Some of the world\u2019s best-known names in artificial intelligence descended on the small ski resort town of Davos, Switzerland, this week for the World Economic Forum (WEF) annual meeting.<\/p>\n
AI dominated many of the discussions among corporations, government leaders, academics, and nongovernmental groups. Yet a clear contrast emerged over how close current models are to replicating human intelligence and what the likely near-term economic impacts of the technology will be.<\/p>\n
The large language models (LLMs) that have captivated the world are not a path to human-level intelligence, two AI experts asserted in separate remarks at Davos.<\/p>\n
Demis Hassabis, the Nobel Prize\u2013winning CEO of Google<\/a> DeepMind<\/a>, and the executive who leads the development of Google\u2019s Gemini models, said today\u2019s AI systems, as impressive as they are, are \u201cnowhere near\u201d human-level artificial general intelligence, or AGI.<\/p>\n Yann LeCun\u2014an AI pioneer who won a Turing Award, computer science\u2019s most prestigious prize, for his work on neural networks\u2014went further, saying that the LLMs that underpin all of the leading AI models will never be able to achieve humanlike intelligence and that a completely different approach is needed.<\/p>\n Their views differ starkly from the position asserted by top executives of Google\u2019s leading AI rivals, OpenAI and Anthropic, who assert that their AI models are about to rival human intelligence.<\/p>\n Dario Amodei, the CEO of Anthropic, told an audience at Davos that AI models would replace the work of all software developers within a year and would reach \u201cNobel-level\u201d scientific research in multiple fields within two years. He said 50% of white-collar jobs would disappear within five years.<\/p>\n OpenAI CEO Sam Altman (who was not at Davos this year) has said we are already beginning to slip past human-level AGI toward \u201csuperintelligence,\u201d or AI that would be smarter than all humans combined.<\/p>\n Can LLMs lead to artificial general intelligence?<\/p>\n In a joint WEF appearance<\/a> with Amodei, Hassabis said that there was a 50% chance AGI might be achieved within the decade, though not through models built exactly like today\u2019s AI systems.<\/p>\n In a later, Google-sponsored talk, he elaborated that \u201cmaybe we need one or two more breakthroughs before we\u2019ll get to AGI.\u201d He identified several key gaps, including the ability to learn from just a few examples, the ability to learn continuously, better long-term memory, and improved reasoning and planning capabilities.<\/p>\n \u201cMy definition of [AGI] is a system that can exhibit all the cognitive capabilities humans can\u2014and I mean all,\u201d he said, including the \u201chighest levels of human creativity that we always celebrate, the scientists and artists we admire.\u201d While advanced AI systems have begun to solve difficult math equations and tackle previously unproved conjectures, AI will need to develop its own breakthrough conjectures\u2014a \u201cmuch harder\u201d task\u2014to be considered on par with human intelligence.<\/p>\n LeCun, speaking at the AI House in Davos, was even more pointed in his criticism of the industry\u2019s singular focus on LLMs. \u201cThe reason \u2026\u00a0LLMs have been so successful is because language is easy,\u201d he argued.<\/p>\n He contrasted this with the challenges posed by the physical world. \u201cWe have systems that can pass the bar exam, they can write code \u2026\u00a0but they don\u2019t really deal with the real world. Which is the reason we don\u2019t have domestic robots [and] we don\u2019t have level-five self-driving cars,\u201d he said.<\/p>\n