At a recent AI summit in New Delhi, Demis Hassabis, CEO of Google DeepMind, addressed the current limitations of artificial general intelligence (AGI). He stated that while advancements are being made, existing AGI systems still do not match human intelligence in three critical areas: continual learning, long-term planning, and consistency.
Hassabis emphasized that current AGI systems are restricted by their training, which prevents them from learning continuously. He described the ideal AGI as one that can adapt and learn from real-time experiences, tailoring its responses to specific situations. He also noted that existing systems struggle with long-term planning, capable only of short-term strategies, which hampers their ability to operate over extended periods like humans can.
Furthermore, he pointed out that these systems exhibit inconsistent performance. For instance, while they can excel in complex mathematical problems, they may still err on simpler tasks, reflecting a lack of the reliability expected from true general intelligence. This inconsistency was highlighted as a significant difference when compared to human experts, who typically do not make basic errors in their field.
Hassabis previously predicted that genuine AGI could be realized within five to ten years. His company, DeepMind, which he co-founded in 2010 and was acquired by Google in 2014, has been influential in advancing AI research and development. The summit featured prominent figures in the industry, including OpenAI’s Sam Altman and Meta's Alexandr Wang, further underscoring the ongoing discourse surrounding AGI and its implications for the future of technology.