A.I. Experiments: Quick, Draw!

A.I. Experiments: Quick, Draw!


[MUSIC PLAYING] JONAS JONJEGAN: Hi, I’m Jonas. HENRY ROWLEY: Hi, I’m Henry. JONAS JONJEGAN: Quick, Draw is
a game a few of us at Google made. You draw, and the computer
uses machine learning to guess what you’re drawing. AI: I see square, or
suitcase, or canoe. Oh, I know, it’s shoe. JONAS JONJEGAN:
It’s an experiment that uses some of the
same technology that helps Google Translate
recognize your handwriting. HENRY ROWLEY: To understand
handwriting or drawings, you don’t just look at
what the person drew. You look at how they
actually drew it. Which strokes did
they make first? Which direction did
they draw them in? You train the computer on
millions of characters, from hundreds of languages. And over time, it learns
whether you wrote look or whether you wrote book. JONAS JONJEGAN:
Training is a big part of how the computer can guess
your drawings correctly. As people, it’s easy for us to
look at these three drawings and know they are all cats. But to a computer,
they’re very different. One is just the head. One has a full body. And one is just facial features. To get the computer
to understand, you have to show it
a lot of cat doodles. And then it starts
to see patterns, like that almost
all doodles of cats have pointy ears, a
small nose, and whiskers. Of course it
doesn’t always work. That’s because it’s only
seen a few thousand doodles. But the more you play with
it, the more it will learn, and the better it
will get at guessing. AI: Oh I know, it’s cat. JONAS JONJEGAN: We put it on
the web for anyone to play with. We hope it inspires other
people to think about fun ways to use machine learning. You can play it at
g.co/aiexperiments.

87 thoughts on “A.I. Experiments: Quick, Draw!

  1. Hi, with this experiment, does it starts collecting the information from users? If I draw a dog right now, it starts guessing and gives me 6,7,8 guesses. In a few days after many trials from users, it would learn and starts giving me less guesses? It would tell me its a dog in less time/guesses?

  2. So it's not really guessing anything, since you draw what it asks for. Right now it's a game about providing a training set for your AI.

  3. Hey, since there seems to be no better way to contact you. It would be wonderful if you set the voice attribute of the text-to-speech to English. Sounds absolutely terrible with the default settings for me 🙂

  4. It asks me to draw phone, I draw phone, it says "that looks like a telephone. I cant guess what you drew". Why doesn't it know a phone and telephone are the same

  5. I think the problem is that if you play it longer than half an hour, it's not the computer that's learning how to decide, but the player is learning how to draw. This problem will become less important when many more objects are added, but with only a few hundred it's probably disastrous for the system.

  6. I suppose that if it's a neural network, it's not difficult to add thousands of words immediately, and just let it run. Only let's say the first 100 drawings of something will not be 'guessed' by the system, but only used to build up some data. However, I think that then the results will be even worse…

  7. It asked me to draw a cake, drew it and it responded with cake…
    I used the same cake drawing for everything, and now it just shows "…" as if it can't recognize the cake drawing it guessed correctly earlier.

    Something is fishy with this.

  8. It doesn't appear to rotate the items as it didn't recognize a perfectly decent saw, but only the second time I drew it at a 90 degree angle pointing up.

  9. Hmm, not really sure if the A.I. is actually learning well… It looks like it's recognizing things drawn only from a certain angle – so crab cannot be drawn from above, needs to be front-facing and some things need to be vertical and others horizontal… And even funnier thing is the cultural aspect – many "ideal" objects seems to be taken straight from American animations for kids, not from the real world 😛

  10. Please answer this question: https://ai.stackexchange.com/questions/2351/is-the-quickdraw-with-google-neural-net-a-convolutional-neural-network

  11. One of it's problems it that is making guesses from other doodles, but not from how we truly see things, now since most people don't draw in perspective, if you start drawing objects in perspective the systems is unable to recognize some of the objects.

  12. YOU GUYS ARE NO MORE THAN A COUPLE OF PATHETIC 3 YR OLDS, WHO DON`T KNOW HOW TO DRAW CORRECTLY!!!!!!!!!!!!! I played your game, and when I looked at pictures of what I was expected to draw of a guitar, and IT LOOKED LIKE A TODDLER DID THAT PICTURE!!!!!!!!!!!!

    | /|
    | / |
    | __ /__|_

    ________________________

  13. What would happen if you drew dickbutt? Would the experiment then start thinking that the Eiffel Tower looks like dickbutt?

  14. My one complaint is that it's very hard to do this with a mouse on your laptop vs using your finger or a stylus on a mobile device. Other than that it's pretty fun.

  15. The translate is inaccurate. It requires me to draw a 扇子(that generate wind by swaying a piece of paper or something), but with other's 電扇(that generate wind with election) pictures

  16. Why donot they make a software that takes the pictures and turns them into drawings of different kinds, so both can learn

  17. Question, does the AI know in advance a limited amount of possibilities? Because why does it gues 'police car' (instead of car) when I haven't even started drawing the police specific parts of a car.
    So that is very strange, the AI should guess 'car' first.

  18. I don't know how to send suggestions but it should be more careful with translations, a for example it traslated bucket as "cubo" but "cubo" also means cube so sometimes you don't really know what it wants.

Leave a Reply

Your email address will not be published. Required fields are marked *