The timeline for AGI depends on who you ask
Matt V
March 19, 2025
DEEPMIND
AGI, aka Artificial General Intelligence - AI that can think, reason and problem-solve like humans, feels inevitable.
But when will it actually happen?
That depends on who you ask.
DeepMind CEO Demis Hassabis believes we’re 5 to 10 years away.
Speaking in London, he pointed out that while today’s AI is impressive, it still struggles with reasoning, planning, and making sense of the real world.
For AI to be considered AGI, it needs to handle everything a human can, not just specific tasks like writing or coding.
While AI has already mastered strategy games like Go, applying that intelligence to real-world problems is a much bigger challenge.
Hassabis isn’t the only one urging patience.
Baidu CEO Robin Li suggests AGI is more than a decade away, contradicting the more optimistic predictions from other tech leaders.
But not everyone agrees with Hassabi’s timeline.
Some believe we’ll see AGI much sooner:
Dario Amodei (Anthropic) thinks we’ll have AI outperforming most humans at nearly all tasks within 2-3 years.
Jeetu Patel (Cisco) expects AGI to show up as early as 2025, with superintelligence, AI that surpasses human intelligence, only a few years behind.
Elon Musk has thrown 2026 into the mix, with OpenAI’s Sam Altman saying it’s coming “reasonably soon.”
So what’s holding us back?
Hassabis argues that the hardest part is teaching AI to understand the world like humans do.
AI can beat grandmasters at chess and Go, but dealing with predictable real-world situations? That’s another story.
There could be a possible solution for this.
Multi-agent AI is AI systems that work together, similar to how human societies function.
DeepMind has already experimented with this, using AI agents to master complex strategy games like Starcraft.
Hassabis also points to world models, AI that can predict and understand real-world events, as a crucial step towards AGI.
Wake me up when AGI learns how to handle a rush-hour parking exit.