Artificial intelligence has moved from research labs to boardrooms at unprecedented speed, reshaping industries from finance to healthcare. Yet, according to Google DeepMind CEO Demis Hassabis, the world remains far from achieving artificial general intelligence (AGI)-the stage where machines can perform any intellectual task that humans can. Speaking at the India AI Impact Summit in New Delhi on February 18, Hassabis said that today’s systems, despite their impressive capabilities, still suffer from fundamental limitations that prevent them from functioning as truly general intelligence.

AGI is widely viewed as the ultimate milestone in AI development and a potential catalyst for profound economic transformation. Governments, technology giants, and venture capital investors have poured billions into research on the assumption that systems capable of reasoning, planning, and learning autonomously could unlock massive productivity gains. Hassabis’ comments suggest that expectations may need recalibration.

Current AI models excel at specific tasks but lack consistent performance across domains. He described this phenomenon as “jagged intelligence” -systems that can perform at extraordinary levels in certain areas while failing at comparatively simple tasks in others. For example, an AI model may solve complex mathematical problems at an Olympiad level yet make errors in basic arithmetic when questions are framed differently. This inconsistency, he said, is incompatible with the reliability expected from a general-purpose intelligence.

Another structural weakness lies in how modern AI systems learn. Most models are trained on vast datasets and then deployed in a largely static form.

Hassabis argued that true intelligence requires continual learning the ability to adapt dynamically based on real-world experience and context. Without this capability, AI systems cannot personalise effectively or evolve in response to changing environments. He noted that once deployed, today’s models are essentially “frozen,” limiting their usefulness in complex, real-time decision-making scenarios.

Long-term planning is another frontier where AI remains immature. While current systems can handle short-term tasks or multi-step reasoning over limited horizons, they struggle to plan coherently over extended periods. Human decision-making often involves strategies that unfold over years or decades, something present-day AI architectures are not designed to replicate. This limitation constrains applications in areas such as policy modelling, corporate strategy, and autonomous systems that require sustained foresight.

Creativity- particularly the ability to formulate original questions or hypotheses-also remains elusive. Hassabis pointed out that AI can serve as a powerful scientific assistant, as demonstrated by systems like AlphaFold, which revolutionised protein structure prediction. However, generating breakthrough ideas requires intuition, judgment, and taste-qualities that machines have yet to demonstrate convincingly. Solving a problem, he argued, is often easier than identifying the right problem to solve, and that distinction defines the difference between automation and true innovation.

Despite these shortcomings, Hassabis expressed optimism about the long-term trajectory of AI development. He suggested that the next decade could usher in a “golden era” of scientific discovery, driven by increasingly sophisticated tools that augment human researchers rather than replace them. His outlook indicates that while AGI remains distant, incremental advances could still deliver significant economic and societal benefits.

For India, his remarks carry particular significance. The country has positioned itself as a major participant in the global AI ecosystem, leveraging its large technology workforce and digital infrastructure. However, the gap between current capabilities and AGI implies that near-term growth will likely come from applied AI solutions-automation, analytics, and decision support-rather than fully autonomous systems. This could favour India’s IT services sector, which specialises in enterprise integration and implementation rather than frontier research.

Globally, the comments may temper speculative valuations in segments of the technology market that assume rapid breakthroughs in autonomous intelligence. Companies building infrastructure for AI deployment-cloud computing, semiconductors, data centres, and cybersecurity-could continue to benefit regardless of AGI timelines, while firms promising human-level automation may face longer development cycles and regulatory scrutiny.