Who is Andrew Ng?
Andrew Ng is a machine learning researcher, educator and entrepreneur. An adjunct professor of computer science at Stanford, he has combined university research, leadership of industrial teams and the creation of educational companies. [1]
Born in the United Kingdom to a family from Hong Kong, he grew up in Hong Kong and Singapore. This international background preceded university study in the United States and a career connecting research institutions, technology companies and distance education. [8]
Education at Carnegie Mellon, MIT and Berkeley
In 1997 he earned two degrees at Carnegie Mellon, in mathematics and computer science and in statistics, with an additional economics major. He completed an electrical engineering and computer science master’s at MIT in 1998. His curriculum vitae dates his Berkeley computer science doctorate to 2003, with the thesis Shaping and policy search in reinforcement learning. [9]
Robotics and learning from demonstrations
His university projects included STAIR, which sought to integrate AI capabilities in an assistant robot; Make3D, concerned with reconstructing three-dimensional scenes from one image; and systems for locomotion and robotic control. This programme linked perception, learning and action, rather than studying only the classification of previously collected data. [10]
With Pieter Abbeel, he presented an inverse reinforcement learning method in 2004. It started from expert demonstrations when writing a reward function was difficult. Under the model’s assumptions, it sought an action policy approaching the expert’s performance without requiring an exact reconstruction of the expert’s internal preferences. [11]
With Abbeel, Adam Coates and other collaborators, he developed experimental helicopters capable of aerobatic manoeuvres under computer control. The project combined pilot demonstrations with learning about vehicle behaviour. Its 2008 and 2010 publications documented methods and experiments, with shared research authorship and contributions from specialist pilots. [12]
Google Brain and large-scale visual learning
Ng was the founding lead of Google Brain. This stage brought his neural-network research into industrial infrastructure, before his subsequent role as Baidu’s vice-president and chief scientist. He also directed the Stanford AI Lab. [5]
In a 2012 paper, Quoc Le, Ng and their collaborators trained a large network on ten million unlabelled images. The system developed detectors sensitive to faces, cats and human bodies. The experiment examined which representations could be learned through data and distributed computation without providing a label for each visual concept. [13]
Baidu and speech recognition systems
Baidu announced his appointment as chief scientist in May 2014, alongside the opening of its Silicon Valley laboratory. Ng took charge of Baidu Research, with teams in California and Beijing. [14]
He coauthored Deep Speech, presented in 2014 with Awni Hannun and other researchers. The system learned speech recognition through recurrent networks and training across multiple GPUs, reducing manually designed components. The research used synthetic variations in training data to address noise and differences between recordings. [15]
Deep Speech 2, presented in 2015 with Dario Amodei and a large team, extended this approach to English and Mandarin recognition. Comparisons used specified evaluation datasets. [16]
In March 2017 he announced his departure from Baidu. His public explanation set out an intention to continue developing AI applications beyond large technology companies. The decision preceded the creation of new educational, business and investment initiatives. [17]
The open course of 2011
In 2011 he taught a Stanford machine learning course online that attracted more than 100,000 students. His subsequent catalogue covers machine learning, deep learning and introductions such as AI for Everyone and Generative AI for Everyone. The offering distinguishes technical training from programmes intended to explain general concepts and uses. [3]
Coursera and platform governance
His Coursera cofounder was Daphne Koller. [28]
He cofounded Coursera in 2012 and was co-chief executive until April 2014. He has chaired its board since 2014. This corporate governance responsibility is distinct from the platform’s day-to-day executive management. [4]
DeepLearning.AI and new skills
He founded DeepLearning.AI in 2017 to develop artificial intelligence education. The organisation combines learning programmes, practical exercises and a learning community. Its activity expanded the provision of specialist courses and connected content published on Coursera with its own additional educational resources. [2]
His Agentic AI course addresses systems that plan tasks, use tools and review their outputs through several stages. This approach shifts attention from a single response to process design. Model reflection is a technique requiring evaluation, rather than an automatic guarantee of correctness. [18]
Books, project decisions and data quality
Machine Learning Yearning organises practical machine learning development decisions: prioritising experiments, diagnosing errors and recognising differences between training and test data. The book also addresses when to use transfer learning, multitask learning or end-to-end systems. Its focus is helping a team navigate specific technical alternatives. [19]
In the 2018 AI Transformation Playbook, he proposed starting with pilots, developing internal capabilities, training teams, defining a strategy and communicating changes. The guide directs these recommendations at business organisations and presents adoption as a sustained process rather than simply purchasing a tool. [20]
In 2021 he advocated a data-centric approach. He proposed benchmarks in which code remained fixed while participants improved the dataset, reversing the usual arrangement of comparing models on unchanging data. The proposal makes another research direction explicit: improving the information used for learning. [21]
LandingAI and document extraction
He founded LandingAI in 2017. The company currently identifies Ng as leading the technical direction of Agentic Document Extraction, an API platform turning business documents into structured data. This work adds document processing to a business trajectory initially associated with computer vision solutions. [22]
In August 2024 he moved from LandingAI’s chief executive position to executive chairman. He has also served on Amazon’s board since April 2024. [4]
AI Fund and AI Aspire
AI Fund operates as a venture studio, working with entrepreneurs on ideas, prototypes, development and early funding. In May 2025 it announced the closing of a second fund of approximately $190 million. Ng signed the announcement as managing general partner. [23]
In July 2025 Bain announced a partnership with Ng and his firm AI Aspire. The agreement combined strategic advice and technical expertise to help businesses develop their AI transformation. [24]
LearnVector and personalised learning
In 2026 he founded LearnVector and became its chief executive. The company is developing individual learning experiences supported by AI. Its public introduction places initial product demonstrations in early 2027, so this stage concerns announced development rather than an already available product catalogue. [25]
Coursera invested $100 million in July 2026 for 33.33% of LearnVector. Its financial filing identifies a related-party transaction because of Ng’s roles at both companies and records approval by an independent committee. In October they signed a framework for future commercial collaborations. [26]
Scientific recognition
In 2009 he received IJCAI’s Computers and Thought Award alongside Carlos Guestrin. The award recognises outstanding young artificial intelligence researchers. In his career, this scientific recognition preceded the expansion of his massive online courses and many of his business ventures. [27]
Computer vision and distancing at work
In 2020, Landing AI introduced a tool to monitor distancing between people at work. Andrew Ng was its CEO at the time, according to the disclosure published by DeepLearning.AI. [7]
Landing AI’s demonstration analysed real-time video to detect whether the intended separation was maintained. The company presented it as support for its customers’ workplace distancing protocols during the pandemic. [6]
From a detection to a preventive measure
The case offers a specific example of AI applied to the organisation of workspace. A useful assessment could compare detections with observations on site, identify routes that repeatedly trigger alerts and check whether changing the layout reduces those situations. It would also need to assess errors when people obscure one another or the camera position changes.
The observed indicator is the estimated separation between people. Assessing a preventive measure would also require examining how the alert is used, what decision it triggers and its effect on exposure. The demonstration documents a technological function; its health outcomes would require a different evaluation.
