Google’s Jeff Dean: AI Already Surpassing Average Humans in Some Tasks

Artificial Intelligence continues to evolve at an extraordinary pace, and according to Google’s Chief Scientist Jeff Dean, it may have already outstripped the average human in several areas. Speaking on The Moonshot Podcast, Dean explained that today’s large language models (LLMs) demonstrate broad competence across a wide spectrum of non-physical tasks, even if they are not yet flawless.

“Most people are not that good at a random task that they’ve never done before, and some of the models we have today are actually pretty reasonable at most things,” Dean said, highlighting the adaptability of modern AI systems compared to human limitations.

Dean noted that while these models may surpass average performance in many cases, they should not be confused with expert-level proficiency. “They will fail at a lot of things, they’re not human expert level in some things,” he admitted. For him, the real breakthrough lies not in perfection but in the AI’s ability to generalize knowledge across diverse domains — something even skilled individuals often struggle to achieve.

This growing general competence has already begun transforming industries. Dean pointed to fields where machines are automating work once thought exclusive to humans. When asked whether AI could soon contribute to scientific or engineering breakthroughs, he suggested that this shift is already underway. “We’re actually probably already close to that in some domains,” he revealed.

He emphasized that areas best suited for AI-driven innovation are those with fast feedback loops, where ideas can be generated, tested, and refined in quick succession. “It’s going to have to be an area that is amenable to a fully automated loop of generating some ideas, trying them out, getting some feedback exploring essentially a very large space of possible solutions to some problems,” Dean explained. According to him, reinforcement learning algorithms and large-scale computational search have already proven highly effective in such scenarios.

However, not all fields can be easily accelerated by AI. Tasks that require lengthy evaluation cycles — for example, scientific experiments that take weeks or months — remain difficult to automate. Still, Dean remains optimistic about the potential of AI to drive progress. “There will be a lot of domains where automated search and computation actually can accelerate progress,” he said, citing both scientific research and engineering design as areas ripe for transformation.

Dean’s remarks underscore a pivotal moment in the evolution of AI. While machines are not yet replacing the world’s top professionals, their growing competence across multiple domains raises questions about how society will adapt. For everyday problem-solving and innovation, AI may already be on the verge of becoming a partner rather than just a tool.

As AI continues its rapid advance, the conversation is shifting from whether machines can match humans to how we integrate these increasingly capable systems into our daily lives and industries.