3
AI-generated election disinformation will likely be in all places
If current elections are something to go by, AI-generated election disinformation and deepfakes are going to be an enormous downside as a document variety of folks march to the polls in 2024. We’re already seeing politicians weaponizing these instruments. In Argentina, two presidential candidates created AI-generated pictures and movies of their opponents to assault them. In Slovakia, deepfakes of a liberal pro-European get together chief threatening to boost the worth of beer and making jokes about baby pornography unfold like wildfire in the course of the nation’s elections. And within the US, Donald Trump has cheered on a bunch that makes use of AI to generate memes with racist and sexist tropes.
Whereas it’s arduous to say how a lot these examples have influenced the outcomes of elections, their proliferation is a worrying development. It should turn into tougher than ever to acknowledge what’s actual on-line. In an already infected and polarized political local weather, this might have extreme penalties.
Just some years in the past making a deepfake would have required superior technical abilities, however generative AI has made it stupidly straightforward and accessible, and the outputs are trying more and more life like. Even respected sources may be fooled by AI-generated content material. For instance, users-submitted AI-generated pictures purporting to depict the Israel-Gaza disaster have flooded inventory picture marketplaces like Adobe’s.
The approaching yr will likely be pivotal for these combating in opposition to the proliferation of such content material. Strategies to trace and mitigate it content material are nonetheless in early days of improvement. Watermarks, corresponding to Google DeepMind’s SynthID, are nonetheless principally voluntary and never utterly foolproof. And social media platforms are notoriously gradual in taking down misinformation. Prepare for an enormous real-time experiment in busting AI-generated faux information.
—Melissa Heikkilä
STEPHANIE ARNETT/MITTR | ISTOCK, ENVATO
4
Robots that multitask
Impressed by among the core strategies behind generative AI’s present growth, roboticists are beginning to construct extra general-purpose robots that may do a wider vary of duties.
The previous few years in AI have seen a shift away from utilizing a number of small fashions, every skilled to do completely different duties—figuring out pictures, drawing them, captioning them—towards single, monolithic fashions skilled to do all this stuff and extra. By displaying OpenAI’s GPT-3 just a few further examples (often called fine-tuning), researchers can practice it to unravel coding issues, write film scripts, move highschool biology exams, and so forth. Multimodal fashions, like GPT-4 and Google DeepMind’s Gemini, can resolve visible duties in addition to linguistic ones.
The identical strategy can work for robots, so it wouldn’t be essential to coach one to flip pancakes and one other to open doorways: a one-size-fits-all mannequin may give robots the power to multitask. A number of examples of labor on this space emerged in 2023.
