It might be a while before AI becomes as good as humans in the real world, but for activities that are done in front of a screen, it might be close to getting there.
In a bold prediction, Stability AI founder Emad Mostaque has claimed that by next year, artificial intelligence will surpass human performance in most screen-based tasks. Mostaque’s forecast points to a rapid evolution in AI capabilities, particularly in the realm of “worker AIs” designed to handle digital tasks with increasing autonomy and efficiency. His comments suggest a seismic shift in the nature of work, where human interaction with AI will become indistinguishable from collaborating with a remote team.
Mostaque’s assertion is grounded in the remarkable progress of generalist AI agents. He highlights the significant advancements in benchmarks designed to test these agents’ abilities. “When you look at generalist technology, this is the really interesting one. You see the rise of agents like Genspark, and Manus and others,” he notes. “When Meta first released their Gaia benchmark last year, the top model scored about 20, 25%. Now we just [saw a] state-of-the-art open framework that beats Manus and Genspark. That was 76% for generalized tasks on the other side of a computer screen.”
He continues, drawing a direct comparison to human capability: “Human level across everything is 92%, and I think that’s broken in the next year. So effectively now, for any job that you can do on the other side of a screen, an AI will probably be able to do it better, faster, and cheaper by next year.”
This transformation, according to Mostaque, will be accompanied by a more natural and human-like interface for interacting with AI. “The way you will interact with those AIs, given the advances in things like HeyGen Avatars and others, will probably be by having a Zoom call with them or WhatsApp,” he predicts. “This real humanization of AI is a very interesting thing. If you use the new OpenAI advanced mode or things like Sesame or others, it’s like talking to a real person in many ways.”
He emphasizes the role of context in this new paradigm: “The key part of this interaction layer is that individual context. And then the more helpful they are, the more we outsource our own thinking. Just like we build teams, we can’t manage everything by ourselves. And again, the way that we interact with them by next year will be in exactly the same way that you interact with your remote team. You won’t be able to tell the difference.”
Mostaque draws a clear distinction between the much-hyped, all-powerful AGI (Artificial General Intelligence) and the more practical “worker AIs” that he believes will drive immediate and significant change. “For me, that’s AGI, to be honest. But the things we trust it for isn’t necessarily, ‘I’ve got a remote worker that I want to come up with a new symphony or do something massively creative.’ That’s what we are there for,” he clarifies. “I want underlings that can go and get jobs done. I want cooks, not chefs. And so I think people focused on this massive polymath AI, when actually these worker AIs is what’s really gonna move the needle.”
He concludes with a sense of imminent takeoff: “Again, we’re seeing the first emergent UIs for that. The context windows, the continuous learning and other elements that I think will mean that next year is the big takeoff year.”
Mostaque’s vision of AI outperforming humans at screen-based tasks within a year is ambitious, yet it taps into a palpable trend of rapid AI advancement. The development of sophisticated AI agents by major tech players, including Google’s “Project Astra” and OpenAI’s advancements in its models, lends credence to this accelerated timeline. These agents are increasingly capable of understanding context, processing information from various sources, and executing multi-step tasks. For businesses, this signals a future where AI assistants could handle everything from complex data analysis and report generation to customer service and administrative tasks, all with minimal human supervision. The implications are profound, suggesting a near-future where productivity is supercharged, but also raising critical questions about the future of jobs that are primarily computer-based and the skills that will be most valued in a world where “worker AIs” become commonplace.
