Google makes machine learning, artificial intelligence top priority

Google has committed to investing in machine learning and artificial intelligence as the top priority for the company. 

Google defines artificial intelligence as making machines intelligent while machine learning is making machines that learn.

Greg Corrado, a senior research scientist at Google, said machine learning was the new way to write computer programs that learn from examples. It is involved in deep learning that is a powerful class of machine learning models. It is the modern reincarnation of artificial neural networks and is the collection of simple, trainable mathematical functions. It is compatible with many variants of machine learning. 

He explained that machine learning was the attempt to get computers to learn from experience and from data. Under machine learning, the three important things are data, models and computational power. 

He said machine learning “is not magic, but it is a tool. It is the new way to make software.” 

One example of Google’s services that use machine learning and deep learning is Gmail Spam, with which Google now intercepts 99.9 per cent of all spam. With Google Photo Search, users can search for personal photos without tags. Two other examples are Machine Translation and Smart Reply.

Google has used machine learning system to improve the quality of searches with RankBrain, an artificial-intelligence system that now works on 15 per cent of Google’s search results. RankBrain, however, is not a replacement for Google’s Hummingbird algorithm, which is a system that processes what people search for and which pages have to be ranked high. 

Google has just launched Smart Reply, which brings machine learning to mobile devices, both Android and iOS.

Gmail’s Inbox uses machine learning to recognise e-mails that need responses and to generate natural-language responses on the fly.

Smart Reply suggests up to three responses based on the e-mails users get. For those e-mails that only need a quick response, it can take care of the thinking and save precious time that would otherwise be spent typing. For those e-mails that require a bit more thought, it gives users a jump-start so they can respond right away. 

Smart Reply was rolled out last week on both Google Play and the App Store in English. 

In a bid to scale machine-learning adoption in wide-ranging software development, Google has launched TensorFlow. It has extensive built-in support for deep learning. Google has used TensorFlow for such things as speech recognition in Google apps, Smart Reply, and to search in Google Photos.

TensorFlow was originally developed by researchers and engineers working on the Google Brain Team within Google’s machine-intelligence research organisation for the purposes of conducting research into algorithm machine learning and deep neural networks, but the system is general enough to be applicable in a wide variety of other domains. 

 

YouTube creators supported

Corrado said TensorFlow was the open software standard for machine learning and deep learning. It is flexible enough for machine-learning research and robust enough for use in real products. It is the same software Google researchers and engineers use to train and operate RankBrain for Google Search as well as to train and operate Smart Reply for Google Inbox. 

“Google puts machine learning in real products,” he said.

He added that machine learning by example worked best. Computers are slow learners; deep learning is growing quickly but more research is needed. 

Another example of a Google product that comes with machine learning is Google Photo, which was launched in May. Currently, Corrado said, around 100 million people around the world use Google Photo each month. 

The Google Photo app is also available on both iOS and Android. Google Photo is set up to be the home for all users’ photos and videos and features automatically organising and helps users find photos quickly. 

Apart from machine leaning, Google emphasises support for its YouTube community, especially YouTube creators.

Eight YouTube Spaces are currently operating worldwide, in Los Angeles, New York, London, Tokyo, Sao Paulo, Paris, Berlin and Mumbai. 

YouTube Space is designed to be the home for YouTube creators to produce video content, attend workshops and collaborate with other channels.

The YouTube Spaces team is focused on helping creators make great content through strategic programmes and workshops largely administered at the YouTube Space production facilities.

David Macdonald, head of YouTube Space Asia Pacific, said YouTube Spaces was designed specifically for YouTube creators to experiment and innovate content for their channels. 

“The concept of YouTube Space is to bring together the most creative people in the world to learn, connect and create,” he said.

At YouTube Space, facilities featuring the latest production and post-production digital video equipment, as well as workshops, seminars and events are available for YouTubers. 

“YouTube Spaces offer creators a place to learn from industry experts and collaborate with other creators in the YouTube community. Across Asia-Pacific, we partner with others to provide similar opportunities for YouTube creators to improve their craft,” Macdonald said.

As of March, creators filming in YouTube Spaces had produced more than 10,000 videos, which had generated more than a billion views and more than 70 million hours of watch-time.

YouTube has more than a billion users, almost one-third of all people on the Internet, and every day people watch hundreds of millions of hours on YouTube and generate billions of views.

On mobile, the average YouTube viewing session is now more than 40 minutes, and the number of hours people spend watching videos on mobile is up 100 per cent year on year. More than half of YouTube views come from mobile devices. YouTube’s mobile revenue doubles annually. 

Partner revenue is up 50 per cent year on year. It has seen this level of partner revenue growth for three straight years. 

As of July, more than 8,000 partners were using Content ID, including many major network broadcasters, movie studios and record labels, which have claimed more than 400 million videos, helping them control their content on YouTube and make money on videos containing copyrighted material.