Quick, Draw!

Objective

Google aimed to demonstrate the practical and engaging applications of machine learning and neural networks to a broad audience. By presenting AI as an interactive, accessible experience rather than abstract technology, Google showcased how neural networks could learn from user-generated data in real-time while entertaining millions worldwide.

Solution

Quick, Draw! is an online guessing game where players draw objects in a series of 20-second sketches (six total per session). A convolutional neural network attempts to identify each drawing in real-time, and after completion, players see how many drawings the AI correctly recognized. The game was built as a web-based experiment and directly incorporated user drawings into the training dataset—the model was trained on 50 million drawings across 345 categories, with new user data continuously feeding back into model improvement.

Results

Quick, Draw! was an interactive game developed by Google Creative Lab where a neural network attempted to guess players' drawings. The game collected 50 million player drawings in its publicly available dataset.