Irene Alvarado // Art && Code: Homemade, 1/14/2021

Irene Alvarado // Art && Code: Homemade, 1/14/2021

Show Video

Hello everyone and welcome back to Art&&Code  Homemade: digital tools and crafty approaches.   I'm thrilled to welcome you now to the session  with Irene Alvarado who is a designer and   developer at Google creative lab where she  explores how emerging technologies will shape   new creative tools. She's also a lecturer at  NYU's interactive telecommunications program.   Irene Alvarado.  

Hi everyone very excited to be here.  I'm going to talk today about a behind-the-scenes story of a particular tool called Teachable  Machine. So as Golan said I work at a really   small team inside of Google. I mean we're less than  you know 100 people which is small for Google size.   And some of us in the lab, the work that we  do includes creating what we call experiments to   showcase and make more accessible some of the machine-learning research that's going on at Google. And a lot of this work shows up  in a site called Experiments With Google.   

And time and time away, we're sort of blown by what happens when you lower the barrier for experimentation   and access to information and tools. And so today I  want to tell you about one particular project   and especially the approach that we took to  creating it. So why am I speaking at a homemade   event when I work at a huge corporation. I think  you'll see that my team tends to operate in a way   that's pretty scrappy, experimental, collaborative.  And this particular project happens to be a   tool that other people have used to create  a lot of homemade projects.    So to begin with let me just talk about what it is.

It's a no code tool to create a machine learning models.   So, you know, that's a mouthful. So I think I'm  just gonna give you a demo. So I believe you can see my screen. This is sort of the home  page for the tool. It's called Teachable Machine.   And I can create different types of what we call  models in machine learning.  

Basically a type of program that the computer has learned to do something. And there's different types.   I can choose to create an image one, an audio one, a pose one.   I'm just gonna go for image, it's the easiest one.  

And I'm just gonna give you a demo so you see  how it works.    I have this section here where I can create some categories. So I'm gonna create three categories. I'm gonna call this  

like neutral or I could call this just like  face or something like that. And I'm gonna give the computer some examples of my face. So I'm going to do something like this, give the computer some examples of my face. Then I'm going to give  the computer some examples of my left hand.   And then yeah that should be enough. And then  I'll give the computer some examples of a book,   Happens to be a book that I like.  

And then I'm gonna go to this middle section to sort of train a model.  And right now, it's taking a while because  all this is happening in my browser. So the tool is very very private. All of the data is staying in  my browser through a library called tensorflow.js   So one of the main points here was to showcase that you can create these models in a really private way.    And so now the model's trained.  It's pretty fast, and now I can try it out.  

I can see that you know it can detect my face. Let's see, it can detect a book, and it can detect my hand. And, you know, some interesting stuff happens,  you know, when I'm half in half out, you see that   the model is trying to figure out okay is  it my face, is it my left hand, you know.   You can sort of learn things about these models  like the fact that they're probabilistic.  

And, you know, so far maybe not so special. I think what really unlocked the tool for a   lot of people is that you can export your model.  So I can upload this to the web and then if  you know how to code, you can sort of take this  code and then essentially take the model outside   of our tool and put it anywhere you want. And then you can build whatever you want with it, a game,   another app, whatever you want.  So let me go back to my slides here. We also have sort of other conversions, right.  So you can create a model and not just  

have it on the web. You can put it on mobile. You  can put it, you know, in Unity, in Processing,   in other formats, and other types of platforms.  And just a little note here of course this is not just myself. I worked on this project with a lot of  colleagues, a lot of very talented colleagues of mine.   Some of them are here. And I'm going  to showcase a lot of work from other people   in this talk. So as much as possible I'm going  to give credit to them, and then at the end I'll  

share a little document with all the links in case  you don't capture them. So okay I want to emphasize   that a lot of our projects have really humble  origins, you know. It's just one person prototyping   or sort of hacking away at an idea, even though  it might seem like we're a really big company, or a really big team.   And just to show you how  that's true, you know, the origin of this project   was actually a really small experiment that looked like this. The interface sort of looks the same, but it was really simplified.

Like you have this sort of three panel view. And you couldn't really add too much data. It's a really really simple project.  But more so than that, even though it was   technically very sophisticated, I think our ideas  at the time were very-   we were kind of exploring really fun use cases.

And just to show you how much that's true I'll show you a project that   one of the originators of this idea  Alex Chen tried out with his kid. So let's see [organ sound effect] [bird chirping sound effect] [quiet organ sound effect] [bird chirping sound effect] So you can see he's basically creating these  paper mache, you know, figurines with his kid.   And so he's training a model  that then triggers a bunch of sounds.  

So it's really kind of fun at the time, just  trying a lot of different things out.   And then we started hearing from a lot of teachers  all over the world who are using this as a tool to   talk about machine learning in the classroom, or  talk about sort of the basics of data to kids.   And then we finally heard even from some folks in policy. So Hal Abelson is a CS professor at MIT,   and he was using the tool to conduct sort of  hands-on workshops with policymakers. So we had a hunch that maybe the silly experiment could become something more.

But really really didn't know how to transform this into an actual tool.  And we also didn't know necessarily what the best use cases would be. And this is where the project  took a really really interesting turn, because   essentially we met the perfect collaborator to  help us and to push us into making it a tool.  

And that person his name is Steve Salling.   He was introduced to us by another team at Google who had been working with him. And Steve is this amazing person. He used to be a landscape architect   and he got ALS which is Lou Gehrig's disease.  And he sort of set out to completely reimagine   how people with his condition get care. And he  created this space that everything is API-fied.  

So he's able to order the elevator with his  computer or turn on the TV with his computer.   It's really amazing. And so he actually found the  original Teachable Machine, and someone else sort   of was using it with him. And, you know, we basically  got introduced to him, and the question was well  

you know, can we figure out if this could be useful to him, and in what way? And, you know, how do we just get to know Steve, and what he might want? So a little pause here to say that, you know, folks like   Steve they're not able to to move or you know  in Steve's case he's not able to communicate.   So he uses something called a gaze tracker or like a head mouse. And he essentially sort of looks at  a point on the screen and then can type into-  can sort of press click and type a word or a letter. So he's able to communicate but it's   really really slow. And so the thought was: okay can   we use a tool like Teachable Machine to train some,  you know, train your computer to detect some facial   expressions and then trigger something. And this of course is not new.

The thing that was sort of new was   not for me to train a model for him, but for  him to be able to load the tool on his computer   and train it himself. Like sort of put that  control on Steve.   And specifically you know,  we basically went down to Boston and worked  with him quite a lot. He became sort of the   first big user of the tool. And we made a  lot of decisions by working with Steve,   and sort of like following his advice. And one of  the things that the tool sort of allowed   us to explore was this idea of immediacy.  So what were some cases where Steve wanted   sort of a really quick reaction, and how could  the tool help for this.   

And one use case was he really wanted to watch a basketball game.    And he really wanted to be able to cheer or to boo depending on what was happening in the game.  And that was something that he was not able to do   with his usual tools, because he had to be like  really fast at cheering or booing when something happened.   And so he trained a simple model that  basically, he could sort of open his mouth and   you know trigger an air horn. So that was  one example.

So you know we kind of immersed ourselves in Steve's worlds. And by getting to know him, we got to know that, you know, maybe other   ALS users could find something like this useful. So we started exploring audio. Like could we have   another input modality to the audio to potentially  help with people who sort of had limited speech.   And that led us to incorporate audio into the tool.   So I actually have a little example here   that I also want to show just so you guys see  how this works.  

I'm loading a project that I had created beforehand from Google Drive. So let's see if it doesn't, you know, this is-   this is some data that I had collected beforehand, some audio data. So there's three classes,   there's background noise, there's a whistle,  and there's a snap. And let's see if it works. [whistle noise] As you can see the whistle works.

[snap sound] You can see the  snap works. So you know, same thing here. I can kind of export the model to other places.   So you know, but the interesting thing here is that   the audio model itself actually came from  this engineer named Shanqing Cai. And he created   that audio model for people like Mark Abraham  who also have ALS through exploring with him   the same idea, like how can I create models for  people like that so that they can trigger things on the computer.    So the technology itself, you know,  also came from this exploration of    working with users who have ALS.

And, you know, you can't see  what's happening here, but essentially Dr. Abraham   has sort of emitted a puff of air and with that  puff of air, he's been able to turn on the light.   So you know, we decided that okay this could be  useful to other people who have ALS.    And we decided to essentially open up a beta program for other people to tell us if they had other similar uses,    and trying to think about maybe other  analogous communities that could find   you know, some interest in Teachable Machine. And that's how we met the amazing Matt Santamaria and his mom Kathy.   

And Matt had come to us and told us  that, you know, he was playing with the tool   and he wanted to try to use it for his mom.  So he actually created a little photo album   that would sort of, you know, change photos  for his mom, and he could sort of load them remotely.   But his mom because she didn't have  really too many motor skills, because she had a stroke,   she wasn't able to control the photo album.  So we actually worked with Matt. And, you know, you can't see it here because it's just an image, but we actually worked with Matt to create a little prototype of, you know, his mom being able to say  previous or next, and then training an audio model   that then was able to sort of like change the  picture that was being shown on the slideshow.   And that was just sort of like a one-day  hack exploration that we did with Matt.   And then ultimately he sort of kept exploring  potential uses of Teachable Machine with his mom.  

And he created this tool called Bring Your Own  Teachable Machine, which he open-sourced and has shared online. And what it allows you to do is to  put any type of audio model, and then link that to   basically a trigger that sends texts-  that sends a text to any phone number.   So pretty cool to see what he did here.   And then finally, you know, just seeing that the tool sort of ended up being useful in a lot of analogous  communities once we launched it publicly. We saw   a lot of uses outside of accessibility. So I wanted to show you a few of my favorite ones today.  

This is a project by researcher Blakeley Paine  She used to study at MIT Media Lab,   and she was really interested in exploring  how to teach AI ethics to kids. So she open-sourced this curriculum. You can  find it online and it just has a bunch of   exercises, really interesting exercises that she  takes kids through. So this one for example  

explains to them the different parts of a  machine learning pipeline. So in this case, you know, collecting data, training your model, and then running the model. She sort of gives them different data sets. So you can see here in the  picture, it's a little blurry, but you can see   the kids got 42 samples of cats and then  seven samples of dogs.    And so the idea is for them to train a model, and then see okay maybe the  model's working really well in some cases   maybe it's working really poorly in some cases. Why is  that? And have a conversation with them about   AI ethics and bias and sort of how training a  model is related to the type of data that you have.  

Here's another example of her workshop.  She sort of asks kids to sort of   label their confidence scores. And you'd  be surprised, you know.   We were invited to join one of the workshops and these kids sort  of established a pretty good kind of insight   into the connection between how the model  performs and the type of data that it was given.  

It's not just Blakeley. There's other  organizations that have created lesson plans.   This one is called and, you know,  again they kind of use these simple training   samples in a lot of schools. You can't use your  webcam, so a lot of them sort of have to upload   picture files or photo files in order to to use  the tool.

And that was a use case that we found out through education that we had to sort  of enable, you know, not having a webcam.   And then more so in the realm of hardware, there's a project called Teachable Sorter   created by these amazing designer technologists Gautam Bose and Lucas Ochoa. And what it is  is it's a very sort of powerful sorter. So it uses this- oops. It uses this accelerated  

hardware board it's kind of like a Raspberry Pi.  And they essentially train a model in the browser    and then they export it to this hardware. And they both were super super helpful   in sort of creating that conversion between web  and, you know, this little hardware device.   Now this is a very complex project. So they made a simple version that's open-source and you can find the instructions online. And it's a tiny sorter,  it's a sorter that you can put on your webcam.  

And so again you can train a model with  Teachable Machine in the browser,    and then export it to this little Arduino, and then sort  of attach this to your webcam and sort things. In a different vein this is a project called  Pomodoro Work and Stretch Chrome extension.   And what it is is it's a chrome extension that  reminds you to stretch. And the way it works is that this person trained a pose model that basically detects when people are stretching, right.   There's a team in Japan that integrated  Teachable Machine with Scratch.    So scratch is a tool for kids to learn how to code.  

It's a tool in a language and environment for children to learn how to code. And unfortunately  a lot of the tutorials are in Japanese.   But the app itself sort of works for everybody. So  you can take any model and run it there. And then a lot of other just  really creative projects like this one   is called Move The Dude. 

I can make this probably a little bigger. So it's a Unity game  that you control with your hand.   And your hand essentially is what's  moving the little character around.   So again because these models work on the web, you  can kind of make a Unity WebGL game and export it. And then just one last example.   A lot of designers and tinkers started using  the tool to just play around with funky designs.  

So Lou Verdun created these little  sort of explorations. In this one he's   doing different poses and matching those poses  to a Keith Haring painting. And then in this one   it's sort of like a cookie store, and then you can draw a cookie and get the closest looking cookie. And then this GitHub user SashiDo  created a really awesome list of   a ton of very inspiring Teachable Machine  projects in case you want to see more of them.   So just to sort of go back to the process of how  we made this tool.   We took a bit of time to think about this way of working   and informally started calling it Start with One  amongst ourselves, amongst my colleagues.   

And it's not a new idea, you know. This is inclusive design, I'm not inventing anything new. Just the word Start with  One sort of was a way to remind ourselves that   you know, the tactical nature of it.   Like you can just choose one community, one person, and sort of start there. And we're really just trying  to celebrate this way of working this ethos.  

Start with One was just an easy way to remember that.    So just coming back to this chart for a second. This idea of like a tool that we created  for one person ended up being useful for a lot more people.   

But I want to clarify that the goal is not necessarily to get to the right of this chart.    A lot of the projects that we make, they just end up   being useful for one person or for one community and that's totally okay. And it's not necessarily the traditional way of working at Google but it's okay for my team.   And, you know, this idea of sort of starting  with maybe the impact or the collaboration first rather than the scale, it doesn't mean it's the only way of working. It doesn't mean it's the best way of working.   It's just a way of working that has worked really well for- for my very small team.    So there are a lot of other projects that sort of fit into this bill.

And if you're curious about them, you can see some of them in this page   It's also a page where you can submit projects, right. So if any of you have a project that fits into this ethos, you're welcome to submit there. And, you know,  right now that times are really hard for people   I think it's easy to be crippled  by what's going on in the world.

I've certainly felt that way, maybe even a little powerless at times. And for me when I think of Start with One   I remember that, you know, even small ideas can  have a big impact if you apply in the right places.   And I don't just mean pragmatic, right. Like  human need is also about joy and curiosity and   entertainment and comedy and love, you know.  

So it's not just a practical view of this. So just a little reminder for all of you I suppose to  to look around you, to collaborate with people   who are different than you, or even people you know  really well: your neighbors, your family. I think the idea is to offer your craft and collaborate  with a One and solve something for them.   You know, even if you're starting small, you'd  be really surprised by how far you can get. 

That's it. There's this link, a tiny url digital  machine community. I'll paste in the discord.   And you can find all the other links in my talk   through that link in case you didn't have  time to copy it down. Thank you very much. Thank you so much Irene. Yeah I'll keep this up  for a little bit just yeah that's great.   Yeah I'll keep this up  for a little bit just in case- Yeah that's great.  Thank you so much. It's beautiful to see all  these different ways in which a diverse public   has found ways to use Teachable Machine  in ways that are very personal and often   you know, single scale kind, of scale to what a single person is curious about.  

Like a, you know, father and a child or, you know, people with different abilities   who can use this in different ways to make easements  for themselves. It's really amazing. I've got a couple questions coming from the chat. So  you mentioned this Start With One point of view.   Is this a philosophy that's just your sub-team  within Google Creative Lab or   does Google Creative Lab have a manifesto or set  of principles that guide its work in general?   And if so, what sort of  guides the work there, and how do you fit into that?   Yeah great question. No I wouldn't say it's  like a general manifesto or even of the lab itself.  

I would say it's a way of thinking within  like my sub-team of the lab. And again like I do want to say like it's not like I'm talking about anything new. Like inclusive design and co-design been talking about this for ages.     It's really just like a short word keyword for us to refer to these types of products, but I would say like the Creative Lab does pride itself in   basically embarking on close collaborations. So we tend to see that projects where we collaborate very closely    like not making for people, but  making with people end up being better.   So not all projects can be done in that way necessarily.   

But I would say like a good amount are, yeah.  And so I know that there's a  there's a sort of coronavirus themed project   that uses the Teachable Machine by I think  it's Isaac Blankensmith which is the ANTI-FACE-TOUCHING-MACHINE.    Where he makes a  machine that whenever he touches his face it says   'no' in a computer synthesized voice.  It's super homemade and it sort of like trains him to stop touching his face.   But I'm curious how  has the pandemic changed or impacted the work that   your team is doing or that people are working  with Teachable Machine,   or that the creative lab is doing?  How have things shifted in light of this big shift around them?  Yeah I mean that's a good question.   I don't know if I have an extremely   unique answer to that except to say that we've  been impacted like anybody else.

I mean luckily I am in an industry where I can work from home. So I feel incredibly lucky and privileged to be able to   be at home safe and not be a front line  worker. It's changed the fact that certainly   collaborations are harder to do. Like you saw in  the pictures with Steve, like we like to go to  

where people are, and like sit next to someone and  actually talk to them and not be on our computers.   And that has certainly gotten harder.   But we're trying to make do with things like Zoom and just like everybody else.

I'd say the hardest thing  is not collaborating with people in person.   And it just takes like a shift, because there's  so much Zoom fatigue and you don't get to know people outside of your colleagues.     Like when I'm trying to know a collaborator or someone outside of Google like anybody else,  you know, you benefit from going to dinner   or grabbing a coffee or just like being a normal  person with them, having a laugh.   And that doesn't really exist anymore. Like Steve actually the person that I was talking about who has ALS,   he's so funny. Like he says so many jokes. He's so restricted with the words that  

he can say because it takes him a long time to  type every sentence, you know, it takes two minutes to type a sentence, but he's so funny, like  he has such a great personality and humor.   And I think that would be very hard to come  off through Zoom starting with the fact that   it's very hard for him to use Zoom because  someone else has to sort of trigger it for him.   So certainly that type of collaboration  would have been very hard to do this year.

Thank you so much for sharing your work and this amazing creative tool.   I like to think that there are several different gifts that each  of our speakers are giving to the audience.   And one of them is the kind of just the gift-  of permission to be a hybrid, right.

Like, you know, you are a designer and a software developer  and, you know, an artist and a creator   and an anthropologist and all these other kinds of,  you know, things that bridge the technical and the cultural, the personal, and the design-oriented, and all this together.    Another gift is just the gift of the tool that you're able to provide to  all of us, you know, me, my students, and, you know,   kids everywhere and other adults. Thank you so much.

Yeah I mean final words is that it's   it's a feedback loop, right. Like I was  inspired by your work Golan.   Like you give a lot of people in this community permission to be hybrids. And I didn't know that that was possible.   And then everyone making things with Teachable Machine inspires us to do other things,   or to take the project in another direction.  

So I think we have the privilege and the honor of working in an era where the internet just allows you to  have this two-way communication. And I think it makes so many things better. So thank you Golan,  and thank everyone for organizing the conference like   Leah, Madeline, Clara, Bill, Linda. You guys all make hybrids around the world possible.

2021-05-03 18:00

Show Video

Other news