Andrew Ng

• 1. Introduction

  ›

  Andrew Yan-Tak Ng was born in London on the 18th of April 1976, to a hematologist father and an arts administrator mother who had both immigrated from Hong Kong. Before he was a teenager, his family had already moved twice: first back to Hong Kong, then to Singapore in 1984. That restlessness, that crossing of worlds, would define how he saw knowledge itself.

  By the time Ng reached his thirties, over a hundred thousand people had enrolled in a single online course he designed. A neural network he helped build learned to recognize cats by watching YouTube videos, with no human ever labeling the images. He cofounded Google Brain, led AI research at one of China's largest technology companies, and taught more than 8 million students. The questions worth sitting with are: how does a researcher end up reshaping both the science and the teaching of artificial intelligence at the same time?

• 2. Formation of a Scientist

  ›

  Raffles Institution in Singapore educated Ng before he crossed the Pacific to Carnegie Mellon University in Pittsburgh, Pennsylvania, where in 1997 he completed an undergraduate degree with a triple major in computer science, statistics, and economics. Between 1996 and 1998, while still a student, he was conducting research on reinforcement learning, model selection, and feature selection at AT&T Bell Labs.

  At MIT, where he earned a master's degree in Electrical Engineering and Computer Science in 1998, Ng built the first publicly available, automatically indexed web-search engine for research papers. It was a direct precursor to CiteSeerX but was specialized in machine learning from the start. His doctoral work at Berkeley, completed in 2002 under the supervision of Michael I. Jordan, produced a thesis titled "Shaping and policy search in reinforcement learning" that remains well-cited.

  During that same graduate period, Ng co-authored the paper that introduced latent Dirichlet allocation alongside David M. Blei and Michael I. Jordan, a technique that became one of the most widely used tools in natural language processing.

• 3. Stanford and the GPU Gamble

  ›

  Ng joined Stanford as an assistant professor in 2002 and became an associate professor in 2009, directing the Stanford Artificial Intelligence Laboratory. His machine learning course CS229 became the most popular course offered on the Stanford campus, with over 1,000 students enrolling in some years.

  In 2008, his research group at Stanford was among the first in the United States to advocate using GPUs for deep learning. The argument was that efficient computing infrastructure could accelerate the training of statistical models by orders of magnitude. At the time, it was considered a controversial and risky call. GPUs have since become a cornerstone of the entire field.

  Around the same time, Ng launched the Stanford Engineering Everywhere program, which published Stanford courses online for free. He taught a machine learning course as part of this program, modeled on the MIT OpenCourseWare approach but aimed at providing a more complete course experience with lectures, materials, problems, and solutions. Between 2009 and 2011, several hundred hours of lecture videos recorded by Stanford instructors were uploaded, with Ng and others drawing inspiration from Sal Khan of Khan Academy and the forum design of Stack Overflow.

• 4. Building Google Brain

  ›

  From 2011 to 2012, Ng worked at Google, where he cofounded and directed the Google Brain Deep Learning Project alongside Jeff Dean, Greg Corrado, and Rajat Monga. The project built large-scale artificial neural networks using Google's distributed computing infrastructure.

  Among its most striking results: a neural network trained on 16,000 CPU cores learned to recognize cats after watching YouTube videos, having never been told what a cat is. That experiment captured widespread attention and illustrated what deep learning at scale could do without explicit human supervision. The same underlying technology later found its way into the speech recognition system in the Android operating system.

  Ng's push to build neural networks at this scale reflected his view, articulated since 2017, that shifting to high-performance computing was essential to accelerate progress in the field.

• 5. Baidu, Departures, and New Ventures

  ›

  In 2014, Ng joined Baidu as chief scientist, where he set up research teams working on facial recognition and a healthcare AI chatbot called Melody. He also developed DuerOS, an AI platform that positioned Baidu prominently in the global conversation around AI development. He resigned in March 2017.

  After leaving Baidu, Ng launched DeepLearning.AI, an online series of deep learning courses that includes the AI for Good Specialization. He also founded LandingAI, which provides AI-powered software products to enterprises. In January 2018, he unveiled the AI Fund, raising $175 million to back new startups. LandingAI went on to secure a $57 million series A funding round in November 2021, led by McRock Capital.

  In October 2024, the AI Fund made its first investment in India, backing Jivi, a healthcare startup using AI for diagnoses, treatment recommendations, and administrative tasks. Amazon announced Ng's appointment to its board of directors on the 11th of April 2024.

• 6. The MOOC Movement

  ›

  In October 2011, the applied version of Stanford's machine learning class launched at ml-class.org with over 100,000 students registered for its first edition. Two companion courses on databases and AI launched alongside it. The platform running the machine learning and database courses was built by students, including Frank Chen, Jiquan Ngiam, Chuan-Yu Foo, and Yifan Mai. Over 40,000 statements of accomplishment were awarded across the three courses, which each ran for ten weeks.

  Ng cofounded Coursera with Stanford computer scientist Daphne Koller in 2012. Koller brought her work on blended learning and peer-grading systems; others inside Stanford contributed a learning management system, a method for using machine learning to sync lecture videos, and the ClassX platform. The AI course in the 2011 Stanford trio was taught by Sebastian Thrun, and it directly led to the founding of Udacity.

  By 2023, an estimated 8 million people worldwide had taken Ng's courses through DeepLearning.AI and Coursera. As of 2020, three of the most popular courses on Coursera were his: Machine Learning ranked first, AI for Everyone ranked fifth, and Neural Networks and Deep Learning ranked sixth.

• 7. Views on Regulation and Risk

  ›

  Ng has consistently argued that the most serious challenge posed by AI is not the rise of dangerous machines but the disruption of labor. His framing: rather than focusing on science-fiction scenarios, academia, industry, and government should hold a serious conversation about the future of work.

  In a December 2023 interview with the Financial Times, Ng raised concerns about regulations targeting open-source AI, warning that reporting, licensing, and liability requirements could unfairly burden smaller firms and slow innovation. He argued that regulating basic technologies like open-source models could hinder progress without meaningfully improving safety.

  In a June 2024 Financial Times interview, Ng criticized proposed AI legislation in California that would have required developers to implement a kill switch for advanced AI models. He described the bill as creating massive liabilities for science-fiction risks and said it stokes fear in anyone daring to innovate. The bill was vetoed by Governor Gavin Newsom in September 2024. Ng writes a weekly digital newsletter called The Batch to continue sharing his views on AI directly with readers.