Google AI Chief: True AGI Remains Elusive, Hassabis Warns

Demis Hassabis, CEO of Google DeepMind, has warned that true Artificial General Intelligence is still years away. In his view, current AI systems fail to adapt in real-time, struggle with planning over decades, and exhibit uneven performance across tasks, unlike human experts. These gaps, he argues, keep AGI out of reach despite rapid advances in machine learning.

He points to core limitations in continual learning, long-term reasoning, and consistency. While an AI model may master a particular task after training, it often cannot transfer that knowledge smoothly to new or evolving situations, and its performance can degrade when faced with unfamiliar challenges. Such shortfalls complicate the prospect of an AI that can think, reason, and plan across a broad range of activities like a human expert.

Despite the optimism surrounding AI progress, Hassabis stresses that surpassing narrow intelligence to achieve a truly general system requires breakthroughs in how machines learn and reason over time. He also notes that solving complex problems—such as Olympiad-level math—does not by itself prove general intelligence, and such systems can still err or become “frozen” after training. The debate remains heated, with some researchers arguing for near-term milestones while others advocate measured, long-term development.

The broader tech community has described the current era as a threshold moment for AI, with expectations of significant breakthroughs in the coming years. However, the path to genuine AGI will likely hinge on advances in continual learning, reliable long-range planning, and robust cross-task adaptability, underscoring the need for careful governance and cooperative global effort as the technology evolves.

#AGI #AIProgress #DeepMind