My coworkers asked me to write another blog post. I am surprised at this. One might think they’d have learned their lessons after I ruined hard-boiled eggs for everyone. One might be wrong.
Instead of asking Google for advice on what to write like last time, I discovered Google has a thing. A new thing. A thing that helps people draw who don’t know how to draw. This new thing uses machine learning algorithms to interpret what you draw and then suggests a picture of what it thinks you were trying to draw.
Naturally, I had to give it a go.
So I drew a carrot. It’s an approximation of a carrot. I even used carrot colours to help AutoDraw get the hint. AutoDraw failed to understand my desire to have a perfect drawing of a carrot. Instead, it suggested a wishbone:
Then a frog:
Beginning to feel dismayed, I tried again.
The algorithm apparently gives up on you when you try and tell it what to give you. Not wanting to give up, I revised my drawing.
AutoDraw still didn’t offer me a carrot. Instead, I got a series of beards, and even a trashcan!
Perhaps I should have taken the hint. But no, my quest for a perfect clipart carrot drove me onwards!
I told myself it would be the last attempt. This was obviously a carrot. Success!
Only four suggestions down, there sat my carrots. Admittedly, the sixth suggestion looked a bit more like a turnip, but I am still pretty impressed. Some say I am easily impressed. These people would be fools. Think about it: machine learning looks at these terrible scribbles, and it understands it is a misrepresentation of an actual thing, and then tries to offer you a better, clearer version of that thing.
Easily impressed, indeed.
Why, yes, AutoDraw, I meant to draw a dragon.