The Neural Network That Found Cats on the Internet
Google's secret project used 16,000 processors to build a neural network that spontaneously learned to recognize cats from unlabeled YouTube videos — proving unsupervised deep learning could discover concepts on its own.
Charles Bell / Public domain
We never told it during the training, 'This is a cat.' It basically invented the concept of a cat.
— Jeff Dean
The Neural Network That Found Cats on the Internet
In 2012, Google Brain researchers Andrew Ng, Jeff Dean, and Quoc Le made a breakthrough that would become a cultural touchstone for artificial intelligence: their neural network learned to recognize cats from YouTube videos without explicit instruction.
What happened: In 2012, Google Brain researchers Andrew Ng, Jeff Dean, and Quoc Le demonstrated that a large-scale neural network could learn to identify high-level concepts like faces and cats from raw data, specifically YouTube videos, without being explicitly programmed to do so. This experiment, detailed in their paper “Building High-level Features Using Large Scale Unsupervised Learning”, showcased the potential of deep learning to extract meaningful features from vast amounts of data.
Why it matters: This milestone was significant because it proved that scale and data could substitute for hand-engineered features in machine learning, validating the approach of using large datasets and neural networks to learn complex patterns. It also marked a turning point in public perception of AI, illustrating that machines could learn from the same internet data that humans consume daily. This realization spurred Google to invest heavily in deep learning infrastructure, setting the stage for future advancements in AI.
Further reading:
Why This Mattered
This experiment demonstrated that large-scale neural networks could learn high-level concepts like faces and cats without ever being told what they were, validating the idea that scale and data could substitute for hand-engineered features. It became a cultural touchstone — the moment the public realized AI was learning from the same internet they used — and helped accelerate Google's massive investment in deep learning infrastructure.
