This article was updated April 10, 2018, and originally published on January 15, 2017.
Alphabet's (GOOG -0.21%) (GOOGL -0.30%) Google is best-known for its search engine, but is also home to some big moves in artificial intelligence (AI). In April 2016, Google CEO Sundar Pichai stepped in as the guest writer of Alphabet's annual founders' letter and noted:
"... the next big step will be for the very concept of the 'device' to fade away. Over time, the computer itself -- whatever its form factor -- will be an intelligent assistant helping you through your day. We will move from mobile first to an AI first world."
Let's take a look at what AI is and how Google is working to be at the forefront of the move to an AI-first world. Most references to AI actually refer to machine learning, and more specifically, deep learning, a specific technique within AI. So what is deep learning? It's when a computer simulation of the human brain called a neural network is used to recreate our capacity to learn. It's fed a multitude of examples and complex algorithms that "teach" it to recognize similarities and distinguish differences. In the simplest terms, the computer learns to recognize patterns from massive quantities of data.
Deep learning uses artificial neural networks to recreate our capacity to learn. Image source: Getty Images.
Machine learning ninjas at Google
Google systems are benefiting from deep learning. Image source: Getty Images.
Engineers proficient in deep learning are currently estimated to be about 10% of the company's 25,000 engineers and a leader of Google's machine learning effort has said he'd like that number to be closer to 100%. To help spread machine learning, the company created the Machine Learning Ninja Program, which identifies talented coders and embeds them in its machine-learning teams. When the six-month program is over, they fan out across the company seeking new ways to apply machine learning to existing systems. This strategy is paying off. For instance, it freed employees from cataloging street addresses captured by Google's Street View cars updating its Maps service: Google Brain now completes the tedious and time-consuming task using image recognition.
Google helped solidify its lead in deep learning in January 2014 when it acquired London-based AI company DeepMind for $600 million. It has since used machine learning technology to arrange its data centers more efficiently, based on how their power use had changed over time. Expanding on that success, Google now uses DeepMind technology to control the power consumption in those data centers. By manipulating servers and cooling systems and controlling over 120 condition-based variables, the company achieved a 15% improvement in power usage efficiency and a 40% reduction in cooling costs. These hundreds of millions of dollars in savings over several years will help recoup the cost to acquire DeepMind and there are plans to roll this strategy out to other potential applications. DeepMind's technology is also available to help companies that run on Google's cloud to improve their own energy efficiency.
Increased functionality for consumers
Consumer products are seeing dramatic improvements. Image source: Getty Images.
Google Assistant is where Google's AI most directly meets people. It powers the Google Home smart speaker and, in future, will be able to stream music, movies, and TV to the growing ecosystem of Google-branded devices including Android TV, Chromecast, and speakers. It will also provide deeper personalization resulting from data located in Maps, Calendar, Keep, and other services. In several recent unscientific tests, Google Assistant was found to be superior to its competitors. One key selling point is the ability to successfully answer follow-up questions.
Other examples abound. Users of Google Translate saw a dramatic improvement in its usefulness and accuracy as applying advances in natural-language processing reduced error rates by up to 85%, making greater progress in six weeks than the service had seen in the previous 10 years. Google Photos now has the uncanny ability to locate a specific photo based on a description given by the user. Gmail's Inbox introduced a Smart Reply feature that suggests three potential responses to an email that users can choose with a tap, rather than having to type an answer.
The Foolish final word
Google has been increasingly focusing its efforts on AI and planning for an AI-centric world. While it typically does not release the terms of acquisitions, it is telling that Google has acquired 11 AI-related companies in the past five years and applying AI technology to increasing its competitive advantage and positioning for the future.
While it will be difficult to quantify some benefits, we already know deep learning can increase revenue and efficiency, and make the ecosystem more productive and valuable to consumers. This moonshot will continue to pay dividends.
