Google Cloud NEXT '17 - News and Updates

Learning Technology & Machine Learning (Google Cloud Next ’17)

NEXT '17
Rate this post
(Video Transcript)
[MUSIC PLAYING] JONATHAN ROCHELLE: Welcome, everybody. It's really late, I know so we're just going to cancel. No I'm just kidding. But I've never– I don't think in my life I've ever used a lectern, but this late I think I'm going to be like this the whole time. Is that all right? I'm really tired. But I'm so excited about this topic that I will re-energize. I'm going to talk about machine learning and learning technology, and just a pre-warning in case some of you would rather just start going out for drinks, I'm not going to teach you about machine learning per se. I'm not a machine learning expert, but I have some thoughts and some ideas on how machine learning connects to learning, human learning that I wanted to share. And you will hopefully learn a little bit about machine learning if you don't know much about it. There's a little bit there. I manage Google Apps for Education right now, and that's what really gets me into this topic.

And I like people to know me as I see myself. So I like to share pictures that I have in my head when I had hair. That's me before my interview for my first job that I did not get. And that's my Toyota Celica in the background, which I still love to this day. I don't have it anymore. But I'm a product manager at Google. I co-founded Google Docs and Drive. I was part of the original team that did that. That was 11 and 1/2 years ago. I started in 2005 at Google. And now I manage G Suite for Education and the Jamboard product, which you might have seen earlier. I'm also a school board member, probably the hardest job I've ever taken on. I knew I was taking on a job that I might regret when, instead of congratulating me, my neighbors thanked me, because that's really the kind of job it is. Nobody wants it. And most importantly, I'm the father of three students. And I think this is what motivates me the most. This is what drives me to care so much, so deeply about education and improving it, and applying technology in a way that is meaningful for the learner, not for the technology.

And so my learning values, I would say, are in three main areas, things that I really feel. First, creativity– creativity, I think, is just the basis for so much and so many things that we should learn and should teach. Discovery of passion– I feel like it's every educator's responsibility to help kids discover what they might be excited about. It's also every person's responsibility and every parent's responsibility to do that for their children or themselves. And learning through engagement– something that is actually I would say more possible now, but always difficult because engagement for one person is different than the next. So I want to step back and look at collaboration because that's where we made some advances in the products that I worked on. And there is a bit of an analogy here, but collaboration 10 years ago could probably be described as the following. First of all, it was incredibly painful. To work together was hard. There was incredible fear about what if I lost this thing?

I'm working for an hour on my machine, and I lost it, right? My hard drive crashes. And that is followed by sadness when it does happen and perhaps anger. And then, of course, you go through the many stages of– and denial is always there. And I love to look at this model and say it's very similar in education. There is so much of this that happens in education, these models that we deny perhaps the things that are wrong with it because it's actually pretty good, and it's better than it was. So when we got personal productivity tools they were so much better that we lived with them for very long. And so when we started the Docs efforts, we were going after something that we felt was very important, and that was the collaboration side. So we took something like spreadsheets, and that's where it started. And we said yeah, it's great. Spreadsheets are amazing. Personal productivity tools were so powerful, are so powerful, but what was painful and frustrating was the sharing.

And so we went after that. And so we delivered something that wasn't just spreadsheets on the web, but we delivered something that was shared spreadsheets on the web. We added that– We basically tweaked two features. We added the share button. And more importantly, we took away the save button. We like to say we killed the save button because there's no reason for us to have to take that other mundane step of saving when there's that risk, that fear that you're going to lose what you have. So we just looked at something that was great, and we tried to make it better. And this is what we ended up with in Docs, something that you're actually collaborating on at the same time. So getting to G Suite, these are the products that I was involved in from that time. I had a small company that was acquired. There was another company acquired called Writely. And we built these products up– Spreadsheets, Docs, Slides, Forms. And we were going after that problem of collaboration.

And over that time, a tremendous number of educators found what they would call magic in those tools because of how they did things differently that suddenly transformed them into not being personal productivity tools, but being learning tools– and most importantly because they were able to give students fast feedback and they were able to let students work together. That wasn't what we expected when we built the tools, but that's what happened. And about four years ago, I changed what I did and stopped managing Docs and Drive and moved over to Google Apps for Education and to move over to try to build new productivity tools. And that's when we launched Google Classroom and Google Expeditions, the VR tool that's used in classrooms, and Jamboard. So that was my involvement. So you look at what's gone on in technology. You now have the world's information. It's at your fingertips. It's literally at your fingertips on your phone. It's even– perhaps you don't even need something on your fingertips to do it, right?

You just use your voice. And you get what you need. You get the answers. The answers are all there– the world's information. And then you add something like Drive, or any product like that, and now you have all of your information. So you have the world's information and your information accessible and useful. Now what? OK, that's all there. When you think about what at least I did in school is I learned by memorizing some of that. The world's information I'm memorizing. I'm working on memorizing, and my kids still the 50 states' capitals in the United States. I don't know why my kids are memorizing the 50 states' capitals. I really have no idea. Because they can walk into my kitchen and say OK, Google– oh, sorry if your phones all light up– what's the capital of Nebraska? I know it's Lincoln, by the way, because I was taught that. So we have that. But now what? How do we get to the self-driving car of education? How do we make it so we take that next leap?

What should our students learn– facts and previously known algorithms and how to do that rote step to do, say, long division? Or should they learn problem-solving skills and tools for problem solving to improve life, to actually discover and create new things, to invent? Now learning, has it changed? When you look at how learning has changed, I want to step back and say, well, what kind of learning? Machine learning versus human learning– how has learning changed in those two areas? So let's look at machine learning. Let's see if we can learn something from that. So computers are great at following instructions. It's what turned me on to computing, to computer science, when I was a junior in high school. Absolutely fell in love within two days of a computer class– the called it "computer math" back then– because I realized that the computer would do whatever I said. I could figure out how to give it a command, and it would do it. It would carry it out. And computers are just great at that.

You tell it what to do, and it does it. And that's great. Computers are good at doing math. You can just ask it a question, and it'll give you an answer. That's just what it does, right? That's what the computers do. And remembering facts– it's got every fact. And I say computers lightly. There's an easy way to get any fact from the internet. And repetitive tasks– give it something repetitive and mundane. The computer will eat it up– humans not so much. They'll do it, but they won't like it. Or most people– some people do actually like repetitive tasks. That's OK. And then we made computers more powerful. You add speed and data capacity. And over the last couple of decades, data collection and machine learning specialization, a miniaturization in cooling and power consumption. All those things came together for us to take a leap in computing. And that leap was actually quite noticeable. And then humans were able to teach computers because of that power to do things like recognize patterns and classify data.

It's why we can point our phones at a sign in another language and get an immediate translation. You can do that today, and it's amazing. And it's natural now, the power we have in our hands in our phones because of that miniaturization. And the power and the computing, in conjunction with the cloud, gives us that. And we can take advantage of that. Clustering data, predicting values, ranking relevance– there's so many things that we were able to teach computers to do because the power was there. And so when you look at machine learning, there are two models. And I'm using something lightly here so don't worry about the semantics of "artificial intelligence." But certainly we called it artificial intelligence for many decades. And that's basically making machines "intelligent" by giving them facts and letting them be able to make the next step by saying, if this, then that. Machine learning is making machines that learn. And you could still call that artificial intelligence, which is why I'm saying don't worry about the semantics.

But those are the two modes of machine learning. But if you look at these two things and you want to compare them, the old way, what I'm calling AI here even though you can say AI is everything, is just basically saying, if then else, but so many times that a human could never do it. So if this pattern exists, then do this, else. Does this pattern exist? Then do this, else. If this pattern exists– and it just looks for thousands and thousands or millions of patterns and is able to mimic. And that's what I would call an expert system. And that is the old way. That's the way we used to think about making computers intelligent. It's the best thing we had. And the new way, the machine learning way, is much more about writing computer programs that learn from examples– so trying to classify data, classify emails. So it actually modifies itself based on patterns it sees, and by giving it answers, by saying this answer was right or this answer was wrong. So if you've never heard of Joel Benjamin, you've probably heard of one of his students.

And one of his students was Deep Blue. Deep Blue was the IBM computer that learned chess from Joel Benjamin. Joel is an expert at chess, and IBM hired him to help teach Deep Blue how to play chess better than any human. Now if Joel Benjamin wasn't better than every other human, how did he teach that computer to be better than every other human? It's because Deep Blue had a much broader capacity. So Mr. Benjamin could give it all of the knowledge through games. Yes, if this happens, this is what I would do. If this happens, this is what I would do. If this happens, this is what I would do. So it's some level of patterning, but it's mostly an expert system. And in 1996, Deep Blue was able to beat the human chess champion. And it was an amazing moment. That was 1996. 20 years later, another product, AlphaGo, using machine learning was able to do something that in 1996 everybody thought would take many decades to achieve. It took two decades. That's a long time, but not that long in the scheme of things.

And that is to beat a champion Go player– does anybody here play the game Go? OK, I respect you. That is a hard game. I don't play it. It's a really hard game, and it's very complex. And I understand why people thought, no, a computer will never beat a human at Go. Very, very complex, but that happened in 2016, just this past year. AlphaGo was the program that was able to beat the human player. And to give you an example of the complexity increase that happened from 1996 to 2016, I just want to show you a visual. So if you look at the dot on the right versus the dot on the left, that's the difference in complexity. The dot on the left versus the dot on the right is the complexity difference between the number of moves in a game of chess versus the number of moves in a typical game of Go. And that means the dot on the left doesn't even justify one pixel on that big screen. That's the difference in the complexity. To put it another way, if you look at the number of moves in a game of chess, it equals the number of atoms in the human body.

But if you look at the number of moves in a typical game of Go, it equals the number of atoms in the universe. That's the difference. That's the leap that was taken from '96 to 2016, 20 years. And there are different flavors of machine learning. So this is just an understanding of what machine learning means. The first is learning by example. They call it supervised learning. So something like a spam detector trained on very simple examples to say yes, this is spam. No, this is not. Yes this is spam. No, this is not. And it's very, very effective and is used in many machine learning examples that you see in your life, including, if you use Gmail, Gmail spam. And it's pretty good. Gmail spam detection is pretty good. The next is unsupervised learning, which is more like pattern discovery. And it's very hard, but it's clustering. It's a method to cluster data, like these things look similar. These things look similar. These things look similar. But if you don't have examples, that's a good way to go.

And then the next one, probably the hardest, is reinforcement learning. And that's feedback, but after a long string of things. And that's more like when you win the game of chess, and saying, OK, that long string of moves that happened, that worked. And you have to look back and understand why it worked. So it's very slow. A lot of people in machine learning would say that might actually be the next big thing. It might require the next leap in capacity and capability. And if you're really interested in these models, there's a ton of research on research.google.com. So just some examples– speech recognition research. Googlers who do this research share a ton of it here. And so pedestrian detection for self-driving cars– that's machine learning because there's so many patterns of what a pedestrian is. There's movement patterns, there's shape patterns, there's things that actually require more than just saying, oh, I recognize a human, because humans come in many forms.

Things like natural language understanding, machine translation, even advertising. There's so much research there. So if you're interested in the topic, definitely go there. Some of these examples– I gave Gmail spam before. It uses of what they call a feed-forward neural network. And it intercepts now 99.9% of all spam. And every time you hit the yes, this is spam button, you're helping teach that model. Thank you for doing that. You keep my inbox clear of spam. And I try to do the same for you. Speech recognition is another example, taking acoustic input and putting that through a deeper current neural network and getting the text output, figuring out what that text is. How many people here spoke to their phone today? Did you speak to your phone? I do it almost every day now– if nothing else, just for a quick reply to an SMS or something like that. And reduced transcription error is now– or it is becoming better and better. Google Photo Search is one of my favorite examples.

I try to be unbiased when I tell people to use Google Photos. It's really hard. They don't believe me that I'm unbiased. But I truly believe it's one of the best products you could use if you care about your photos and you take too many, which we all do now. I was much better when I actually had to print the photos and get them in print at managing sharing and things, because I only took a few and I shared those, and I cared more about them, I think. Now I just snap millions of photos, and I don't really look back at them that much, but I started to when I started using Google Photo Search because now I can look for things that were not tagged because it actually recognizes patterns in the photos through machine learning. So this has improved my life. So I can search now for "beach," and it'll find pictures of my kids on the beach. I didn't tag these. I didn't do anything to tag them, but it can use machine learning and the algorithms that are there to do that for me.

Or "skiing"– and I can find pictures of my kids skiing. so if I say to my mom, you got to see your grandkids skiing, I don't have to actually think, oh let me go tag them and whatever. I just do a search and I hand her my phone. The other way it improves our lives, and maybe you guys will recognize this, is– anybody you know what that is? Snapchat? Otherwise known as Snap at this point. Did you buy the stock? I didn't either. But this improves my life, right? Now I can sit at breakfast with my daughter and have a lot of fun. This is like literally– I'm not kidding when I say this improves my life. I'm definitely laughing more than I used to. But it actually is something that I appreciate, but obviously it's not improving my life necessarily. It's entertainment. But I will say those same concepts are improving human life. And this is where I want to make an association to learning. This woman is Britney Wenger, and in that picture she's 17 years old, when she won the Google Science Fair for discovering an incredibly effective method of diagnosing breast cancer.

And she did that by learning coding, machine learning, big data, and big data analysis, tools that were not taught in school. But she took it upon herself, not because she wanted to win the science fair, but because she wanted to improve life. And that's truly impactful, a little bit more than Snapchat. So we've done an amazing job teaching machines, really amazing job, and it has helped us as humans. But what about humans? How has learning changed over those 20 years? We saw how much it's changed, and who knows what it's going to do over the next 20 years? Somebody asked me today, do you think your 10-year-old daughter will get a driver's license? And of course at first I'm thinking, why wouldn't she? And then I realized what he was asking me was will she need to drive? Or will self-driving cars be the norm? If they're truly better, why are we driving? And he's in San Francisco, and he said, you ever notice how many cars are parked on both sides of every street?

Imagine if that wasn't so. Imagine all the room we would gain. Imagine the efficiency. Imagine the change in pollution. Everything changes, when you think about it, if humans don't have to drive, park, do all those things. And literally the accident rate goes way down because humans are really bad drivers. We don't set a high bar for machines, let's just say. But if you compare human learning to machine learning, and this is just a visual, not necessarily fact. I don't have numbers to prove this chart. But where have we gone in 20 years? How have we changed learning in 20 years? I think it's improved, but it has not improved or changed enough. And it has not improved greatly. it's not noticeable. I can't really show it. There are instances I can show. I've seen amazing educational instances– people, situations, classes, methods. But they feel like they're spot changes. And where is that graph going to go? How are we going to change that graph? I want to give you an example that I feel really proves the point of something that I think is wrong with education.

This is a math problem. There are 26 goats and 10 sheep on a boat. And the question that's asked is, how old is the captain of the boat? This was a question asked in the study of a bunch of fifth graders not long ago. 75% of the fifth graders asked gave a numeric answer to this question. They actually answered the question. They tried to answer it. Sometimes I like to give away a Jamboard or something to person anybody who can get the answer. Because I don't think there is an answer. These are the answers the kids gave, though. 75 percent, most of them, gave the answer 36. [LAUGHTER] Clearly the captain of the boat is 36. And you know why? Because I have two numbers. When they were asked, why do you think the captain is 36? Well, in these types of problems, we usually have to add the two numbers. 16 is another answer they give. You know, sometimes it's tricky, and you've got to subtract them. And then there was one or two instances of 260, where they're doing advanced math and thinking, well, I might not have learned that in school yet, but multiplication– I can do this.

But this is the proof, right? They're programed. This is the irony. Our kids are programmed, even though they haven't really advanced as much as machines have improved in their learning, because we teach them with rote methods. We teach them to do things in rote ways. Follow steps, not solve problems, and not to understand. And what can we do to help solve that? What are some of the things? And these are some basic things. But first of all, we could help train teachers more, give them the tools. Some of this happens because teachers are teaching the way they were taught to teach. And our teaching programs haven't really changed that much, or haven't changed enough to include today's tools, new tools, new ways of teaching. And giving children and students and people– this is not just about kids– experiences, not just facts. Let's let them experience something. Like I said before, with some of the data model, the machine learning, we're giving the machine experiences, saying try this.

Oh, fail. OK, now we told you it failed. Try again. Success! Try again, fail. Try again, success. You can't get that necessarily unless you have experiences. And there are methods of giving experiences. I even show Google Expeditions here. It's a precursor to something I'll show later. And experts help, too. Connect experts with students, with learners. Let them experience with that expert has experienced. And I show basically Hangout, Skype, Twitter as examples of the technology that's there to help with experiences. Sometimes it's just social networks. Like, sure, they're useful in some ways. But they're useful to connect you to experts. And educators can actually reach out on those networks to find experts when they are not the expert. The teacher just has to be the guide, not the expert. And games can help, too. Games help a tremendous amount because they motivate learning. They help to motivate learning for everyone, not just students. And I just have a quick example.

I have a friend who's a driver who's not so good at driving, one of these examples of setting a low bar. But then she got a Prius, and the Prius gives immediate feedback and a score. And as soon as you see a score, everything changes. I want a better score. Slow down. OK, I slow down, I get a better score. Oh my gosh, I'm learning. And it happens immediately. I actually did this in my car recently. I just changed what my dashboard would show to show my efficiency, my fuel efficiency, and it's a gas car, unfortunately. But I found myself driving so well. I was so careful to not accelerate too much and to not necessarily overuse the brake. And that means I'm going to keep a bigger distance because then I can be more efficient. And I started playing a game to see how high I could get my miles per gallon efficiency. And not only was it fun, but the people in the car were like, wow, you're driving so carefully. What happened? And it really was an obvious change. And then my son, I talked to him from where I was driving to, and he said, what was the efficiency?

What was your score on the way down? He even cared. It was just a game because it was fun. And grades could reflect just learning, not everything else. So it is important to know if you're right or wrong. We said experiences are important. And to know whether or not it was the right answer or wrong answer, sometimes it is important to get that feedback. But we so often incorporate everything else into a grade that we give to a student. It's like what happened? I thought you really knew your chemistry. Well, I was late on three assignments. It's like, oh, so we don't really know if you learned chemistry. No. Well, that's a pity. Oh and I was late four days. I had an A+, but I was late four days, so I got a B. There's a good idea. So what are we doing in learning technology today? And I would say learning technology today has been focused, and there are some instances that this is not true, and they are more advanced. They're just not there enough. They're not everywhere.

But what we're doing generally in learning today is focusing on the basics. So one thing we know is amazing in education is immediate feedback. Teacher-student feedback is so important. That's why we saw something like Docs, and why I started with that being so impactful for teachers because they realized they could give immediate feedback to 100 students without being there. But what was frustrating them was the ability to share. And that's actually why– in other words, they have 100 students. How do I share to 100 students and get that feedback to 100 students quickly? And it was super hard. That's why we created Classroom. Google Classroom was created on that foundation to say, if we could get them there faster, they'll be better teachers. They'll be happier. They'll spend more time with students, and they won't be so inefficient. We tried to get them off paper and onto something that was easy to give feedback faster. Same thing with field trips. Field trips are amazing experiences.

It's the kind of thing that when you look back at school, and you think about what your remembered, field trips are memorable. And it's not just the noisy bus. That's memorable in a bad way. But it's incredibly enriching. You really feel it. And there is some proof and research that shows that when your body is involved, when your senses are involved, you remember more, you learn more. You're open to learning. But what's frustrating about that is it's hard to go on field trips, really hard. They're expensive. Not every kid has the same opportunities. I've seen in my own schools, we've cut back on field trips for cost and just for access reasons. You don't have enough children participating, so you can't really do it. And you're usually limited to where your bus can go, even if it's a three-hour ride. And that's why we created Google Expeditions specifically out of the team that's focused on education, so that we could at least try to give children more experiences using virtual reality.

And now it's easy to argue that that's not the same, but it's pretty good. And we know it's pretty good because we've tried it in hundreds and hundreds of classrooms. And especially, even before we launched it, we tried it with some teachers and students, and invariably the experience was amazing and the engagement was super high. And when we combine that with good lesson planning, it's really taking off in a way that is great for students and great for teachers. And there's hundreds and hundreds of expeditions now being used in classrooms as part of lessons, and the children are experiencing more. They can go to Rome. They can go to the Great Wall of China. And it can be for a math class, not just a history class. It's not just about getting to that place. It's about learning using that place. Whiteboards– and it's funny because I never really made this connection until I saw what was happening, but whiteboards could be amazing. But again, there's things that would protect the teacher from using it in the way we would like to see them use it.

There's no save or recall, no sharing. You can't add imagery necessarily to a white board, unless you're an incredible artist. It's not legible. You've got ink and all the things that go wrong. So we produced Jamboard. That was really for a business environment, but we're seeing and learning obviously, and this is not a new concept, it's something that really enhances learning if you can use it the way it should be used. And that's what we're doing in learning technology. But what could we do if we really focused? What could we do to enhance learning using technology and advanced technology? First of all, there's an incredible number of tools. I mentioned Brittany Wenger, who's now a researcher. I think she's still with Georgia Tech, but now a researcher, a medical researcher in her 20s now. So it was I think it was 2012 that she won the science fair, if you want to look that up. But there are tools out there now, and you saw probably some of them announced today and produced even before today.

TensorFlow, natural language APIs, big cloud data, analytics, Vision API to pull information out of videos, something that would take just forever if you had to do it manually. And maybe there's many things that can be done if students are taught some of these tools. And if you're thinking, yeah, but what student could actually learn those tools? Isn't that too advanced? The answer is no, they're not too advanced anymore. The bar for using those tools has come way down. And what we often do is we project what we feel is our own limitation on our students. And we think, wow, I don't get that. That doesn't mean the student won't get it. I found that my 16-year-old learns much faster than I do. And he'll get things before me now. And I actually remember that moment with my dad, when he was frustrated with me passing him in my ability to do things. And it always made me feel good to teach him. And now I'm lucky because I feel good when my son is able to teach me.

But we have to give them the benefit of the doubt, which means letting them learn things that are more advanced than we knew and what we were taught. What we learned in school is not what we should be teaching our students in school. It's just not the same thing. And unfortunately, that's not where we are today. We're basically teaching them what we were taught. So these tools are actually useful. And imagine how many Brittany Wengers we could have if we taught everyone those tools that she had to go learn on her own. She had to even know they were available. One of the things we could do with those tools is match learning to the learner. I like to say if you've seen one student, you've seen one student, but we teach them all the same. And I know that's not true in every case. Sometimes there are breakouts. There is a way to group children or give them personalized learning with some tools. There are some amazing tools out there. Like I said, they're just not pervasive enough.

But we could do a lot more if we'd collect the data. So first we have to recognize what data should we collect, and then we have to say, what can we do with that data? But this is one thing that we can make incredible advances on if we really focus. And there's three areas that I would say you could match the learning to the learner. First is the content. What are you actually using to teach? Then the format. How is it delivered? Is it physical? You're out in the field. Or is it video, or is it reading? I know myself and my wife, we learn in very different ways. She's a reader, and I lose consciousness when I read basically. I'm a visual learner. And the pace, the pace is always different. But a slow-paced learner is not a worse learner. They are just a slower-paced learner. They might absorb more. The other thing we could do is provide more immediate feedback. And here I want to give you an example of something that actually exists. It was launched very recently. And I just took a few screen shots, just earlier today, to show what can be done with this.

It's called the Perspective API. And when you leave here today, or if you have your laptop and you're really bored, you can actually go to the Perspective API and try this. There's a lot of things you can try there. One of the things you can try struck me immediately as something that can be applied, and not just education generally, but specifically educating people, and especially students, how to communicate more effectively. What could be more important than helping us communicate more effectively with each other? How often do you think, wow, I can't deal with that person. Or that was such a tough discussion. Or, wow, did you see that email? Can you believe that? I can't believe that was the response. Or the reply all that you wish the person could read before they sent. The Perspective API was launched out of the Jigsaw team at Google. And Jigsaw used to be called the Google Ideas team. And they have very high I would say aspirations to help humanity, literally.

And one of the things they were trying to do is reduce toxicity in communication. And so if immediately you're thinking, oh, you mean YouTube comments? You're actually right. That's one of the things that they actually tested it on. But others, like the New York Times and Wikipedia, they tried it. And they tried to train that model to measure toxicity. And they have a tool online that I would love to see wired to every communication method that we use, even Snapchat. And I'll give you an example. If I said, "you are so stupid" in response to somebody, that gets a 97% toxicity rating– 97%. But I bet that comment is out there in the ether a million times. You are so stupid. And basically it's saying 97% similar to comments people said were toxic. So they had people, just like the spam thing, say yep, toxic, toxic, toxic, not so toxic. You know, how would you rate this? And that model was trained against this data. If I just change that to "that is so stupid," not you, that, it goes down a little bit– 94 percent.

All right, not quite enough. I should be a little less toxic still. But if I change it again to say "that is so dumb," now that gets a 76% toxicity rating. I'm learning. I'm learning to be less toxic. And then if I change that to say, "that might be a bad idea," that goes down to 12%. And if I go even further and say I want to be even less toxic, I say "that is not a great idea,"– 2%. Send. Imagine if every student, instead of saying, no, you cannot use chat, or you cannot use email– how many schools shut down email? They say no, the students can't send email. They can't talk to each other on email. A lot of schools, if you're not involved in education, a lot of schools basically say, we don't actually give our kids email at school, especially in the younger grades. But imagine if we did and we helped them learn how to communicate better. How many bad adult communications could you avoid in the future? So that's just an idea of something you can do with Perspective API.

And there's so many other things out there that you could do that with. There are so many other applications. And if you think through some of the things that are announced here or at the next industry API conference or machine learning conference, there are so many ways you can look at those things and say, how can we use that in education? How could a learner use that? And I would compel you to do that because there's also opportunity there. But if I look at learning technology, I like to think of it as there is the now, learning technology, and that's where we've been. So providing engaging experiences, reducing teaching friction, and relevance for students futures– in other words, learning technology, the stuff that we use to learn, is relevant because kids will use it. If you look at the verb learning technology, you want to learn technology, which also means teach technology, if teachers model modern practices, students will learn it better. If they see a teacher handing out paper, collecting paper, and taking a week to give feedback, that is not a good model for a student to think, oh, that's how it works.

No it doesn't. It doesn't work like that at all. And if they're forced to sit there, every single student, and learn cursive writing, and feel free to come up here and attack me later about cursive writing if you want, I'll argue till I'm done. Every single student learning cursive writing– I would argue that that's not necessarily relevant to their future. It might be if they're an artist. But teachers modeling modern practice is important for learning technology. In other words, learn it. And discovering advanced tools– learn it. Learn those tools. Teach those tools. And then learning how to apply that technology. How should I use it? Using relevant examples and saying, this is how we use this technology. I want my kids to learn spreadsheets in the third grade or the second grade so that they understand, oh, yeah, that's how you do that. That's how you actually– first of all, that's what data is. They don't even know what data is usually when they get to, say, the ninth grade sometimes.

And it doesn't matter what tool they use. If they really want to use Excel, I'll allow it. But if they want to use Google Spreadsheets, I'd be happier. But I'm just giving them as an example. Spreadsheets are super useful in our lives. They should learn it super early, and they can. Trust me, I've seen it. First graders, second graders, third graders learning spreadsheets is totally doable. And that's just one example, a simple one, but learning how to apply it. But I would say you go to the next level. And what we're at today, what we're trying to do, is improve and just make learning better. And to get better learning, we take those two things and we go to the next step and just basically focus on learning to solve unknown problems, not learning to follow the examples of our predecessors that solve problems and say, oh, yeah, this is how this problem was solved. Follow these steps. Learn these steps. Maybe there's a new way to do it. And learning to learn– figuring out what do you do when you don't know something.

And trying not to just answer students with, oh, here's the answer. This is the answer. No. Here's how you go about finding an answer. So if we don't allow our students to use their phones in school, they don't know that they could look for that answer much quicker much more quickly. And they don't know that that's just a fact that they could look up. Why don't we let them do that? We find value in them memorizing that fact instead? I don't. So I want to see us teaching students to learn. And I'll end it there. Thank you very much. [MUSIC PLAYING]

 

Read the video

In this talk, Jonathan Rochelle provides a history lesson of how humans taught machines to be intelligent. He then applies those same concepts to how humans teach each other, and what this means for the future of our schools, classes, and the technology we use for learning.

Missed the conference? Watch all the talks here: https://goo.gl/c1Vs3h
Watch more talks about Collaboration & Productivity here: https://goo.gl/q3WbLc


Comments to Learning Technology & Machine Learning (Google Cloud Next ’17)

  • I was in the Google Next, and this conference of Jonathan was the best of all in my opinion.

    Gabriel Marticorena March 11, 2017 6:55 pm Reply
  • confirm prev. post. I was there at next and this one was the Absolutely the best! Great Jonathan!

    Fabio Perotti March 16, 2017 11:07 am Reply

Leave a Comment

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

Loading...
1Code.Blog - Your #1 Code Blog