From South Park v Trump to AI slopaganda: deepfakes are now part of the news cycle, for better and for worse | Anna Broinowski

"But on our screens from film, art and satire to the algorithmically turbo-charged, factually opaque, monetised churn of the 24/7 news cycle AI is already making us laugh. Deepfakes synthetic audio and video of people doing and saying things they never said or did are the chief comedic disruptors in a suite of increasingly persuasive AI tools shaping the post-truth reality envisioned by the Microsoft engineer Eric Horvitz, where fact and fiction are indistinguishable."

"In eight short years, deepfakes have risen from cultural outlier to mainstream meme, embodying the futurist Roy Amara's Law: we overestimate the effects of new technology in the short run, but underestimate its long-term impacts. And with some experts predicting 90% of online content could be AI-generated by 2027, the extent to which synthetic replicas will change how we trust and interact with on-screen depictions of real people is now an urgent question for creators, policymakers and viewers."

"But as soon as deepfakes started being used a political tool, concerns snowballed into panic. Philosophers labelled them an epistemic threat to evidentiary systems; corporations mounted costly (and unsuccessful) detection programs; and pundits warned of an info-apocalypse, in which a convincing deepfake of a world leader could start world war three. Less dramatic but equally chilling was the prediction that the mere awareness of deepfakes would collapse civic trust altogether."

AI will not be a threat to authors until it can write genuinely funny books. On screens, AI already shapes comedy across film, art, satire and the algorithmically turbo-charged, opaque, monetised churn of the 24/7 news cycle. Deepfakes—synthetic audio and video of people doing and saying things they never said or did—serve as primary comedic disruptors and blur fact and fiction. Deepfakes rose from nonconsensual porn in 2017 to mainstream meme within eight years, illustrating Amara's Law about short-term overestimation and long-term underestimation. With predictions that up to 90% of online content could be AI-generated by 2027, synthetic replicas pose urgent challenges for trust, creators, policymakers and viewers.

Read at www.theguardian.com