在周二的Meta伦敦活动中，图灵奖得主、Meta AI部门负责人Yann LeCun公开表达了他对通用人工智能（AGI）的怀疑态度。LeCun认为，人类智能的普遍性概念是错误的，因此追求“人类水平的AI”更为实际。他指出，当前的AI系统在推理、规划、持久记忆和理解物理世界等四个关键认知领域存在显著局限，这限制了它们的应用并可能导致错误。
LeCun特别提到了大规模语言模型（LLMs），尽管这些模型在语言流畅性上表现出色，但它们对现实世界的理解极其有限。他举例说明，自动驾驶汽车的安全性问题和家用机器人的家务处理能力，以及智能助手仅能完成基础任务的现状，都是AI能力不足的体现。他强调，人类通过与环境的互动学习，而不仅仅是阅读文本，这使得一个四岁孩子接触到的数据量可能比最大规模的LLMs还要多50倍。
LeCun的观点凸显了AI研究的挑战，即如何使AI系统不仅能处理语言，还能真正理解并适应复杂的世界。他的言论引发了对当前AI发展方向的深入反思，暗示了在追求人工智能与人类智能相匹敌的道路上，研究者需要更加注重模型的全面学习能力和对现实世界的理解。
英语如下：
**News Title:** “Turing Award Winner Yann LeCun: Large Models Fall Short of Human Intelligence, AI Faces Four Major Cognitive Challenges”
**Keywords:** Large Model Limitations, Turing Award Winner, AI Cognitive Challenges
**News Content:**
### Yann LeCun, Turing Award Winner, Questions Large Models’ Path to Human-Level Intelligence
During a Meta event in London on Tuesday, Yann LeCun, a Turing Award recipient and head of Meta’s AI division, publicly expressed skepticism about the feasibility of achieving artificial general intelligence (AGI). LeCun argues that the concept of human intelligence’s universality is misguided, making the pursuit of “human-level AI” a more realistic goal. He points out that current AI systems exhibit significant limitations in four key cognitive domains: reasoning, planning, long-term memory, and understanding the physical world, restricting their applications and potentially leading to errors.
LeCun specifically mentioned large language models (LLMs), which, despite their linguistic fluency, have a limited understanding of the real world. He illustrated this with examples, such as safety concerns with autonomous vehicles, the capabilities of household robots, and the basic tasks performed by intelligent assistants, all of which demonstrate AI’s limitations. He emphasized that humans learn through interaction with their environment, not just by reading text, suggesting that a four-year-old child might be exposed to 50 times more data than the largest LLMs.
LeCun’s perspective highlights the challenges in AI research, which center on enabling AI systems to not only process language but also genuinely understand and adapt to a complex world. His remarks prompt a deeper reflection on the direction of current AI development, implying that researchers should focus more on models’ comprehensive learning abilities and real-world comprehension in their quest to match human intelligence.
【来源】https://thenextweb.com/news/meta-yann-lecun-ai-behind-human-intelligence
Views: 2