# Andrew Ng

> 1976– · Computer Scientist, AI Researcher, Co-founder of Coursera
>
> **Recorded contribution:** Google Brain co-founder; Coursera co-founder; Stanford ML course; deeplearning.ai; democratized ML education

## How to use this dossier

Read for a causal chain, not a hero story: inherited problem → contribution → mechanism → downstream capability → limit. Then close the page and complete the reconstruction exercise from memory.

## 1. Historical orientation

Computer scientist Andrew Ng led the Stanford AI Lab, co-founded Google Brain in 2011, later led AI work at Baidu, and co-founded Coursera and DeepLearning.AI. His combined historical role is research leadership, production advocacy, and unusually broad education: Stanford's online machine-learning course helped catalyze the modern MOOC movement.

## 2. The problem inherited

Machine learning knowledge and large-scale training infrastructure were concentrated in a small number of universities and technology firms, while many practitioners lacked a disciplined path from objective and data to evaluation and deployment.

## 3. The central contribution

Ng helped scale deep-learning research and democratize machine-learning education through open online courses, curricula, and institution-building.

## 4. Reconstruct the mechanism

1. Frame a prediction or perception task with an explicit dataset and evaluation split.
2. Build a simple baseline before increasing model and compute complexity.
3. Train at scale by parallelizing data processing and numerical optimization.
4. Use error analysis to decide whether data, labels, model capacity, or deployment process is the next bottleneck.

## 5. What changed downstream

- Mass online courses made machine-learning instruction available to millions of learners.
- Google Brain and subsequent organizations helped normalize large-scale deep-learning engineering in industry.

## 6. Attribution, limits, and uncertainty

- Research and platforms are team achievements; Ng's leadership should not absorb the work of students, collaborators, engineers, and course staff.
- 'Democratization' through courses does not remove compute, data, language, labor, accessibility, or institutional barriers to meaningful participation.

## 7. Reconstruction lab

Choose a small classification task, write the decision and error costs, train a baseline, produce an error table, and justify one—not five—next interventions from the evidence. Split the evaluation set before inspecting errors and preserve a final untouched test set. Label each mistake as a data, representation, optimization, threshold, or objective failure; only then choose an intervention. Calculate whether a one-point accuracy gain changes the decision cost. Finally, write a short explanation suitable for a beginner and another for a practitioner, reflecting Ng’s dual influence on practical iteration and scalable education without confusing pedagogy with algorithmic invention. Publish the baseline, split, metric, and chosen intervention before rerunning the experiment so educational clarity is paired with protection against post-hoc storytelling.

## 8. Evidence trail

- [Andrew Ng](https://ai.stanford.edu/~ang/) — Stanford University
- [Machine Learning course](https://see.stanford.edu/Course/CS229) — Stanford Engineering Everywhere
- [Building High-level Features Using Large Scale Unsupervised Learning](https://proceedings.mlr.press/v27/le13a.html) — ICML

---

*Research checked 2026-08-09. Dates, roles, and claims about living people are historical snapshots. Linked sources remain the authority; this dossier is original instructional synthesis.*