**图灵奖得主Yann LeCun对通用人工智能持怀疑态度，强调AI距离人类智力还有距离**
在图灵奖获得者兼Meta AI负责人的Yann LeCun最新公开发言中，他对人工智能的发展提出了独到的见解。在参与Meta伦敦的一次活动时，LeCun表示，通用人工智能（AGI）这一概念可能并不存在。他认为目前所称的“通用人工智能”并不具备真正意义上的普遍性，人类智能远比目前的人工智能系统复杂。
LeCun指出，目前的人工智能系统面临四个认知挑战：推理、规划、持久记忆和理解物理世界。这些能力的缺失限制了AI的应用范围，并可能导致错误频发。以自动驾驶汽车为例，它们在公共道路上的表现依然面临挑战。此外，家用机器人在执行基本家务时的困难以及智能助手只能完成有限的任务，也说明了AI的现实局限。
关于大型语言模型（LLMs），LeCun强调了其局限性。虽然LLMs在语言流畅性方面表现出色，但它们对现实世界的理解仍然非常有限。他指出，人类通过与环境的实际互动获取知识和经验，而非仅仅通过文本。令人震惊的是，LeCun估计一个四岁孩子所接触的数据量是当前世界上最大的LLMs的50倍。这意味着即使是现有的大规模模型也无法与人类的感知和学习速度相比。
对于未来的人工智能发展，LeCun更倾向于追求“人类水平的AI”，他认为真正的挑战在于如何让人工智能系统更深入地理解现实世界，并不仅仅是模仿人类的语言模式。他的观点和看法无疑为人工智能领域的研究者和开发者提供了新的思考方向。
英语如下：
News Title: Turing Award Winner LeCun Talks about AI Limitation: Large Models Fail to Match Human Intelligence
Keywords: AI Expert View, Limitations of Large Models, Comparison with Human Intelligence
News Content: **Turing Award Winner Yann LeCun Expresses Skepticism on AGI, Stressing the Distance between AI and Human Intelligence**
Turing Award winner and Meta AI leader, Yann LeCun, has shared his unique insights on the development of artificial intelligence. During an event at Meta’s London office, LeCun expressed怀疑关于通用人工智能（AGI）的概念，认为目前所称的“通用人工智能”并不具备真正的普遍性，人类智能远比当前的人工智能系统复杂。
LeCun pointed out that current AI systems face four cognitive challenges: reasoning, planning, long-term memory, and understanding the physical world. The lack of these abilities limits the application of AI and may lead to frequent errors. Taking autonomous vehicles as an example, their performance on public roads still faces challenges. In addition, the difficulties faced by household robots in performing basic household tasks and the limited tasks that smart assistants can complete also illustrate the practical limitations of AI.
Regarding large language models (LLMs), LeCun highlighted their limitations. While LLMs excel in language fluency, their understanding of the real world is still very limited. He pointed out that humans acquire knowledge and experience through actual interaction with the environment, not just through text. Surprisingly, LeCun estimated that a four-year-old child’s data exposure is 50 times that of the largest LLMs currently in existence. This means that even the existing large-scale models cannot compare to humans in terms of perception and learning speed.
Looking ahead to the development of AI in the future, LeCun prefers to pursue “human-level AI,” believing that the real challenge lies in how to make AI systems understand the real world more deeply, and not just imitate human language patterns. His views and opinions undoubtedly provide new directions for researchers and developers in the field of artificial intelligence.
【来源】https://thenextweb.com/news/meta-yann-lecun-ai-behind-human-intelligence
Views: 3