AI Already Surpasses Average Human Ability In Many Domains: Google DeepMind Scientist
Jeff Dean emphasises that although artificial intelligence can manage a broad spectrum of cognitive tasks with competence, these systems still have shortcomings and shouldn't be considered flawless.
Artificial intelligence has already exceeded human abilities in certain areas, according to Google DeepMind's chief scientist, Jeff Dean. He observes that many of today’s leading AI models are capable of handling a wide variety of "non-physical tasks" better than an average person can.
During his appearance on the Moonshot Podcast, Dean highlighted that most individuals find it quite challenging when faced with tasks they are not used to doing. In contrast, he pointed out that modern AI systems are often able to tackle these unfamiliar problems with a fair degree of success.
"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," he said.
Dean made it clear that there is a significant difference between outperforming everyday individuals and reaching the standards set by leading experts. He emphasised that although AI can manage a broad spectrum of cognitive tasks with competence, these systems still have their shortcomings and should not be considered flawless.
“They will fail at a lot of things, they’re not human, expert level in some things,” he said.
Dean noted AI’s remarkable capacity to apply its knowledge across a variety of fields. This is a skill that many people find difficult to achieve as effectively, according to the seasoned software engineer.
When questioned about whether computers might soon outpace humans in generating scientific or engineering breakthroughs, he indicated that such a transition is already underway in certain specialised areas. “We’re actually probably already close to that in some domains,” he said.
Another aspect he touched upon in the podcast is his hesitation to discuss artificial general intelligence. “The reason I tend to steer away from AGI conversations is lots of people have very different definitions of it, and the difficulty of the problem varies by factors of a trillion,” the top scientist highlighted.
Demis Hassabis, CEO of Google DeepMind and Dean’s boss, holds a more hopeful view on AGI. In a recent interview with WIRED, he expressed confidence that a major breakthrough in AGI could be realised within the next five to ten years, signalling a significant advance in AI capabilities.
Google DeepMind is the primary IA research lab of the tech giant. It was formed in 2023 after the merger of DeepMind and Google Brain, the company’s in-house AI research team.
