Duration 40:11
16+
Play
Video

Opportunities, challenges, and strategies to develop AI for everyone

Daphne Luong
Director of Engineering at Google
+ 3 speakers
  • Video
  • Table of contents
  • Video
2018 Google I/O
May 10, 2018, Mountain View, USA
2018 Google I/O
Video
Opportunities, challenges, and strategies to develop AI for everyone
Available
In cart
Free
Free
Free
Free
Free
Free
Add to favorites
7.95 K
I like 0
I dislike 0
Available
In cart
Free
Free
Free
Free
Free
Free
  • Description
  • Transcript
  • Discussion

About speakers

Daphne Luong
Director of Engineering at Google
John Platt
Software Engineer at Google
Rajen Sheth
Senior Director of Product Management at Google
Fernanda Viégas
Research Scientist at Google

Daphne Luong is a director of engineering at Google, where she leads efforts in natural language understanding and human computation. Her expertise is in building technologies from the ground up, with experience across a range of startups and established companies. Daphne also worked at Nest where she was responsible for IoT cloud platform, security, eCommerce, energy services, and Work with Nest program. At Microsoft, she led the spin-out of the Tellme IVR business into [24]7.ai where her global team was r

View the profile

John Platt is best known for his work in machine learning: the SMO algorithm for support vector machines and calibrating the output of models. He was an early adopter of convolutional neural networks in the 1990s. However, John has worked in many different fields: data systems, computational geometry, object recognition, media UIs, analog computation, handwriting recognition, and applied math.

View the profile

Rajen Sheth leads the Google Cloud Artificial Intelligence product line, and he focuses on making Google’s AI technology easy and useful for developers and enterprise customers. Previously, Rajen led the development of Android and Chrome for business and education, including the Android for Work products, the Chromebooks for Education product line, and Chromebooks and Chrome browser for work.

View the profile

Fernanda Viégas is a research scientist spearheading the convergence of design thinking in AI practices. She co-leads Google’s PAIR (People+AI Research) initiative, which focuses on improving human/AI interaction and responsibly democratizing machine learning technology. Fernanda is well known for her contributions to data visualization, and the systems she and her team have created are used daily by millions of people.

View the profile

About the talk

Fernanda Viégas is a research scientist spearheading the convergence of design thinking in AI practices. She co-leads Google’s PAIR (People+AI Research) initiative, which focuses on improving human/AI interaction and responsibly democratizing machine learning technology. Fernanda is well known for her contributions to data visualization, and the systems she and her team have created are used daily by millions of people.

Share

Thanks for coming. My name is Fernanda Villegas. I'm part of Google brain. I'm also a leader in the pair initiatives which stands for people plus AI research and I'm very excited to be joined today by this wonderful panel. So here to my side is rajen sheth. He's the director of product management at Google Cloud AI. Definitely along engineering director at Google and John plant director of applied science in Google AI and today we're going to talk about opportunities challenges and strategies to develop AI for everyone. So if you could please I would love for each

one of you just to talk very briefly about what do you do a Google Driving shaft and I I run the product team for cloud Ai. And so what we're doing is within Google Cloud, we're trying to figure out how can we bring the best of AI to developers and turn into Enterprises? And so how do we give give developers a great platform to build on but then how do we take some of the best of Google's Ai and make it so that are available is services for for Developers? Thank you. All my name is Daphne long. I work in Google AI he went country

Tatian Sada and natural language understanding. Lunch on Platte and I help run the applied science or inside of Google AI we're super excited to buy all the opportunities in science related to things like biology or physics and we believe that computer science, especially in machine learning and AI can really help accelerate that that's what that's what my work desk things that is very inspiring to me about this panel and about a lot of the themes going on at I owe this year is a i for everyone and so I know that impair one of

the one of the main research themes we have is how do you design human-centered AI technology? How do you start with a user in mind and then design machine learning technology that stay with that one of the things I'd love to ask each one of you two to talk about a little bit is what does AI for everyone mean to you? So I for me, it means really how do you make a eye easy and useful for for for people that want to use it and I think they're two big problems that we're seeing right now with AI one is that there isn't as much skill

out. There are as much knowledge about how to build models especially deep learning is there should be and so we're trying to do both built tools for that and touch her to make it to the more people can access AI but then but then also can also learn about to learn about a giant in India and figure out how to use that what we're finding for example is that there are probably in the tens of thousands of people out there that know how to use deep learning problem the order of a couple of million to date data scientist out there, but there are 21 million developers. And so our

goal is how do we get AI to be accessible by the 21 million developers in the second part of that is useful. That's how do we actually make it so that way I is useful. How do we go especially for businesses Beyond kind of where they can only do cool things but they can do things that are useful that actually are vital to their business to me getting everyone involved and aware. That's what that algorithms have representation for all users right to own the data type. This means that we have to say that the representative of all users

for everyone is trying to get the benefits of a i to a large part of society or across the world. So we do is we look for leverage points, right? We're not trying to get everyone in the world to use AI we're trying to get them to have their fruits of deserve what they I can provide so if we can find these leverage points, which where I ate local application of a I can really help the whole world. That's what we're looking for. That's why I'm super excited about a flying Ai and machine learning to problems in science because it was

science. I need to breakthrough Technologies. I think one of the interesting things here for this fan off and the steam is that each one of you is coming from a very different perspective, right? So, right and I think about you is coming from the product / business side of things. Definitely your you're helping build some of the fundamental blocks like the data the fundamental data creation blocks for it being able to do this technology at all. And then John you're coming from a research perspective, right? How do I enable the scientist Halloween able science at large? So one of the things

that comes up whenever we talk about AI whenever we talk about machine learning is this notion of representation and I want to start at the beginning with that. I want to talk about data and so deafening. I was really excited when I heard that you were going to be part of this panel cuz I think about you as the data star of Google and so your work is a lot about Creating better ways to scale up for instance. How do we get other data? How do we do it in an insightful way? And so one of the things I'd love for you to talk about a little bit is

that the representation of data sets at Google? What is what is that look like today? So having some meaningful and systematically correctly calibrated label data escalates really crucial for machine learning and from a representation not you know, well balance, for example, if you don't get Wikipedia Theta Chi the pronouns he is mentioned so much more than she right. So I'm inside Google. We have a lot of data and examples set that I want to talk about Easter audio. They just said that we open source last year. The data says had

2.1 million video with 8.5 hour of audio battery annotated we around 500 classes of Staff. That's pretty interesting really like you have sounds so unlike speech and music all the way to light up engine sounds or burping right or Godly. So then from that perspective it's really important to have audio and sound from all over the world. Right? I haven't found some kids from India's very different from China and the us so that kind of thing is what I think you know, what you think about when

you make Play-Doh sets So that's when it went when I went when I became aware of all the different open open source data sets that we have put out. I always really it's really exciting to see audio video tagged you have translation you have you have all of those? But whenever you're talkin about data in machine learning, you're also talking about challenges, right? So one of the things you kind of started touching on a little bit and I want to pull that out more is this notion of representation making sure that the data set has a good

representation from different users different communities and that's a hard thing to do. So, I'm curious. What are we a Google doing about that? How are you thinking about that? So, you know Google users and communities, right? They have your very generous donating data and so forth to make out of a product better and example is Google guy. How many of you that Google guy in the audience? Right? So yes, so on the theater front we are also do you need something very similar? We have a crowdsourced app, which is a nap and on the web right now since the last two year. We actually

had two million user donating to the app. I'm from all over the world that you need 233 Tenchu. That's a lot of different people. Right and we had 200 million donation from people answering questions all the way from. Hey, can you see the sign from the street for a 11 Indonesia City versus hey, how do you eat this sentence? What's the sentiment of this sentence in Hindi? Right so that all those data and Toes. Actually help us make products better help people navigate Street where there's no sign life. Unlike Landmark navigation making the kibo

more accessible as well as making your heavy more detailed labeled beta said so I think that's really important. If you don't have to crowdsource app, please download it great. Thank you. I owe one of the things about audio that I had never thought about until I read this is that there's a huge discrepancy between even voices of adults. Versus children and children is it so you have to think about many many different dimensions of diversity, right? And then falling from the app the country and initiated

Cruz stats in 35 countries are actually really excited about the whole contributions. Awesome. Okay, moving on a little bit about from data one of one of the things whenever we talk about AI for everyone one of the I think first questions that comes to mind is how do we make AI accessible and better for our users and Rajan here? I'm going to turn to you and Since you you are helping lead these services on cloud, right? How do you think about designing these ml tools?

So that the so that they are a easy for users to use and then ask a question. There are two ways you could talk a little bit about what are the difficulties that you're starting to see what are the patterns that we're students get users got stuck to two big parts of that one is how do I make it easy for people to create models? But then the second hardest how do we make sure that those models are fair are not biased are are are a really kind of serving the purpose in the right way because it is very easy for people to just take what they've

been doing manually and end up including by us into a into a model. So on one part of it, what we're doing is worth providing of of writing tools. One of the most interesting things they were working on is cloud automl and soap. Mel is a way by which you can give us a dating site the other side and then we'll use machine learning to create a machine learning model for that dataset the first area we're doing this with his image recognition where you can give us a set of images that are labeled and then we'll give you a highly actor and model on top of that to to predict for 4 images. And so

that is really help us expand. We have 15,000 people that have signed up for access to that and we're singing Amazing use cases everything from really large companies to very small companies who are doing things around the upper example helping Farmers manage their field or helping track litter and and making sure the track litter back to to the producers Freaky's do I as a user need to have if I want to use auto aim out to be able to use it. Are you why it's very similar

to say Google photos you upload your pictures. You label them you can we actually are integrating with that 11:00 services Daphne and team are using to have us label it if you want that and then you just hit train and then we'll create a model for you. However, that's sad getting the right data set is the key part. They're so even if you don't have to go create the model, you have to create the right data set and said there were trying to provide tools to make sure that people are are able to create the right to do so, I can figure out where their flaws in the data set and then on the other

side of this around things like Baez, we're trying to figure out Tools around Batman also helped advise customers with with best practices to make sure that the day that they're putting in can actually produce the right results data continues to be one of those bottlenecks reduce the foundation of you put in the wrong day that you end up with the wrong results. John one of the one of the things as you all have been seeing is diversity diversity and datasets. Also, how do you create these experiences for

users who are not themselves machine learning experts to be able to use this technology. I'm also curious to hear from you about diversity and representation on the side of machine learning develop reason researchers. The reason I'm looking at you is because you're working with scientist. And so these are people who are going to be potentially needing custom model. They're going to be pushing the envelope envelope sometimes but they may not themselves be machine learning experts. So I'm I'm curious about your thoughts. There will be increasing with

data-driven a lot of scientists like physicists to geologists to biologists. They do their job by gathering large amounts of data, so they're already kind. Getting into data science and so is our big opportunity right now for them to use machine learning to kind of make themselves be much more productive the way I look at it is it's kind of like giving a scientist thousands of undergraduate assistant another look at you fly. If you if you if you could examine no fly to cells you can just have thousands of assistance to look at it for you. That's enough one example. So

I'm I really I don't have answers your question, but I really excited that we can build kind of model to help scientists use specific things to help accelerate their their research and when we we actually collaborate with Scientists to help build these kind of normal, so I don't know what kind of scientist well of the Light Side people. There's a bunch of us working with biomedical research trying to understand the core research of disease. Very generous diseases. There's a bunch of us working on climate and energy.

So we're working with a company called Tau Technologies trying to figure out whether we can make Fusion Energi make it a commercial Development Across biology physics chemistry, we work with a bunch of people who seem to like quantum chemistry the same question. I asked Rajan. I'd like to ask you which is what do you see as the biggest difficulties for scientists to do this kind of work today. Are there are there are kind of pattern. Is it understanding how to put together models? Is it what what do you see

between scientist and a computer scientist, but I think they would prefer to have a productivity tool. Can we help them build a productive? For their own research and either they can try to build themselves if they know machine learning or we can help them. Very excited about is a few months ago folks in in pair in collaboration with others that Google launched tensorflow. J. Yes, and this is really significant in the sense that we what we were talking about it in terms of democratizing the technology and bringing it

bringing machine learning to the web right tensorflow. JS speak the native language of the web speak JavaScript. And so now you have a whole new set of developers who can start to take advantage of of this technology and we are already starting to see some interesting applications for instance. Developer who decided to create to train a model based on his web camera to make the mouse move in different directions just fight where he was looking. So if he looked up the mouse would go up if you look

down to Mouse would go down and the reason he did that was because he was trying to help a friend who had suffered a stroke and became a paraplegic so that to me again starts to talk about this possibility of bringing this technology into your into your personal at what do you need? How can you train these models have fruit for your specific purposes? So I think As we talked about democratizing the technology as we talked about bringing it to everybody.

When is the things that I think we had Google are very mindful of his that we we have to do that with insight and responsibilities rights. You want to be thinking ahead. So one question I have Rajan is it is since you are providing this service, right? What happens if you have Bad actors trying to use your service and what states guards could you provide is something that is very much talk of mind for us because as we seen more people use AI we want to make sure that we're providing those safeguards with whether it's kind of safeguards or or

best practices to make sure that the Fitz using right way on one hand. When you have things like open source Technologies, it's very hard to prevent people from from doing bad things if they want to do bad things but one advantage we have a cloud is that we we can we can put in terms of services are acceptable use policy of things like that to make sure people are using this in in the right way. And I think one thing we're discovering with with AI is that we have to rethink that it's it's not an infrastructure technology anymore. We have to rethink their the uses and that's something

good that we're doing quite a bit. Another thing that worries me to is beyond Bad actors people unintentionally doing things that that might be bad and not even knowing that they are and it goes back to some of the things we talked about with bias In fairness. I'm in a lot of cases people aren't intending to create something that's by us. But they end up creating something that's buys because they have because of the day that it's coming in. And so that's another area that we're looking at quite a bit which is how do we help people out there? Our customers

developers have the tools to be able to do this. Well such that such that they're not unintentionally creating bad Moms 2. Tooling right? I think it's also important to like get people trained all the way from the beginning and two are like Willie Mae format the data at diversity what you need for data annotator rice Hotel. We're having diversity along that, you know the whole path. So I think that's important and then I'm just out for other practitioners to understand about that stayed up as practicing eLearning. Absolutely. I mean, I

really think about this from the research respected cuz I'm a researcher and we write and publish papers and we have to think very carefully about when we write a paper. I mean the intent is to have a positive impact on the world. Right? So before you release the paper we say is this going to have a positive impact in the world or or not. Now you can't always predict who's going to be unintended consequences of releasing technology, but we try to do the best we can and whichever have a bias towards open this because it's just

generally made the prior is that it's it better to share these Technologies. I loved just publishing and and open sourcing and I have never been in a field like this before where the whole idea is you publish as soon as you have. Finding and then you submit to a conference like that. I had never seen that before and I think it's wonderful because it it means that the field progresses much faster, right? I want to come back to to one point that we were talking about, which is this notion of there might be unintended consequences or even with the best of intentions. You may not

know that your data is biased or is that there might be a skew or something and they're 110? once I died, I'd love to put out is this notion that building tools so that people can inspect the data and also opening up the community of people who inspect these things. So impaired for instance. One of the things we've been doing is building tools visualization tools that visualize your entire data said and allows allow you to very easily see things like oh, this is the shape of my data. This is

the distribution. This is what female datapoints look like for a male datapoints or children datapoints and what this means is that because it's because it's this kind of user interface where the experiences of visualization it means that not only developers can look at this. It means that product managers can look at those Executives who are different. Stakeholders can start having a conversation sit together and be like, oh, I I don't know. I saw something strange in the dataset or what do you think is going to happen when we Fest it by this

Dimension here. I thought I'd love to hear a little bit of your thoughts on on the importance of broadening the conversation around doing a high for everyone so that it's not a conversation that is happening only amongst developers. Obviously, the developers are super important, but I also think it's the conversation that needs to happen with a broader set of stakeholders. I guess one thought on that and it's interesting. You kind of the previous car rapidly. This is this is going one

customers is the first flight. I show a picture of the Mosaic browser from 1994 and my point with it is that we all of us and what we remember back to what that was. And before that point the internet had been around for a very long time before that point, but that was when the internet started to go out to many more people for many more things you start to see everything from checking sports scores on the way to the e-commerce all kinds of things and it opened up many more possibilities and open up a lot more questions and also questions

where they're needed to be more tools tools to make it easier for tools to make people safe. And I think we are at 1994. We are at that point for a I right now it's been used for years and years and years, but now it's been used in many many more ways. I think we need to think about you laying in that way. We need new problems are starting to come up every single day and new technologies are coming out every single day and we're going to need to think about that not just as a developer Community, but then in coordination with how it's being used in the people, that is

affecting. I wish that we were in a rush and I were just talking about the first spam detector before we came up in 1997 in that really, you know, I remember the first spam email that got sent in the in the 80s it in up and so it is as as new problems come up. You need to have your new tools to kind of help fight the problem. So you did and then you have to have tools to be able to work with it. Another thing another aspect of this of broadening the conversation, especially when we're talking about products.

Also is this notion of having in the same room your technical team your design team for instance Enterprise at Europeans and so forth. And so one of the things we're doing at Google is finding ways of of involving the designers say you waxers from the beginning and and educating them around machine learning as as a design material. How do you think about that as as literally a material you're going to design with and if you think about it, it's a very challenging material because it's something that each person in a product might have a

slightly different user experience. My experience may be very different from yours certain things might be automated in different ways for different people at different times. And so it makes that you wax that user entry. Action much more challenging that space becomes a lot bigger as so I can when we are talking about AI for everyone it really need this conversation that needs to happen between multiple multiple parties have to come back to before we run out of time. John is is

this notion of science one of the things that I think there's a lot of excitement around is that machine learning could be a new paradigm for science, right? That's finally we could solve certain very complex scientific problems. Using these new tools, so I'm curious about your thoughts as why is it that a I make such a difference in these very complex spaces. Well, that's an excellent question right now. The current state of AI and machine learning is still very perceptual. We're very very good at looking at images listening to sounds and so you can

imagine as I said before I like to talk about these thousands of undergraduate they would be listening or looking at the data. And so that's a great productivity term for science in the long run. But in the long run what we really want us and they called, where where the machine learning and AI Auburn actually try to understand what cause what we don't mean that still very very early in in research, but that's only the essence of science. Model explains Nobis cause that and you can extrapolate if you know what the causal

model is. So I think that's part of the long-term trend for a eye, but we're not there yet because we don't models don't really understand causation yet. Don't want to do things along those lines. We had a visiting scientist in our group in Cambridge. He's an earthquake size. You know who I'm talking about. He's a professor at Harvard who came and spent half a year with us. And one of the things I saw it was incredibly inspiring about what he was doing. So obviously as an earthquake scientist, do you want to predict earthquakes right? It's

incredibly hard we can't do it. Well yet but the thing that he was able to do so I did not know anything about earthquake science. He explained to me that this is the kind of science that lives in HPC high-performance Computing. So huge huge computers that will spend like entire weeks doing very very gnarly computation trying to stimulate where might a Aftershock happen after an earthquake. Okay things that Californians should care a lot about and and I'm sure

you do so imagine something that takes a week running on huge computers. Okay, he was able to get These kinds of stimulation results that were just as good using a simple neuron that okay. It was so simple the neuron that was so simple that it was the kind of neural network. Are you could count the neurons. That's how how simple it was and so he won he was done and Incredibly happy because it meant he didn't need to wait wait a whole week and using a ton of compute power

to do the same kind of simulations he was doing but then so that in itself is a win but then the deeper question and one of the things that gives me a lot of excitement is what exactly was this neural-net figuring out about the physics of the earth that we haven't figured out yet. so And again, if we could interpret what the system was doing, what does neuron that was doing if we could learn from it could we become better scientists ourselves? Right and you're starting to see this in science you starting to see this in

medicine. So for instance that the brain work with diabetic retinopathy where these systems are looking at the image of the back of your eye or funded and not only are they being able to understand whether or not you have the diabetic retinopathy, but They're also being able to do things like understand the gender of the patients and the cardiovascular risk of cardiovascular disease disease risk things that doctors themselves couldn't necessarily see in these images. So again part of the race now is to understand what are these systems learning that we can learn? So can we become

better doctors that are scientists right by using this kind of can we learn back? Can we listen back from from these machines? So we have a few minutes to go but I am I have two more questions one is for you Regin. What are you seeing of the ability of a guy to really help businesses have I guess I'm curious about like what have been some of the most successful cases but also some of the most surprising and you're like woah, I didn't know we could do a i for this or is there a bunch of things that they received their definitely certain

industries that that are farther ahead tech industry. Of course, there's a lot then use their financial services. There's a lot already being used but then there are many emerging use cases which I think a pretty amazing. I'm some of the examples we see like on the manufacturing line. Can you use Vision to figure out if a part is okay or not, like a tire coming down the factory line is going to be a good Tire or it might actually Bear risk due to the person those kinds of things are things that that we're starting to see retail is another case. How do you actually Make it so that you can

make the experience of the other user better Saul Give an example. Where were you using automl with Disney? My son happens to be a massive Lightning McQueen fan from cars and I'm sure many of you have have kids or family members that are that are big Disney fans. He will be he has an insatiable appetite for Disney stuff and the stuff with Lightning McQueen now with the technology that we've developed you can search on Disney's on shopdisney for everything with Lightning McQueen on it whether or not his cousin description or not. I'll do it like visual inspection those kinds of

things are really interesting interesting guy use kisses and so we're starting to see that We need to get a business to Point Oregon thinking about hey, I'm using AI to do this but where the users are feeling magic and it doesn't matter if it's AI behind the covers or not. And that's where I think things like what we've done with Google photos and stuff like that is really incredible because it's not about AI it's about the user experience. so we talked we talked about some of the challenges some

of the strategies I went to end on the opportunities. We talked about some of the opportunity also, but I wanted to end on that note on the opportunities in the last year. We seen a I being apply to to various domains in really interesting ways. So you have medicine science you have farming, right? We were talking about some of those examples and I'd love to hear from each one of you what opportunity are you excited about for a for the next however many years. John Deere want to start. Okay. Well, you mentioned

biomedical research to mow. I can't talk about that. So let me talk about another project. I find really exciting it which I mentioned before which is actually working to see if we can get a Fusion Energy to be a real commercial source of energy in order to displace fossil fuels because right now the world is burning way too many fossil fuels and flooding the atmosphere with carbon dioxide. So we're actually using machine learning to help this company t a Technologies and we're helping in two ways. We're trying to optimize other experiments are actually helping them designed

experiments optimization Amber using Bayesian methods to actually help the debug the plasma inside their machine. So the goal is going to be trying to make this plasma about as hot as the center of the Sun and if we can do that we can measure its heat loss rate and see if we can actually get to commercially relevant Fusion. Would really I think revolutionize the world when can we expect that? Well, I don't want to promise anything. We'll have some good scientific results. I think in 2019. We won't have what they call Breakeven Fusion then but will have some very solid results

predicting the heat loss rate. So we'll know a lot more in about a year. Yeah. For me I am very excited about two things. The first thing is really will continue to work with the council's contributors to build like really diverse covel diverse dataset, right that not a single kindly couldn't build myself, but actually leveraging everyone who's passionate about you don't like representing. There are the world to create dataset that we can open source to use by everyone and then similarly for machine learning since I do a

lot of volunteering outside of work with social entrepreneurship when you are across the world. I really want to see that and now he's available for like nonprofit air quality models or maybe like a present. They do you like using translate to make books available for like a thousand different languages and and for mother tongue, so I think that's exciting. I'm looking forward to that nice. For me there two things I'm passionate about one is Healthcare in the other such acacian. And so Healthcare Justice idea like you talked

about being able to use AI to do early disease detection. You have this incredible impact on one human being and on many human being sent that and I'm families around them and things like that in that that's that you could be really amazing education. I spend most of my time here is Google working on our education products and the idea of imagining an environment where every student can learn in the way that they want to learn at their own pace in their own style and having a I help power that can be transformative. You can pull the potential and actually make a harness the potential of every

single individual in a way that was never possible before. I'm I'm going to Segway from that and say again one of the things that has been really inspiring to see with things like tensorflow. Jas is that we have now Professor, is that a number of universities who were building educational materials around this and then I feel like that's really important. I'm really excited about this because it is not only the lowers the barrier for entry in the second ology. But it also I think starts to hit

a much more diverse set of of developers who are interested in in using this technology for four different things that we haven't even dreamt up yet. So I'm very excited about that. The other thing that I think his is really good to see is how the conversation around machine learning has been. Has started to touch on things like ethics and and fairness, and I think that's really important. I'm really happy that he rui Google we think very deeply about

these these questions and both from a research perspective what are research things we can do to help address some of these problems, but also, you know at policy product-wise design-wise. What're it it's it's a broader conversation. So I am going to thank our Panos. Thank you so much. It's been so inspiring to hear not only your insides, but to have you share your experience. As I said from many different parts of Google on this topic today with us. Thank you. Thank you.

Cackle comments for the website

Buy this talk

Access to the talk “Opportunities, challenges, and strategies to develop AI for everyone”
Available
In cart
Free
Free
Free
Free
Free
Free

Access to all the recordings of the event

Get access to all videos “2018 Google I/O”
Available
In cart
Free
Free
Free
Free
Free
Free
Ticket

Interested in topic “Software development”?

You might be interested in videos from this event

September 28, 2018
Moscow
16
159
app store, apps, development, google play, mobile, soft

Similar talks

Martin Görner
Software Engineer at Google
+ 1 speaker
Yu-Han Liu
Software Engineer at Google
+ 1 speaker
Available
In cart
Free
Free
Free
Free
Free
Free
Yufeng Guo
Developer and Machine Learning Advocate at Google
Available
In cart
Free
Free
Free
Free
Free
Free
Sundar Pichai
Chief Executive Officer at Google
Available
In cart
Free
Free
Free
Free
Free
Free

Buy this video

Video

Access to the talk “Opportunities, challenges, and strategies to develop AI for everyone”
Available
In cart
Free
Free
Free
Free
Free
Free

Conference Cast

With ConferenceCast.tv, you get access to our library of the world's best conference talks.

Conference Cast
558 conferences
22059 speakers
8245 hours of content