The New Race Toward Human-Level Intelligence
Artificial intelligence is advancing at a pace few could have predicted even a handful of years ago. With every new model, every breakthrough, and every technological leap, the industry gets closer to a milestone that once belonged only in science fiction: artificial general intelligence, or AGI. This refers to an AI capable of understanding, learning, and solving problems across nearly any domain—much like a human mind.
At the center of this global race stands Google DeepMind, long recognized as one of the world’s leading AI research organizations. Following the highly praised release of Gemini 3, its most powerful model so far, DeepMind’s CEO Demis Hassabis made a bold and attention-grabbing statement at the Axios AI+ Summit in San Francisco: if the world truly wants AGI, then AI systems must be scaled to their maximum potential.
His message was straightforward but striking. According to Hassabis, scaling is not just one of many strategies. It may be the strategy. The backbone—or even the entirety—of how AGI will emerge.
In this article, we break down what he meant, why scale matters, and what this philosophy could mean for the future of AI and society.
Why Scaling Is at the Heart of DeepMind’s Strategy
Understanding What Scaling Really Means
In AI, scaling refers to increasing the size of a model—its dataset, its number of parameters, and the compute power used to train it. Bigger models tend to learn more patterns, reason more effectively, and perform more tasks with surprising versatility.
Over the past decade, a clear pattern has emerged: when you make models bigger, they suddenly unlock unexpected abilities. These are often called emergent behaviors. Systems start reasoning better, solving harder tasks, analyzing longer contexts, and performing more complex logic.
Hassabis argues that this pattern has been too consistent to ignore. If larger models continuously unlock new capabilities, then pushing scaling further may eventually unlock something far more profound: the moment AI transitions from narrow skills to general intelligence.
Why Hassabis Believes Scaling Could Produce AGI
Hassabis has spent decades exploring both biological and artificial intelligence. His view is that AGI does not necessarily require brand-new scientific discoveries. Instead, it may be a natural result of continuing to expand today’s model architectures and training systems.
In other words, we might already have the ingredients. We simply haven’t pushed them far enough.
He suggested that the same forces driving today’s breakthroughs—massive datasets, enormous compute clusters, and extremely large neural networks—could eventually form the entire foundation of AGI.
This is a controversial stance in the AI research world. Many experts believe scaling alone won’t get us there. But Hassabis is convinced that scaling is not only part of the answer; it might actually be the whole answer.
Gemini 3: A Glimpse of What Scaling Can Achieve
DeepMind’s latest model, Gemini 3, delivered significant improvements in reasoning, multimodal abilities, and real-world problem solving. Its success strengthened Hassabis’s argument: each jump in scale tends to deliver leaps in capability.
To him, Gemini 3 isn’t just a product. It’s evidence.
As models grow, they begin to take on characteristics closer to human-level cognition, including:
- Better long-term planning
- Stronger abstract reasoning
- Deeper understanding of nuanced language
- Improved ability to handle complex tasks
- More reliable decision making
Hassabis sees these developments not as isolated upgrades but as milestones on a single, continuous path toward AGI.
The Global Race: Why Speed and Scale Matter Now
Competition Is Pushing Boundaries at Record Pace
The AI world is more competitive than ever. Companies like OpenAI, Meta, Anthropic, xAI, and others are racing to build the most powerful models. What once happened over decades now happens over months.
Hassabis recognizes that this environment does not allow for hesitation. Innovation is accelerating, and any organization that slows down risks falling behind. Scaling, he argues, is both a scientific strategy and a competitive necessity.
The Stakes Are Higher Than Ever
The potential of AGI is immense. It could transform medicine, science, education, transportation, communication, and virtually every industry. The organization that achieves AGI first could shape the future of global technology and set standards for how intelligent systems are controlled and deployed.
DeepMind wants to be at the forefront of that transformation. And Hassabis believes scale is the way to get there.
The Debate: Can Scaling Alone Truly Produce AGI?
Not everyone in the AI field agrees with Hassabis. Some argue that large models still lack essential cognitive features such as robust reasoning, memory, and understanding of cause and effect. They believe that new architectures or hybrid systems may be needed.
Despite this, scaling skeptics cannot deny the extraordinary progress seen over the last few years. Many breakthroughs that were once dismissed as impossible have already emerged from simply making models larger and training them more effectively.
Hassabis acknowledges the debate but remains confident that scaling is the main driver. He believes the next stages of AI evolution will come from pushing current techniques to their absolute limits.
The Challenges Ahead: Cost, Compute, and Responsibility
The Enormous Cost of Scaling
Training state-of-the-art models is extremely expensive. As models grow, so does their appetite for computational resources. Building AGI-level systems may require colossal investments in data centers, energy, and specialized hardware.
The question is whether even the most well-funded companies can sustain this pace indefinitely.
Safety and Control
With great capability comes enormous responsibility. As AI systems become more powerful, concerns about misuse, unintended behavior, and safety continue to rise.
Hassabis emphasizes that scaling must go hand in hand with strong ethical frameworks and robust safety measures. In his view, moving fast is necessary, but moving responsibly is essential.
What This Means for the Future of AI
If Hassabis is correct, then we may be much closer to AGI than most people realize. Rapid scaling could unlock breakthroughs within years, not decades.
The next generation of AI systems could:
- Perform scientific research
- Design new medicines
- Automate complex industries
- Assist with large-scale decision making
- Generate new creative solutions to global problems
Whether or not scaling alone achieves AGI, it is clear that it will shape the future of AI development. The push to build ever-larger, ever-smarter systems is not slowing down. And DeepMind’s CEO wants the world to understand exactly why.
The Road to AGI May Be a Matter of Scale
Demis Hassabis has placed a bold bet on the future of artificial intelligence. His belief in the power of scaling reflects years of research, thousands of experiments, and the phenomenal progress of models like Gemini 3.
In his view, the path to AGI is not mysterious. It is straightforward, measurable, and technological. Build bigger models. Train them longer. Give them more compute. Push them to their limits. And somewhere along that path, something extraordinary may emerge.
Whether the world agrees or not, one thing is certain: the race toward AGI is accelerating, and DeepMind plans to lead it.
