What Google AI Could Say About Your Doodles?
…or how people draw elephants
No one analyzes doodles. If your friend draws something like this:
You say “Haha, you’re funny” or “Go practice, dude!” Ok, if you really like someone you might start discussing deep feelings attached to that smiling dog. You both can have a good laugh and that’s pretty much it.
How often do you see other people’s doodles anyway? Usually it’s a small circle of friends playing Pictionary at a party, or colleagues, tainting each other’s business presentations.
The only doodle we see more or less constantly is male genitals on our neighbor’s garage. Not much to analyze there.
But things are changing.
I’m sure many of you heard of Google’s “Quick, Draw!” game that recognizes your doodles. The game went viral, and people created thousands, if not millions doodles in it.
There is just one problem.
Google won’t show us these doodles. It only shows 15–20 of them. On top of that what you get is whether your doodle was recognized or not, that’s all. What if Google could tell more? In future, maybe. But not now.
15 drawings is enough for a good laugh or a boost of self-confidence, but not enough to see the bigger picture.
What if I’d like to know how many other people decided to draw a boomerang with a hangover? Or how many of them were angry? Or southpaws? Another way of getting answers to such questions could be simply to analyze hundreds of them, to reveal patterns. But we don’t have such an access.
This article could easily end here with an outburst of a missed opportunity.
However, like in every good Hollywood movie, there is a dramatic coincidence lurking around the corner. Our company, Icons8, developed exactly the same doodling game, only we didn’t present it as a game.
And before you accuse me of ripping off Google’s idea, I should mention it was released on our website 2 months before theirs.
I won’t shatter the skies by saying that Google’s “Quick, Draw!” is not exactly a game, but a neural net that learns to recognize visual images. So is ours. It just wasn’t presented as a game, rather, as a visual search for icons.
And, just like Google’s, with every doodle people made for it, the neural net became better at recognizing things.
Our neural net, however, had neither a game-component (no track of success) nor a social component (no social buttons, no sharing ), so it never went viral. We just asked people from our mailing list to draw different things for us and they did.
Overall they made 45,621 doodles. Animals, vehicles, icons, gadgets — all kind of stuff.
I saw this as a chance. Instead of 15 similar drawings in Google I could see hundreds.
Besides being a big-time analytical nerd, since college I have been learning about drawing and the science behind the process, so when I was left with all these drawings I couldn’t resist.
I spent 3 weeks analyzing them, observing, looking for patterns. And I found a few. These patterns could be used for a deeper analysis of a thought process behind drawing and the way people use their brain. Eventually, neural nets would be able to make similar observations or correct mine.
To demonstrate them, I’ll use one of the most diverse and broad categories — animals.
How People Draw Elephants
People are usually not that good at drawing animals. And why should we be? It takes time to master drawing them, so we tend to depict them stereotypically.
Why we use these stereotypes? It’s thoughtfully explained in a very popular Betty Edwards “Drawing on the Right Side of the Brain” book. In short, our brain consists of two parts: left-brain and right-brain. Left- and right-brain do things differently — they perceive information differently, and they process it differently.
It has an impact on drawing as well. If I was to ask you to copy a horse from reference, the result would vary whether you used your left-brain or right-brain for that.
The left-brain is analytical, logical. It simplifies and structures information and then assigns labels to it, so we can recall it easier. In short: It takes a horse, breaks it into parts (head, legs, and body) and simplifies them into ovals and sticks (see the above illustration).
This results in a stereotypical drawing, and these stereotypes are very persistent. Once learned, you may use them throughout your whole life, drawing horse exactly same way every time.
Right-brain is all about intuition and observation. It sees things exactly the way they really are. It has no words and no labels. No heads, no legs, just curved lines and contours. In right-brain mode you can copy an object very accurately.
Trained artists can deliberately switch between the two modes — they can create structure using stereotypes of left-brain mode (simple geometrical shapes) and then recreate more complex parts of the drawing using right-brain mode while portraying something.
But that’s trained artists. Untrained people use left-brain by default. Talking, logically thinking, calculating, structuring information — most daily tasks require left brain. So when you ask untrained people to draw something from memory, a vast majority of them will be using their left-brain as default. That’s why 95% of drawings will consist of stereotypes that were formed long ago:
But what happens if you show people a reference before they doodle? There are 3 possible scenarios:
Right-mode copy: people switch into right-brain mode and copy the reference quite accurately.
Mixed: people start copying the contours or big shapes in right-mode, but somewhere in the middle left-brain mode kicks in and they start using stereotypes. They will “attach trunk somewhere to the head” or draw 4 parallel legs.
Own thing: No matter what the reference is, people would draw something of their own. In all cases it’s a prepared, stereotypical drawing that people learned before. i.e. left-brain
I analysed about 70 drawings of elephants. Here’s a chart:
Most people followed the reference and switched into right-mode. Thus their copies ended up being very accurate. But here’s the catch. It’s not the same for other animals. Let’s bring out the whole zoo.
There is a lot going here, so let me dive into it.
The More People Know, The Worse They Draw
Why do people copy elephants in right-mode much more often than any other animal? This question really made me think. At first I thought — it must be it’s unique, exotic shape. But then again, turtles are exotic too. Horses have pretty unusual shapes in them as well.
Then it hit me. Turtles are not that exotic to us. We know them very well. Or we think, we do.
People more and more often have turtles and piggies as pets — we see them quite often.
It’s not about knowing, it’s about perceived knowledge. The more we think we know what pigs and turtles look like, the more stereotypes we form about them. We ignore reference and apply those stereotypes in our drawings.
I’m not even talking about cats, dogs, and horses. Horses are in every 2nd movie, cats and dogs are everywhere. That’s why they are mostly drawn in mixed mode or people use their own prepared drawing for them.
Elephants are not very trendy. We don’t know much about them, we don’t see them often (most recent movie about elephants, anyone? Jungle Book? Alexander? The whole Kung-Fu Panda series population consists of 90% pigs).
The unfamiliar shape of an elephant forces us to gain knowledge about them from observation, switch into right brain mode, and copy them more willingly.
Just like in countless social psychology experiments, when people are unsure of how to act in a current situation, they tend to follow the external example.
The more perceived knowledge you have about something, the bigger chance you’ll draw stereotypically.
There is also a bright side to it. Seeing that people draw different animals differently, I say everyone can learn to draw better by working on their stereotypes and switching to right mode deliberately.
When we talk about emotion in art, people tend to image a furious artist penetrating the canvas with red eyes and a mad smile. That’s a common stereotype. The truth is, right-mode is quite peaceful.
However, left-mode isn’t. When we talk about doodles, people literally channel their emotions into their doodle, when they do them in a left-mode.
Take a look at these drawings.
All these drawings contain stereotypes. They all were made in left-brain mode. But can you see the abrupt, unfinished lines in pigs compared to elephants? It’s as if the drawing was made in hurry, meanwhile the elephant’s shapes are more smooth, polished, and neutral?
Given that all of these drawings were made with a mouse, it’s not about just coincidence, it’s a pattern.
The most “abrupt,” emotional drawings were made for pigs, cats, dogs, and horses. As if people were emotionally attached to the picture, but for different reasons. In some cases positively, in some, negatively.
I think the most typical emotion here is a despair.
People more often completely give up while drawing a horse than with any other animal. As if it was a failed attempt from the beginning, even though they had a reference right in front of them.
I was already talking about persistent stereotypes in our drawings, but it seems like emotions attached to these drawings are just as persistent. I haven’t seen such behavior for any other animal.
Pigs are very controversial animals in our society. For some people they are cute domestic pets, for others, pigs are dirty animals. Sometimes it’s even a verbal insult.
You can find out for yourself which is which.
Also, if you look at the comparison chart again, you can see that pigs are the leader in “drawing my own thing ignoring the reference” section. However, these drawings are not very diverse. Most of them look like this:
Yet, there are animal doodles that are even more diverse in “my own drawing” category.
Cats & Dogs
Cats and Dogs are the most controversial of all animal categories. They have both beautiful drawings and abrupt, stereotypical ones. As if our society is really divided into cat lovers and dog lovers.
Here are the people who definitely love cats:
And those, who equally hate them.
It seems like emotions and stereotypes are deeply connected. Emotions affect our drawings when we’re in left-brain mode. That’s why we’re trying so hard to exclude emotions from our logical, business decisions of our left, analytical brain.
The situation is pretty much the same with dogs. Careless, erratic lines give away the attitude.
How do I see that people love dogs? Or cats? It’s not just about the quality of lines or stereotypes.
They add smiles.
The stronger the emotion (hate, despair, love) people experience, be it an emotion towards an object or towards themselves (e.g. inability to draw a horse), the bigger chance they will switch into left-mode and start drawing stereotypically, ignoring the reference altogether.
At this point in time, neural nets like Google’s “Quick, Draw,” are still learning to recognize people’s drawings. And that’s just the beginning. Soon they will be able to analyze them. Just like I did, but using bigger data. I hope my findings will prove useful and may be used in more advanced neural networks. I’m sure people’s drawings can provide much more information than just a “similar or not” or “good or bad” binary response.
In the meantime, there is so much we can learn from these nets as well. So the next time your friends are drawing a doodle, don’t be quick to judge them. I assure you, there is always something you can see there.
Thanks for reading! I hope you’ve found this article interesting, and I’ll be glad to hear your feedback in the comments.
About the author:
Andrew started at Icons8 as a usability specialist, conducting interviews and usability surveys. He desperately wanted to share his findings with professional community and started writing insightful and funny (sometimes both) stories for our blog.