The race to build AI that can think like a human may be moving faster than expected.
Google DeepMind co-founder Demis Hassabis now believes artificial general intelligence (AGI) could arrive within just two to three years, potentially putting human-level AI within reach by 2028.
Speaking to The Wall Street Journal Magazine in an interview published Wednesday, the Nobel Prize-winning AI pioneer shortened his previous estimate of three to five years.
"It's soon," Hassabis tells the publication, adding, "I'm updating nearer and nearer."
That's a striking statement from one of the researchers behind Google's Gemini models. But it also raises a bigger question: If AI is potentially just two years away from matching human intelligence, why does today's technology still struggle with seemingly simple tasks?
Why Google's AI chief is moving up his timeline
Hassabis has been predicting the arrival of AGI for years, but his estimates have become increasingly aggressive.
In December 2025, he told Axios that AGI could still be five to 10 years away. By May 2026, he was suggesting it might arrive around 2030, or possibly even sooner.
Now, he's putting the timeline at 2028 or 2029.
Part of his growing confidence comes from the progress AI companies have made with autonomous agents. Unlike traditional chatbots that wait for instructions, these systems can carry out multistep tasks, use tools and make decisions along the way.
Following Google I/O in May, Hassabis told Axios that today's AI agents were essentially a practice run for AGI.
He also acknowledged that some of his urgent language was intended to encourage governments, economists and the public to take the approaching technology more seriously.
What's still missing before AI can think like us?
Despite the impressive capabilities of ChatGPT, Gemini and Claude, today's AI systems aren't nearly as flexible or reliable as human intelligence.
During an April interview with Y Combinator, Hassabis identified several problems that researchers still need to solve.
One is continuous learning. Humans can learn from new experiences and apply that knowledge to future situations. AI models, by comparison, don't automatically update their underlying knowledge every time they encounter something new.
Memory is another challenge. Even with increasingly large context windows, AI assistants can lose track of important information or struggle to connect details across extended conversations and tasks.
Then there's reasoning. AI can solve difficult math problems and write sophisticated code, yet it can still make basic logical mistakes or confidently produce incorrect answers.
Hassabis believes solving these problems, along with making AI more consistent, will be essential before AGI becomes a reality.
That also helps explain why having a chatbot that scores highly on exams isn't necessarily the same as building genuinely human-level intelligence.
What would AGI actually mean for you?
The biggest difference between today's AI assistants and AGI would be flexibility.
Right now, you might use ChatGPT to plan a vacation, Gemini to summarize emails or Claude to help write a document. These tools are powerful, but they still need supervision and can make mistakes.
An AGI system could potentially handle a much broader range of intellectual tasks, adapting to unfamiliar problems rather than relying on capabilities learned during training.
Imagine giving an AI assistant a complicated goal, such as planning a family vacation, comparing prices, coordinating schedules and making reservations, and having it manage the entire process with minimal guidance.
In the workplace, more capable AI could take on increasingly complex research, analysis and administrative tasks. In healthcare, it could help scientists accelerate drug discovery or identify promising new treatments.
There's already some precedent for the scientific potential. Hassabis shared the 2024 Nobel Prize in Chemistry for his work on AlphaFold, an AI system that transformed scientists' ability to predict protein structures.
But there are also legitimate concerns about jobs, privacy, security and what happens when increasingly autonomous AI systems make decisions incorrectly.
Google DeepMind has published a framework for measuring progress toward AGI, recognizing that intelligence includes different levels of capability, breadth and autonomy. There's still no universally accepted test that determines when AGI has officially arrived.
The takeaway
Hassabis isn't the only AI leader predicting a dramatic transformation in the next few years, but his latest timeline is significant because of his role in developing some of the world's most advanced AI systems.
For everyday users, the takeaway here is that AI tools we're already using could become considerably more capable, autonomous and deeply integrated into our lives over the next few years.
Whether that happens on Hassabis' timeline remains to be seen. But if he's right, the next two years could look very different from the last two.
More from Tom's Guide
- OpenAI unleashed thousands of AI agents to study a Parkinson-like disease — here's what they found in 32 minutes
- 7 ChatGPT plugins that are a must-have for anyone with a side hustle
- ‘Babe, wake up, the internet is collapsing’: Two massive flaws just broke the web, and the data of billions is in the crossfire