Ciarán O’Connor on the challenges of spotting AI-generated images

ISD Senior Analyst Ciarán O’Connor spoke to RTÉ about the growing challenges for experts and the public coming across fast-improving AI-generated images. The concern arises amid AI images of celebrities and politicians generating millions of views on social media platforms such as X. 

“Generative AI is accelerating at such a quick rate that the tells – the less refined, clumsy giveaways – are quickly being addressed,” said Ciarán. “The commonplace clues may even be outdated by next year.” 

While AI technology has rapidly improved their image quality, there are still signs that viewers should still look out for. Ciarán first notes in the article that poor editing is often an early tell-tale sign, such as “edges of clothes, body parts that don’t line up, sleeves or wrist or ears that don’t sit where they’re supposed to.” In addition, skin textures that are suspected of being “too smooth or perfect” could be another indicator, as well as background details where the technology puts less effort.  

With regards to timing and context of disinformation spreading, Ciarán highlights the importance of high-profile topics. “Election campaigns can be primed for a way to sway public opinion by any means necessary by those involved in the campaign.”

Ciarán adds that when viewing images, people should scrutinise the source behind the image, who is sharing it, and whether someone could benefit from misleading viewers.