It usually takes two (or more) players to play a game of Pictionary, or “guess what I drew”. Now, thanks to Google, it only takes one!
Introducing Quick, Draw!, an A.I. experiment created by Google designed to teach itself how to recognize drawings and match them to words. Today, I’ll be providing a very basic walk through on how the program works. If you saw the title and are just interested in learning about it yourself, skip to the bottom of this article for the link.
Quick, Draw! is a neural network, an AI program designed to use machine learning to learn and remember the information it receives, so that it can better recognize it in the future. In Quick, Draw!’s case, it asks users to draw a picture of something to the best of their ability. Once the AI program is able to sufficiently recognize the picture, it moves on to the next one. If it cannot guess the picture within 20 seconds, it simply moves on to the next one. This is the first screen that will show up after starting:
This is the screen that appears right before it asks you to draw whatever it asks for. Here, it wants us to draw a bridge. You will then be asked, to the best of your ability, to create your best interpretation of a bridge. If the computer can recognize it, you’ll automatically move to the next one!
Once you have completed (successfully or unsuccessfully) all six of your pictures, you will come to a results screen where you can analyze and see what other people drew for each picture:
From here, you can click on each picture to see what other people drew and what the computer recognized.
Here we see that the AI was able to recognize the bridge, as well as showing what other words or items it may have thought you were drawing. Rainbow and fish were a couple of the other potential matches!
Finally, if you scroll down, it will show you what other people drew that were successfully interpreted. As you draw, the Google AI uses a database of these pictures to try and help it identify what you are trying to draw. Naturally, the more pictures it has in it’s data bank, the “smarter” it will become, and the faster it will be able to detect what you are trying to draw. With some pictures, you can draw as little as two lines before it can tell what you are trying to draw!
Neural network AI is a newer technology that is just starting to get its legs. Earlier this year, Tay.AI, a Twitter neural network bot created by Microsoft, was “taught” to be incredibly rude in less than 24 hours, thanks to user input. As time goes on and the technology improves, we will begin to see a whole variety of uses for this type of technology.
It’ll only be a matter of time before the “Skynet” becomes a very real possibility.
To get started with Quick, Draw!, head over to https://quickdraw.withgoogle.com/ to give it a try.