During a session at the India AI Impact Summit, Demis Hassabis, CEO of Google DeepMind, stated that the world may be entering a golden era of scientific discovery, fuelled by rapid advances in artificial intelligence. He discussed how the human brain served as the inspiration for modern AI systems. He also stated that one of the most significant advances in the field has been reinforcement learning, a technique that allows machines to learn from experience rather than relying solely on pre-programmed instructions.
According to him, allowing systems to learn directly from data has greatly increased their power and adaptability. At the same time, he stated that AI research has revealed how efficient the human brain is in terms of learning without consuming an infinite amount of information.
Hassabis stated that the industry is now entering what many refer to as the ‘agentic’ era, in which AI systems become more autonomous and capable of acting on their own. However, he cautioned that this progress carries serious risks. Concerns about misuse by bad actors, including nation states, as well as emerging biosecurity and cybersecurity threats, require immediate attention. He pointed out that strong safeguards and international cooperation will be critical as AI systems become more capable. He added that global forums like the India AI Impact Summit are important for aligning stakeholders on safety and governance.
Despite his optimism, Hassabis stated that artificial general intelligence (AGI) has not yet been achieved. He explained that current systems are typically extensively trained before being deployed in largely fixed form. What they lack is true continuous learning, or the ability to learn and adapt even after deployment. He also pointed out problems in long-term planning and consistency. While some AI models can solve extremely complex mathematical problems, they may still make basic mistakes depending on how a question is phrased. This uneven performance points out the difference between narrow expertise and genuine general intelligence, he added.
Still, Hassabis described the past decade of progress as remarkable and suggested that AGI could be on the horizon within the next five to eight years, if breakthroughs continue.