Google DeepMind CEO Demis Hassabis described current artificial intelligence as “jagged” and “frozen” Wednesday. Speaking at the India AI Impact Summit 2026 in New Delhi, the Nobel laureate argued that reaching true superintelligence requires solving critical architectural gaps.
Add businessleague.in as a Preferred Source
While modern systems like Gemini outperform humans on many benchmarks, Hassabis noted they remain inconsistent. He emphasized that the road to Artificial General Intelligence (AGI) depends on moving past narrow task-solving toward long-term reasoning. The summit, held at Bharat Mandapam, serves as the first major global AI gathering hosted in the Global South.
Also Read |Tamil Nadu Voter List Purge: 97 Lakh Names Deleted in SIR Phase 1
The “Jagged Intelligence” Problem
Hassabis introduced the term “jagged intelligence” to describe the uneven performance of today’s models. He pointed out that systems can win gold medals in the International Maths Olympiad yet fail at elementary arithmetic if a question is rephrased. A true general intelligence would not exhibit such sharp inconsistencies, he noted Tuesday.
Meanwhile, current models are effectively “frozen” after their initial training phase. Therefore, they cannot learn continuously from real-world experiences or adapt dynamically to new contexts once deployed. Hassabis argued that AGI needs “continual learning” to mirror human cognitive growth.
AGI Timeline: The 2030 Threshold
In a bold prediction, Hassabis stated that AGI is on the horizon within the next five to eight years. This places the arrival of human-level intelligence by roughly 2030. He described the current era as a “threshold moment” where AI is transitioning into autonomous, agentic systems.
Next, Hassabis envisioned a “Golden Era” for scientific discovery. He cited AlphaFold as the proof of concept for using AI to solve complex natural-world problems. By accelerating research in climate modeling and material science, AI will act as a “force multiplier” for human experts over the next decade.
Also Read |Tamil Nadu Voter List Purge: 97 Lakh Names Deleted in SIR Phase 1
Dual-Purpose Risks and Bio-Security
Hassabis also warned of the “dual-purpose” risks associated with advancing intelligence. He highlighted the threat of bad actors repurposing autonomous systems for harmful ends, specifically in biosecurity and cybersecurity. “We need to make sure cyber defenders are more powerful than the attack vectors,” Hassabis said.
Therefore, he called for urgent international cooperation to build technical guardrails. As AI becomes more agentic, ensuring these systems perform exactly as expected is a primary safety challenge. The DeepMind CEO stressed that international dialogue is the first step toward mitigating these long-term technical hurdles.
Reality Check
Hassabis claims AGI is only five to eight years away. Still, his own “jagged intelligence” label suggests the fundamental engineering hurdles remain immense. Therefore, the 2030 timeline may be more of an aspirational target for investors than a confirmed hardware roadmap. In fact, many researchers argue that “System 2” reasoning—which Hassabis says is missing—requires a total shift away from current transformer-based architectures.
The Loopholes
The “continual learning” Hassabis describes is currently limited by massive compute costs. In fact, retraining a model on-the-fly in a personalized way would require energy resources that most companies cannot afford. Therefore, “frozen” models remain the industry standard not by choice, but by economic necessity. Still, Google continues to push the “agentic” narrative to justify the high subscription costs of its advanced AI suites.
Also Read |Tamil Nadu Voter List Purge: 97 Lakh Names Deleted in SIR Phase 1
What This Means for You
If you use AI for professional tasks, understand that your tools are still “jagged.” First, double-check AI-generated math or code, even if the model sounds authoritative. Then, start experimenting with “agentic” workflows where AI manages multi-step tasks rather than just answering questions.
Finally, realize that the 2030 AGI threshold will fundamentally redefine the job market. You should focus on cross-disciplinary skills and high-level strategy, as these remain the hardest for “jagged” systems to replicate. Before the end of the year, expect to see the first “breakout moments” in robotics as Google integrates foundation models into physical hardware.
What’s Next
Google DeepMind is expected to release a paper on “continual learning” architectures by late 2026. Then, the next phase of Gemini will likely feature improved long-term planning modules for enterprise users. Finally, the India AI Impact Summit will conclude on February 20 with a final report on global AI safety standards.
Also Read |Tamil Nadu Voter List Purge: 97 Lakh Names Deleted in SIR Phase 1
End….
Add businessleague.in as a Preferred Source
