Google’s AI is exceptionally good at Pictionary
- Fariha Khan
- November 22, 2016
- 551
When Google open sourced its deep learning engine in 2015, its researcher partnered with The New York Times to create this data visualization showing neural networks. The firm has now rolled out AI Experiments, a collection of games and tools on the web in order to help you understand how machine learning actually works.
Consider Quick Draw game which works similar to Pictionary. This game provides you with 20 seconds to draw an object on screen, and Google makes guesses alongside. You may be asked to draw an umbrella, a microwave, a baseball etc. and Google will guessed it correctly. Every time.
While the game has great accuracy, it is not a powerful tool for learning. Observing how Google responds to your doodling can give you an idea how its technology works.
Actually what happen is that when you are asked to draw a thing such as a tree you start making leaves. In its robotic voice, Google guesses: “squiggle.” As you draw more leaves, it sees a bush and then you draw the trunk and it works. In other words, AI works and gives you results by guessing data one by one.
AI Experiments is not simply a free education for neural network nitwits. Every interaction improves Google’s ability to more quickly identify images and language. It makes the company’s products stronger and serves the user better as well. The data strengthens apps such as Google Photos that make use of AI to organize your pictures in a flash.