge in deepfakes and how easy it is to create them we talk spotting fakes and how you can stay vigilant.
The widespread availability and ease of use of image and video generators powered by artificial intelligence (AI) means what you see on the internet may not be what it seems.
These programs and apps can create increasingly realistic-looking pictures, videos, and sound with only a few simple text prompts.
In one recent example, a popular indie rock band that had more than a million monthly listeners on Spotify turned out to be AI-generated. In another case, officials in the U.S. warned of attempts to use AI tools to impersonate Secretary of State Marco Rubio.
The Canadian Anti-Fraud Centre also recently warned about a rise in AI-generated voice messages being used to impersonate high-profile figures to extort victims for money and personal data.
Even CTV News Ottawa was subject to impersonation using these kinds of tools. Following a report about an online scam that affected a local family, a fake version of our report circulated online, changing the words our anchors said to make it sound like they were advertising the scam in a positive light.
Technology analyst and journalist Carmi Levy told CTV Morning Live that, for now, most AI-generated images and videos have imperfections that can give the game away.
'The two words that we usually use that seem to work for now are 'lean in.' So, when you're watching a video, slow it down, lean in and watch for what we like to call 'tells' that give away the fact that it's not real, it is synthetically generated, AI-created,' he said.
Some things to watch out for include how the subject of the video moves.
'The movement might not be as smooth, it might be kind of stop-and-start,' said Levy. 'There might be little hitches here and there; sometimes the lips, the audio doesn't quite match what you're seeing on the individual's face.'
The lighting or shading of the video might not look natural, or there might be other artefacts that seem out of place, Levy says.
'For example, if someone puts their arm around the back of somebody, there might be a placement issue or extra pieces; sometimes there are additional fingers or not enough fingers.'
Levy warns, however, that these tools are becoming more sophisticated and are better able to replicate a realistic scene.
'With each new generation of these video generator apps, they get better and so it gets even harder to tell that something was AI-created. In fact, the new version of Veo, Veo 3, is so good that for a lot of people, even if they lean in, it's increasing difficult to tell,' he said.
The federal government says rapid advances in AI have underscored the risks associated with the technology.
'The Government is taking actions to ensure that AI is developed in a safe and responsible manner,' a spokesperson for Innovation, Science and Economic Development Canada said in a statement to CTV News Ottawa.
'Just last November, the Government of Canada launched the Canadian Artificial Intelligence Safety Institute with the mandate to advance scientific understanding of the risks associated with the most advanced AI systems, develop measures to reduce those risks, and build trust to foster AI innovation.
'In addition, the government also launched a Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems in September of 2023. The Code calls on AI developers and managers to take risk-mitigation measures such as developing and implementing reliable and freely available methods to detect content generated by their AI system. Greater transparency on AI generated content can help protect Canadians against AI-enabled scams and fakes. The Government is protecting Canadians from fraud and impersonation, including cases involving AI, and will continue to explore all options to strengthen these protections.'
Levy says when you see something online, before sharing it, ask questions and make sure it's coming from a trustworthy source.
'How do you tell if it's real or not real? You check the source. You make sure it's coming from a legitimate source that you can trust. In many cases, if you can't trace it back to its origin, you should start with the assumption that it's not real,' he said.
'That's what I do; I start cynically and require the content to prove to me that it is, in fact, real.'