‘AI that helps you in the real world’: Google’s DeepMind boss on merging Gemini and Veo AI models

Google

Google is dreaming big when it comes to artificial intelligence. And now, it’s taking a major step toward building AI that actually understands the physical world — by teaching it through YouTube videos.

In a recent episode of the Possible podcast, Google DeepMind CEO Demis Hassabis explained how the company plans to combine its two powerful AI tools: Gemini, which is built to understand and generate text, images, and audio, and Veo, which creates videos. Together, they could form a super-smart assistant that doesn’t just answer questions, but actually understands how the world works.

“We’ve always built Gemini to be multimodal from the beginning,” said Hassabis. “And the reason we did that is because we have a vision for this idea of a universal digital assistant — one that actually helps you in the real world.”

In simpler terms, Google wants to create an AI that doesn’t just chat or draw pictures — but watches, listens, and learns from everything, including videos.

And yes, a big chunk of that video training is likely coming from YouTube. Since Google owns the platform, it has access to an enormous library of content showing real-life activities — cooking, building, sports, science experiments, you name it. “Basically, by watching YouTube videos — a lot of YouTube videos — [Veo 2] can figure out, you know, the physics of the world,” Hassabis said.

The goal? Smarter AI that can “see” and interpret the world more like humans do. Think of an assistant that doesn’t just give you a recipe but understands how ingredients behave when cooked, because it has seen thousands of videos showing that in action.

Google isn’t alone in this race. OpenAI and Amazon are also working on similar “omni” models — AI systems that can handle anything from text to images to sounds, all at once.

Of course, all of this requires massive amounts of data. Google has said some of its models may be trained on YouTube content, depending on creator agreements. The company even updated its terms of service last year to allow broader use of content for training AI.