Dr. Fei-Fei Li discusses the evolution of vision, how human brains learn, and the importance of keeping AI human-centered.
要点 · TL;DR
视觉是智能的基石，大脑皮层一半的活动都涉及视觉功能。 Vision is the cornerstone of intelligence, with half of cortical activity involved in visual function.
现代 AI 的成功源于神经网络、大数据和 GPU 计算的融合，ImageNet 就是例证。 Modern AI's success stems from neural networks, big data, and GPU computing, as shown by ImageNet.
AI 应增强人类能动性而非取代，提升学习、健康和创造力。 AI should augment human agency, not replace it, enhancing learning, health, and creativity.
核心观点 · Key points
视觉是智能的基石，无论是在进化中还是在人工智能中，大脑皮层活动的一半都涉及视觉功能。 Vision is a cornerstone of intelligence, both in evolution and in AI, with half of cortical activity involved in visual function.
现代人工智能的成功源于神经网络、大数据和GPU算力的融合，ImageNet就是例证。 Modern AI's success stems from the convergence of neural networks, big data, and GPU computing, as exemplified by ImageNet.
人工智能从海量数据中学习模式，但人类的直觉和情感是高度个人化的，常常无法被AI获取。 AI learns from patterns in massive data, but human intuition and emotion are deeply personal and often inaccessible to AI.
人工智能应增强人类能动性，而非取代它，我们必须确保它促进学习、健康和创造力。 AI should augment human agency, not replace it, and we must ensure it enhances learning, health, and creativity.
在人工智能讨论中，教师和家长常被忽视，但他们对于引导下一代至关重要。 Teachers and parents are often forgotten in AI discussions, yet they are crucial for guiding the next generation.
科学发现是人工智能的一个有前景的领域，能够实现跨学科综合并加速医学突破。 Scientific discovery is a promising area for AI, enabling cross-disciplinary synthesis and accelerating medical breakthroughs.
反共识 · Contrarian takes
人工智能的创造力，如AlphaGo的第37步，是真实的，但仅限于规则明确的领域。 AI's creativity, like AlphaGo's Move 37, is real but limited to well-defined domains with clear rules.
互联网捕捉了人类行为，但并非所有人类认知；深度个人化的思想对AI而言仍不可及。 The internet captures human behavior, but not all human cognition; deeply personal thoughts remain inaccessible to AI.
AI在某些诊断任务上可以超越医生，但它缺乏人类的具身共情和生活经验。 AI can outperform doctors in some diagnostic tasks, but it lacks the embodied empathy and lived experience of humans.
我们不应拒绝学生使用AI工具；相反，我们必须教他们用AI增强学习和能动性。 We should not deny students AI tools; instead, we must teach them to use AI to enhance learning and agency.
AI的未来不仅仅是语言；空间和物理智能将是下一个前沿，实现具身AI。 The future of AI is not just language; spatial and physical intelligence will be the next frontier, enabling embodied AI.
关于AI的公共讨论被极端的末日论和乌托邦主义所扭曲；我们需要平衡的、多方参与的对话。 Public discourse on AI is skewed by extreme doomerism and utopianism; we need balanced, multi-stakeholder conversations.
本期章节 · Chapters（共 34）
开场独白与书籍推广Opening Monologue and Book Promotion
视觉作为智能基石Vision as Cornerstone of Intelligence
神经科学与视觉对AI的贡献Neuroscience and Vision's Contribution to AI
拐点与人类错误Inflection Point and Human Error
赞助商插播Sponsor Break
超越视觉：声音与自然语言处理Beyond Vision: Sound and NLP
儿童学习与概率Child Learning and Probability
物体恒常性与AIObject constancy and AI
AI与猫动作的合理性AI and the plausibility of cat movements
思想与创造力：AI的差距Thoughts and creativity: the gap in AI
互联网作为人类行为库The internet as a library of human behavior
AI创造力与第37手AI Creativity and Move 37
未来脑机接口Future Brain-Computer Interfaces
AI作为能动性工具AI as a Tool for Agency
赞助商插播：AG1与ElementSponsor Break: AG1 and Element
技术作为连接器Technology as Connectors
公共传播与教育Public Communication and Education
AI在医学与科学发现中AI in Medicine and Scientific Discovery
AI在医学诊断中AI in Medical Diagnosis
直觉与AIIntuition and AI
直觉与上下文Intuition and Context
人类与AI的差异Human vs AI Differences
监管与多方治理Regulation and Multi-stakeholder Governance