字幕表 動画を再生する 英語字幕をプリント Let's get back to the Google I/O developers conference in Mountain View, California. With the spotlight on hardware this year, Google CEO Sundar Pichai announced a new artificial intelligence supercomputer chip looking to transform the search giant into an AI first company and a real cloud computing contender. We caught up with Scott Huffman, Google's vice president of engineering, and asked just what this supercomputer chip means for Google. Well we're really excited to be able to have the computing power to be able to really harness all of the newer machine learning algorithms. One of the things that is exciting about these new algorithms is they're very — what in computer science terms we call — "highly parallelizable". So you can do many computations at once and get very high scale and process a lot of data that way. And the new chips are really designed to do that from the ground up. So really designed to do the kinds of machine learning processing that we're using a lot of. Digital assistants are all the rage now, but Google Home sales are still dwarfed by Amazon Echo. How is the Google Assistant different from Siri, different from Alexa, or Cortana? So, one thing that we're very excited about with the Google Assistant, is the ability to actually go across all the different devices and contacts in your life. So as you go from your house, to your car and your commute, to out and about on your day, we want the same assistant to really be available to help you in all those different places. And so today, we're really excited to deploy the assistant out to all the iPhones, make it available to iPhone users in the US, and we're in the process, of course, of rolling out across all the Android phones, Google Home, Android Auto, Android TV, Android Wear, so really making that assistant always available to you no matter what you're doing. Now what is it gonna take for voice technology to actually improve? Because, you know, I've used all of these devices, and in my own experience, it's still rather crude. Well, so we think we're making a lot of progress, but one of the big things and Sundar talked a little about it today, is really using broad-scale data and neural algorithms in order to improve the technology. So we've been actually pretty significantly overhauling, kind of, all of our algorithms under the hood every couple of years to take advantage of the new computing power that we have and new and larger amounts of data. And every time that we do that, we see a pretty big jump in improvement. One of the things that we did as we worked on bringing Google Home to the market that was really an exciting thing, is because Google Home needs to work at a distance, I might be standing far away, there's a lot more noise in the microphone signal. And so by adding essentially artificial noise into our training data, we were able to have our neural network actually be able to recognize things at a far distance away. So, these kinds of algorithms are very powerful for kind of, shaping recognition in different environments. What is one place that you see the assistant going, where it hasn't gone yet? Today we're enabling a voice conversation into a great set of functionality but that... the users are doing sort of the obvious things on every device. But I don't think we fully realized yet the vision of having any kind of conversation you want, having it really be understood, and then having the assistant tap in to all the different services in the world and a seamless way to do that. That's really the vision. And so I think we have a long way to go. One example that we showed some beginnings of today that we're really excited about is something we call Google Lens. And this is just the realization that you know, speaking out loud is great, but when I'm talking with my friend, a lot of times what I do is point at something, and then we talk about that. We talk about what we see. And with Google's advances in computer vision and computer, kind of, image understanding, the assistant is actually going to begin to have that capability over the next few months, so that I'll be able to open my camera, my viewfinder, and then begin to talk to the assistant about what I see. And so we're really excited about that.