美国国家工程院院士、斯坦福大学教授李飞飞近日在参加亚洲美国学者论坛时表示，实现真正的通用人工智能（AGI），需要超越二维图像处理的能力，进入三维空间智能的领域。她认为，现有的AI模型如Sora虽然能够生成视频内容，但本质上仍停留在二维平面，缺乏对三维空间的理解和操作能力。
李飞飞强调，空间智能是理解世界和让机器人执行任务的先决条件。她以Sora模型为例，指出该模型无法改变视角来展示同一场景的不同角度，这表明模型缺乏对三维空间的理解。她认为，真正的AGI需要能够理解物体之间的关系、三维空间中的几何形状，以及物体间的相互作用。
李飞飞表示，空间智能的应用范围广泛，包括增强现实（AR）、虚拟现实（VR）、机器人操作以及应用程序设计等。她认为，自然进化赋予了动物在三维空间中生活、预判和互动的能力，这是人类和其他动物能够生存的关键。
李飞飞的研究和创业活动表明，她对空间智能在AI领域的应用充满信心。她的公司World Labs在7月底宣布完成两轮融资，估值达到10亿美元。她希望通过她的研究，能够推动AI从“看到”到“做到”的转变，实现更加智能和实用的AI技术。
英语如下：
Title: “AI Matriarch” Li Fei-Fei: 3D Spatial Intelligence is the Key to Achieving AGI
Keywords: 3D Intelligence, AGI, Li Fei-Fei
Content: At the recent Asian American Scholars Forum, Li Fei-Fei, a member of the National Academy of Engineering and a professor at Stanford University, stated that to realize true general artificial intelligence (AGI), it is necessary to transcend the ability to process two-dimensional images and enter the domain of three-dimensional spatial intelligence. She argues that existing AI models such as Sora, while capable of generating video content, fundamentally remain in the realm of two-dimensional planes, lacking the understanding and manipulation capabilities of three-dimensional space.
Li Fei-Fei emphasizes that spatial intelligence is a prerequisite for understanding the world and for robots to perform tasks. She uses the Sora model as an example, pointing out that the model is unable to change perspectives to show different angles of the same scene, indicating a lack of understanding of three-dimensional space. She believes that true AGI must be able to understand the relationships between objects, the geometric shapes in three-dimensional space, and the interactions between objects.
Li Fei-Fei notes that the applications of spatial intelligence are broad, including augmented reality (AR), virtual reality (VR), robotic operations, and application design. She points out that natural evolution has endowed animals with the ability to live, predict, and interact in three-dimensional space, which is a key to the survival of humans and other animals.
Li Fei-Fei’s research and entrepreneurial activities indicate that she is confident in the application of spatial intelligence in the AI field. Her company, World Labs, announced at the end of July that it had completed two rounds of financing with a valuation of $1 billion. She aims to push AI from being able to “see” to being able to “do” through her research, achieving more intelligent and practical AI technologies.
【来源】https://www.tmtpost.com/7193753.html
Views: 2