我眼中的 AI 进化史
C
Chris
作者
AI 进化史
双眸、数据与空间
人工智能的演进绝非一蹴而就，而是一部由好奇心、海量数据与人类价值交织而成的壮丽史诗。
李飞飞教授在《我眼中的世界》里呈现的双重旅程——一个移民科学家的成长，以及计算机视觉的蜕变——恰好对应了现代 AI 发展的核心主线。
1
初光启蒙与物理的追问
1990s — 2005
个人里程碑
1999年毕业于普林斯顿大学物理系
2005年获得加州理工学院电机工程博士学位
2005年获得加州理工学院电机工程博士学位
AI 时代背景
此时的 AI 正处于"连接主义"与"符号主义"暗暗较量的后冬时期。李飞飞在神经科学实验（如猫的视觉皮层电信号重建）的启发下，坚信"视觉"是认知的入口，并将目光投向了让计算机"看见"这一终极命题。
2
ImageNet 与深度学习的奇点时刻
2006 — 2012
个人里程碑
2006年起着手构思并推进 ImageNet 数据库的建设
2009年在普林斯顿任教期间正式发布这一包含上千万张标注图片的数据库
2009年在普林斯顿任教期间正式发布这一包含上千万张标注图片的数据库
AI 时代背景
在全行业都在卷算法优化时，李飞飞敏锐地意识到"数据"才是训练神经网络的燃料。这一孤注一掷的信念在 2012 年的 ImageNet 挑战赛上迎来了奇点：Geoffrey Hinton 团队的 AlexNet 凭借卷积神经网络（CNN）与 GPU 算力一举夺魁，彻底引爆了持续至今的深度学习革命。
3
产业跃迁与人文反思
2016 — 2019
个人里程碑
2016–2018年出任谷歌云（Google Cloud）AI/ML 首席科学家
2019年与哲学家、经济学家等多学科专家共同创立斯坦福大学以人为本人工智能研究院（Stanford HAI）
2019年与哲学家、经济学家等多学科专家共同创立斯坦福大学以人为本人工智能研究院（Stanford HAI）
AI 时代背景
AI 开始从学术殿堂走向千行百业。与此同时，算法偏见、隐私泄露和伦理失范等问题抬头。李飞飞大声疾呼：
"AI 必须是一门以人为本的科学"——技术应当服务并增强人类，而非取代或削弱人性。
4
从生成式大模型到"空间智能"的重塑
2022 — 2026
个人里程碑
2023年出版自传《我眼中的世界》
2024年成立专注于空间智能与物理世界理解的 AI 初创公司 World Labs 并担任 CEO
2024年成立专注于空间智能与物理世界理解的 AI 初创公司 World Labs 并担任 CEO
AI 时代背景
2022年底 ChatGPT 的爆发开启了大语言模型（LLM）狂飙的时代。而到了2026年，AI 的前沿趋势正在从单纯的"文字与像素生成"，迈向具身智能（Embodied AI）和空间智能（Spatial Intelligence）。正如李飞飞所倡导的，AI 不仅要理解语言，更要像人类一样，在三维物理世界中生存、感知、推理和交互。
In her memoir The Worlds I See, Dr. Fei-Fei Li illustrates two parallel journeys—the coming-of-age of an immigrant scientist and the metamorphosis of computer vision—which perfectly trace the spine of modern AI.
1
First Light & The Quest of Physics
1990s — 2005
Personal Milestone
1999 Graduated with a BA in Physics from Princeton
2005 PhD in Electrical Engineering from Caltech
2005 PhD in Electrical Engineering from Caltech
AI Landscape
During this post-winter era, AI was stuck in a quiet tug-of-war between connectionism and symbolism. Inspired by neuroscience (such as experiments reconstructing visual inputs from a cat's cortex), Li believed "vision" was the gateway to intelligence, dedicating her research to teaching computers how to "see."
2
ImageNet & The Singularity of Deep Learning
2006 — 2012
Personal Milestone
2006 Conceived the ImageNet project
2009 Officially launched this database of millions of labeled images at Princeton
2009 Officially launched this database of millions of labeled images at Princeton
AI Landscape
While the academic community focused primarily on tweaking algorithms, Li realized that "data" was the actual fuel required to scale neural networks. This contrarian bet paid off in 2012's ImageNet Challenge, where Geoffrey Hinton's team introduced AlexNet. Powered by Convolutional Neural Networks (CNNs) and GPUs, AlexNet crushed the competition and ignited the deep learning revolution.
3
Industrial Integration & Human-Centered Reflection
2016 — 2019
Personal Milestone
2016–2018 Chief Scientist of AI/ML at Google Cloud
2019 Co-founded Stanford Institute for Human-Centered AI (Stanford HAI)
2019 Co-founded Stanford Institute for Human-Centered AI (Stanford HAI)
AI Landscape
AI transitioned from academic labs into global enterprise. At the same time, issues of algorithmic bias, privacy, and systemic risk emerged. Li became a leading voice advocating that:
"AI must be human-centered" — technology should augment human capabilities rather than diminish them.
4
From Generative LLMs to Spatial Intelligence
2022 — 2026
Personal Milestone
2023 Published memoir The Worlds I See
2024 Co-founded and became CEO of World Labs, focused on spatial intelligence
2024 Co-founded and became CEO of World Labs, focused on spatial intelligence
AI Landscape
The explosion of ChatGPT in late 2022 kicked off the LLM boom. By 2026, the frontier has expanded beyond generating text and 2D pixels to Embodied AI and Spatial Intelligence. As Li champions, AI must move beyond passive observation to active, 3D perception and reasoning—understanding the physical layout of the world just as humans do.