DeepMind's CEO said there are still 3 areas where AGI systems can't match real intelligence

The journey toward achieving true artificial general intelligence (AGI) is ongoing, but significant hurdles remain, as articulated by DeepMind's CEO, Demis Hassabis. During an AI summit held in New Delhi, he responded to inquiries about whether existing AGI systems can rival human intelligence. AGI represents a theoretical form of machine intelligence capable of reasoning and problem-solving in ways it hasn’t been specifically trained for. Hassabis conveyed a clear message: "I don't think we are there yet." He identified three critical areas where current AGI systems still fall short. The first issue is what he termed "continual learning." He explained that these systems are essentially static, limited by the training they underwent prior to deployment. Ideally, these systems would learn continuously from their experiences, adapting to their environments and personalizing responses based on specific tasks. The second limitation he pointed out is the lack of long-term planning capabilities. While current AGI can manage short-term tasks, they do not possess the ability to plan for extended periods, a skill that humans naturally have. Lastly, Hassabis mentioned the inconsistency of these systems. They may excel in specific domains, such as performing exceptionally well in complex mathematical challenges, yet falter on basic arithmetic problems when posed in certain ways. He emphasized that a truly general intelligence should not exhibit such uneven capabilities. In a previous interview with "60 Minutes," Hassabis predicted that genuine AGI could emerge within the next five to ten years. Since co-founding DeepMind in 2010, which was later acquired by Google in 2014, Hassabis has been at the forefront of AI research, contributing to significant advancements, including the development of Google's Gemini. His recent accolade includes a joint Nobel Prize in Chemistry for his innovations in protein structure prediction. The discourse around AGI remains vibrant and contentious within Silicon Valley. For instance, Databricks CEO Ali Ghodsi expressed at a recent conference that many current AI chatbots already fulfill the definition of AGI, while others in the industry seem to keep moving the benchmarks toward a more advanced form of intelligence that could surpass human cognitive abilities. The AI Summit in India, which runs from Monday to Friday this week, features an impressive lineup of prominent figures from the tech and AI industries. Noteworthy speakers include Sam Altman, CEO of OpenAI; Dario Amodei, CEO of Anthropic; Sundar Pichai, CEO of Google; and Alexandr Wang, Meta's chief AI officer.