Demis Hassabis, CEO of Google DeepMind, firmly believes that AI is going to transform education, coding and even drug discovery. In his recent podcast interview with Rowan Cheung, the founder of The RundownAI, Hassabis spoke about the biggest announcements, the AI as a companion conundrum, and how the next decade of technology will shape considering the rapid advancements in AI. The CEO asserted that AI is here to make us smarter.
Last week, Google unveiled a plethora of AI applications at the Google I/O 2025. The search giant which is briskly moving forward in AI advancements showed a range of possibilities with its new AI Mode in Search to its universal AI assistant – Project Astra. Talking about the things that most excite him from Google I/O 2025, Demis Hassabis, CEO of Google DeepMind, said, “If I had to pick a top three: Gemini 2.5 Pro DeepThink is a super frontier model on reasoning… Veo 3 is the first time we’ve combined audio and video… And Flash is probably going to surprise a lot of people.”
Hassabis placed Gemini 2.5 Pro DeepThink at the top of the list, terming it a ‘super frontier model on thinking’. He shared that Google’s Veo 3 is the most advanced video generation model ever. “It’s the first time we’ve combined audio and video together—and we’ve made big strides in improving video quality,” he said. The CEO rounded up his top picks by describing Gemini Flash as a faster, lightweight model for mobile and embedded devices. Hassabis also mentioned Gemini Diffusion, which is a research milestone in speed and image generation.
At Google I/O, perhaps one of the major highlights was Project Astra, which is an evolving AI assistant programmed to be proactive and multimodal and to be operational across phones and wearable devices. However, proactivity also comes with some challenges. “You want something to be helpful, not annoying,” Hassabis explained. “It’s a complex research problem, understanding when you’re busy, whether you’re speaking to the assistant or a human, even your physical context.” According to him, getting it right is critical for the universal assistant vision, especially as Google works towards memory-sharing across devices. “That’s firmly on the roadmap in the following months,” he confirmed.
AI as companions
Since big tech and AI startups are working towards making AI more personalised, the way people interact with these systems is bound to change. The Nobel Prize laureate acknowledged that users are likely to form bonds with their AI assistants. “It’s clear users want systems that know them well, understand their preferences, and carry on conversations from yesterday. But we’ll also have to think about things like upgrades, especially after people spend time training their assistant.”
He said that assistants could become indispensable not just for casual users but in professional workflows. When asked if overreliance on AI tools makes a user lazier or dumb, the CEO said he does not think of it that way. “It’s about teaching the next generation how to make the best use of these tools. They’re already part of education, so let’s embrace it and use it for better learning.”
Hassabis is particularly optimistic about the potential of AI in education, especially through Google’s LearnLM initiative. He told the host that with LearnLM, one could create flashcards on the fly, get suggestions on YouTube videos tailored to what they may be struggling with and even help them identify gaps in their understanding.
When asked what advice he would give to educators on tailoring curriculum around AI, not as a replacement but as a tool, Hassabis reasoned that curricula need to evolve rapidly. “Personalised learning where a student learns in class and continues at home with an AI tutor could be incredibly powerful.” He views AI as a tool to democratise education globally: “You could bring much higher-quality learning to poorer parts of the world that don’t have good education systems.”
Story continues below this ad
Coding in the Age of AI
During the interview, Cheung mentioned that one area where AI is dramatically impacting has been software development, especially with the emergence of tools like Jules and Vibe coding. With these tools, AI is writing most of the code. In this scenario, Hassabis was asked, What makes for a good developer?
The Google DeepMind executive responded, saying, “I think the next era will be a creative one… Top engineers will be 10x more productive because they’ll understand what the AI is doing and give better instructions. And hobbyists will get access to powerful tools previously out of reach.”
Hassabis went on to predict that natural language could become the next programming language. “When I started, I was coding in assembly. Then came C, Python… Now, natural language might be the final step.” On a similar tangent, if coding becomes easy, how will startups stay competitive? To this, Hassabis said that the competitive edge could come from “distribution, execution speed, or deep vertical integration with specialist data.”
Can AI cure disease?
Hassabis believes that hybrid AI systems will rise in importance, pointing at AlphaFold, the AI model that combines deep learning with biology and physics. The CEO, in a segment from 60 Minutes, claimed that AI may help cure all diseases in the next decade. When asked what he meant, he clarified that he meant AI could design hundreds of potential drugs. However, regulatory approval will still take time, but the possibilities are real.
Story continues below this ad
He explained that when early AI-designed drugs are validated and back-tested for safety and efficacy, regulations might evolve to trust AI predictions more. “We’ve done it before,” he said, referencing AlphaFold. “Mapping one protein used to take a PhD student five years. AlphaFold mapped 200 million in a single year. That’s a billion years of PhD time saved.”
In his short conversation, the Google DeepMind boss made it clear that AI is not just changing software; it is essentially redefining how we learn, work and treat disease. The 48-year-old British scientist is known to be a chess prodigy. He was knighted in 2023 for his services to AI. In 2024, Hassabis and John Jumper won the Nobel Prize in Chemistry for their work on AlphaFold, an AI system that predicts 3D protein structures. He co-founded DeepMind in 2010 with a mission to build Artificial General Intelligence (AGI). While most people perceive AGI to be smarter than humans, Hassabis defines it as systems that can do anything the human brain can do.
