Home / Google DeepMind: Latest AI Advances in AGI, Drug Discovery, and Robotics

Google DeepMind: Latest AI Advances in AGI, Drug Discovery, and Robotics

By: Jasmin Espinas
May 7, 2025

Google DeepMind is a leading AI research lab focused on building systems that can learn, reason, and solve complex problems.

Recent updates highlight its work in general intelligence, drug discovery, music tools, and robotics. From AlphaFold 3 to the Gemini models, each project shows how DeepMind is applying AI in useful and practical ways.

This roundup gives a snapshot of where DeepMind is heading and what its latest research could mean for science, technology, and everyday life.

Google DeepMind Believes In the Potential of AGI

Google DeepMind sees artificial general intelligence (AGI) as a huge leap in the future of AI and technology. AGI refers to AI that can think, plan, and solve tasks at the level of a human. Some experts believe it could be developed in the coming years.

If AGI can act on its own and handle complex problems, it could improve how we manage health, education, and the environment. DeepMind believes AGI has the potential to bring practical solutions to issues that affect millions of people.

At the same time, the company recognizes that AGI must be built with care. Even a small error in how it’s designed or used could lead to serious problems. That’s why safety remains part of the process from the beginning.

Furthermore, AGI could support faster and more accurate medical decisions. It could also help students learn in ways that suit their needs. Smaller teams or groups could even take on difficult work using tools that were once limited to large institutions.

How Google DeepMind Plans to Keep AGI Safe

As AI becomes more powerful, the risks grow, too. Currently, some people already misuse AI to create harmful content or share false information. In the future, stronger systems could influence opinions or behaviors in ways that cause real harm.

Because of this, Google DeepMind is building stronger safety measures early on. One step includes limiting who can access certain features, especially those that could be used in cyberattacks.

The company is also testing several tools to reduce risks. These include advanced security systems that stop people from bypassing safety checks, rules that control how AI is used once it’s live, and research that helps spot risks before they grow. DeepMind recently introduced a cybersecurity testing framework to catch threats early and protect against misuse.

Even now, DeepMind checks models like Gemini to see if they can cause harm. These regular checks help the team stay ahead of potential problems while the technology continues to grow.

DeepMind and Isomorphic Labs’ AlphaFold 3 Could Speed Up How New Medicines Are Made

In May 2024, Google DeepMind and Isomorphic Labs introduced AlphaFold 3, their latest AI model. Unlike earlier versions, this one goes beyond predicting how proteins fold. It can also predict how different molecules, like DNA, RNA, and small molecules that attach to proteins, interact with each other.

This makes it more useful for drug development. By seeing how molecules work together, scientists can find better ways to treat diseases faster and more accurately.

Isomorphic Labs, created by DeepMind to focus on drug discovery, recently raised $600 million from outside investors. The company believes AI can help find better treatments more quickly.

At a panel during the World Economic Forum Annual Meeting in Davos, DeepMind CEO Demis Hassabis explained that using AI for science offers more possibilities than just language tools. He also noted that some of the drugs designed with AI at Isomorphic Labs will also begin trials by the end of this year.

Lyria 2: Google DeepMind’s Improved AI Model Helps People Create Music

Google DeepMind recently introduced Lyria 2, an AI music model that can create clear, high-quality audio for Music AI Sandbox. This model can handle different styles and complex sounds while keeping the small details that make music feel real.

The team also built a tool called Lyria RealTime. This tool lets users shape music as it plays. Users can mix styles, adjust sounds, and build songs moment by moment with this feature. It’s useful for testing ideas quickly or making long, flowing tracks without starting over.

To ensure people use this tech responsibly, DeepMind adds a digital watermark to all music made with Lyria through SynthID. This watermark helps show which tracks were made using AI. It’s part of the company’s effort to build tools responsibly.

Google DeepMind’s Gemini Models Add New Tools for Robots and Developers

Google DeepMind has introduced new AI models designed to help robots understand and interact with the world more effectively.

The first model, called Gemini Robotics, builds on Gemini 2.0. It adds the ability to control physical actions, which means it can help robots move, see, and respond to their surroundings.

Next is Gemini Robotics-ER. This version can understand space and movement better, so developers can use it to run their own programs. It’s designed for tasks that involve real-world reasoning, like picking things up or moving around a space safely.

Both models make it easier for robots to handle practical tasks. DeepMind is already working with Apptronik to develop new humanoid robots using these tools. It’s also working with selected testers to improve Gemini Robotics-ER before it’s more widely available.

Alongside this, DeepMind previewed Gemini 2.5 Flash. It builds on the earlier 2.0 Flash model, but now includes stronger reasoning features while keeping things fast and affordable. Developers can choose when the model should think through a task, or keep it off for quick results. They can also set limits to balance quality, cost, and speed.

Gemini 2.5 Flash doesn’t immediately generate output. It pauses to think through complex questions, like solving math problems or analyzing research. This “thinking process” leads to better answers, especially on difficult prompts. In tests, it performed well, just behind DeepMind’s more advanced 2.5 Pro model.

You May Also Like: Studio vs. 1-Bed Unit: Which Is Right for You?

What Do Google DeepMind’s Latest Models Mean for the Future of AI?

Google DeepMind has released several new AI models focusing on science, creativity, robotics, and advanced reasoning. Each one helps AI solve more complex problems, work with physical tools, or support real-time tasks like music creation and coding.

These models show how fast AI is moving forward. From drug research to more intelligent robots, DeepMind’s work could shape how people use AI in health, education, and daily life.

To stay informed about these updates and where AI is heading next, subscribe to Financial Daily Update. Get the latest on DeepMind, tech news, and how AI is changing the world.