Artificial Intelligence in Practice Read online

  Self-Driving Cars

  Alphabet's autonomous vehicle division, Waymo, has one of the most mature self-driving car platforms in the world, having recently become the first to make rides available commercially.5

  Alphabet has gone down the road of developing its own vehicles, which are so automated they don't even include steering wheels or any driver controls. Designed for a new age of urban motoring where car ownership is often expensive and inconvenient, Waymo's service is aimed at the ride-sharing networks, which it predicts will make up transport networks in smart cities of the near future.

  Captioning Millions Of Videos

  Google also uses machine learning natural language algorithms to automatically create subtitles for the hard of hearing (or those who value peace and quiet) for videos on its YouTube video-streaming service.

  As well as speech, the system uses deep neural networks to identify ambient sounds, including applause, music and laughter and automatically displays text telling the viewer what sounds are occurring.6

  Diagnosing Disease

  Alphabet's AI (specifically deep learning) technology has also been extensively deployed in the medical field. One recent breakthrough involves diagnosing eye conditions. For this, it applies learning algorithms to 3D infrared scans of eyeballs known as optical coherence tomography scans.7

  The system relies on two deep learning algorithms, one of which builds up a detailed map of the eye's structure and learns about what is normal and what could be indicative of a problem such as age-related macular degeneration. The other makes diagnoses based on medical data and provides medical professionals with assistance in diagnosing and treating the illness.

  Google Brain

  Google's AI research division is known as Google Brain. It was formed by Google's Jeff Dean and Greg Corrado along with Andrew Ng of Stanford University in 2011, and their work has established them as pioneers of the current wave of practical AI technology.

  Google Brain realized that the vast, super-fast storage networks it had built up, as well as the huge amount of data flowing through the internet (and therefore its servers), were the keys to unlocking the usefulness of machine learning and deep learning.

  Since it was established, this group has been responsible for developing many of the core technologies, such as computer vision and natural language processing, driving the current wave of AI adoption in business.8

  Deep Mind

  Another key weapon in Alphabet's AI arsenal is Deep Mind, which it acquired in 2014. The UK start-up specialized in building neural net “simulations” of the brain, which it trained to play games. The focus on gameplay enabled Deep Mind's researchers to study the way the brain tackles various cognitive problems, and use the data to build machines that attempted to tackle problems in the same way. The technology made headlines in 2016 when it powered the first computer that was able to beat a professional human Go Player.9

  Today AI technology developed by Deep Mind powers a number of Alphabet's smart applications, including optimizing the efficiency of cooling machinery in its data centers, and managing battery life on mobile devices running the Android operating system. It's also the brains behind the eye imagery in the healthcare application mentioned above.

  Key Challenges, Learning Points And Takeaways

  Alphabet and Google clearly believe that AI is the launchpad that will drive the next wave of transformative computer technology.

  As well as this, they believe the societal impact of this next wave will be even greater than that of previous waves – including the development of the internet.

  Having more data than anyone else is a key advantage, which has enabled Alphabet to continue to develop first-in-class services – from search, to ad serving, language translation, speech processing, smart homes and autonomous driving.

  Having the infrastructure in place to move that data around, and the processing power to query and access it at super-fast speeds required to power its search engine, enabled Google to apply the same infrastructure to AI applications.

  Where Alphabet has seen breakthrough development in leading edge AI, such as deep learning, by research groups and start-ups, Google has used its financial resources to bring it on board and add their expertise to its own.

  Notes

  1Alphabet, 2017 Founder's Letter: https://abc.xyz/investor/founders-letters/2017/index.html

  2Search Engine Land, FAQ: All about the Google RankBrain algorithm: https://searchengineland.com/faq-all-about-the-new-google-rankbrain-algorithm-234440

  3Google, Google Duplex: An AI System for Accomplishing Real-World Tasks Over the Phone: https://ai.googleblog.com/2018/05/duplex-ai-system-for-natural-conversation.html

  4The Verge, The Pixel Buds’ translation feature is coming to all headphones with Google Assistant: https://www.theverge.com/circuitbreaker/ 2018/10/15/17978298/pixel-buds-google-translate-google-assistant- headphones

  5Financial Times, Alphabet's Waymo begins charging passengers for self-driving cars: https://www.ft.com/content/7980e98e-d8b6-11e8-a854-33d6f82e62f8

  6Google, Adding Sound Effect Information to YouTube Captions: https://ai.googleblog.com/2017/03/adding-sound-effect-information-to. html

  7Nature, Clinically applicable deep learning for diagnosis and referral in retinal disease: https://www.nature.com/articles/s41591-018-0107-6

  8Google, Using large-scale brain simulations for machine learning and A.I.: https://googleblog.blogspot.com/2012/06/using-large-scale-brain-simulations-for.html

  9Wired, Google's AI Wins First Historic Match: https://www.wired.com/ 2016/03/googles-ai-wins-first-game-historic-match-go-champion/

  3

  Amazon: Using Deep Learning To Drive Business Performance

  Jeff Bezos founded Amazon as an online book store, but in reality he could have sold anything. His main focus was establishing a technology company that could dominate during the predicted boom in online retail, which he saw coming. Today, Amazon is a multinational e-commerce giant and the world's leading cloud computing provider, making it the third most valuable public company in the United States. Beyond its core retail and cloud business, the company also has a publishing business, a film and television studio operation, and produces consumer products such as the Kindle e-readers, Fire tablets and TV sticks, as well as the Amazon Echo.

  Amazon has used predictive analytics since those earliest days in the 1990s. It has experience deploying these systems across its entire business – from its famous recommendation engines to optimizing the routes of robots working in its order fulfilment centers.

  However, the growing power of machine learning has caused the online retail giant to reassess every aspect of its operations since the start of the current decade. Not content with merely competing with Walmart and Target for the retail market, it has always positioned itself as a rival to Google, Facebook and Apple, seeking leadership in the tech sphere.

  This meant implementing deep learning technology into its core services, as well as expanding into new areas such as home automation with its Alexa-powered Echo devices and cashier-free retail stores.

  Looking ahead, Amazon has grand plans involving automated delivery drones and “anticipatory shipping”, which will attempt to read your mind and ship products to you before you even order them!

  How Does Amazon Use Artificial Intelligence?

  Amazon pioneered the recommendation engine – search engines designed to sell us things – which has been the core of its business strategy since the beginning. Over the years, the analytics behind the scenes have become more sophisticated but it has always worked by segmenting customers according to the data it collects about them, modelling their behavior and matching them with items popular with others who fit a similar pattern.

  In early 2014, the company began the single biggest overhaul of its recommendation system to date, when it started to implement deep learning algorithms into its prediction tools.1 Deep learning is now built into many of the site's features, which are de
signed to present the user with a more personalized shopping experience, such as its “frequently bought together” and “customers who bought this also bought …” recommendations.

  Deep learning uses deeply layered neural networks that mimic human brains in the way they “learn” from the data that passes through them. These algorithms are capable of adapting themselves to become increasingly efficient at spotting patterns and relationships in data, in this case Amazon's transactional and customer behavioral data. They now power Amazon's recommendation engine just as they do Google's searches, Facebook's feeds and Netflix's movie suggestions. Like its rivals for the tech crown, Amazon is confidently backing deep learning as the technology that will power the artificial intelligence (AI) revolution.

  Another key use case at Amazon is found in its fulfilment centers – warehouses where the millions of customer orders placed every day are picked and packed by humans working alongside sophisticated, AI-powered robots. When observed as a static, stand-alone piece of machinery, Amazon's warehouse robots may not look like much – simply squat, mobile platforms.2 But driven by deep learning algorithms, they are able to efficiently route their way around labyrinthine stacks of portable shelves, locate whatever items are required and move them to the human picker who completes the assembly of each order. As robots can operate in far tighter conditions than humans, this initiative helps Amazon maximize the space available for stock in its warehouses, increasing revenue as orders can be filled more quickly. One hundred thousand of these robots are currently deployed in Amazon fulfilment centers around the world.3

  Amazon Alexa

  It's strange to think that the AI-powered personal home assistant device almost seemed like a novelty when Amazon first introduced it in 2015. As of 2018 they are a feature in 16% of US homes, and that figure looks set to increase as Amazon, along with Google, continue to improve, refine and market their devices.4

  Amazon's breakthrough was to realize that the biggest factor limiting uptake of AI in the home wasn't the technology itself, which had matured to the point where it is more than capable of assisting with basic domestic tasks. It was the interface itself – while smartphones have become increasingly useful, they often still aren't as simple to use as, say, a light switch, kettle, radio or recipe book.

  Echo made it straightforward for us to communicate using our voices with smart home devices, as well as a handy portal for quick information, or playing background music while we go about household chores.

  The accuracy with which it can interpret our voice commands is due to Amazon's implementation of deep learning within its natural language algorithms.5 Neural networks are used to detect the user's “wake word”, which tells the device to start listening for and interpreting a command. As it processes voice commands it becomes increasingly efficient at understanding the nuanced ways human beings use spoken language. Effectively, the deep neural network “learns” about how we talk from the voice data it processes.

  Amazon's Artificial Intelligence Flywheel

  Amazon's model for propagating the use of AI across its variety of business operations has been referred to as a “flywheel”.6 The name is taken from a class of mechanical devices designed to efficiently store energy generated by a power source and moderate its rate of release. The idea is that excess “energy” generated by successful deployments of AI in one part of the business will fuel research and investment in another part.

  This approach helped to foster an environment where data and technology are shared between departments and business units, which are able to learn from best practice guidelines established by others. For example, the improvements in recommendation engine accuracy brought about by the deployment of deep learning became a key driver behind its adoption by the team working on Echo's voice capabilities.

  In turn, other units at Amazon realized they could capitalize on the widespread adoption of Alexa-equipped devices into homes – particularly the ability to create custom applications known as “skills”, which can be invoked through the device. This led to skills being added to let users access services such as Amazon Prime Video and Amazon Music Unlimited with their voices. And, in turn, deep learning has also been integrated into the way Alexa decides which of its 40,000 skills a user will find most useful, based on the words they speak.7

  Amazon realized that successful deep learning initiatives pay for themselves, not just by enabling the business processes that they are deployed in to operate more efficiently, but also by generating more data with which to train algorithms deployed in different processes.

  Amazon Web Services

  Just like its competitors Google and Alibaba, Amazon sells cloud-based computing services to its business customers under its Amazon Web Services (AWS) brand. In recent years it has built machine learning services into this offering, meaning businesses can “hire” AI capabilities at a fraction of the cost of building their own infrastructure.

  With the race to adopt AI among businesses in all sectors, providing tools to help smaller businesses compete has become a core business strategy for Amazon. After all, as the old adage goes, those most likely to get rich out of a gold rush are those who are selling the shovels!

  AWS provides access to core machine learning technologies, such as natural language processing and computer vision, as well as tools that can put them to use extracting actionable insights from unstructured voice or video data.8

  Amazon Prime Air

  One of Amazon's more ambitious projects involves rolling out fleets of airborne drones to deliver packages directly to our homes. When it was announced in 2013, its mission was to enable Amazon to make deliveries within 30 minutes of a customer placing an order.9

  Since then Amazon has carried out the first trial deliveries by drones from its fulfilment center in Cambridge, England.

  Machine learning is a fundamental part of the systems that control the drones.10 Although it has been going for several years, the project is still a long way from general use and there are regulatory hurdles to be overcome. Amazon has not publicly spoken in detail about the technology of the drones, but it is likely that they employ computer vision to help them navigate around obstacles and identify safe landing spots.

  Key Challenges, Learning Points And Takeaways

  Amazon was one of the first online businesses to harness the power of predictive analytics. AI – which promises more accurate predictions than any technology so far – is a natural next step for it.

  Amazon has built a corporate strategy that it calls a “flywheel” to encourage distribution of energy, momentum and data generated by AI initiatives throughout its network of business operations.

  Advances won through building deep learning capabilities into its recommendation engine algorithms inspired further use of the technology, feeding into development of its Alexa voice assistant and its Amazon Prime Air drone delivery service.

  Amazon is enabling other businesses to automate and take advantage of AI. It does this by leasing its machine learning and deep learning technology as a service through its AWS platform.

  Notes

  1Wired, Inside Amazon's Artificial Intelligence Flywheel: https://www. wired.com/story/amazon-artificial-intelligence-flywheel/

  2Robots, Drive Unit: https://robots.ieee.org/robots/kiva/?utm_source= spectrum

  3IEEE Spectrum, Brad Porter, VP of Robotics at Amazon, on Warehouse Automation, Machine Learning, and His First Robot: https://spectrum. ieee.org/automaton/robotics/industrial-robots/interview-brad-porter- vp-of-robotics-at-amazon

  4Tech Crunch, 39 million Americans now own a smart speaker, report claims: https://techcrunch.com/2018/01/12/39-million-americans-now -own-a-smart-speaker-report-claims/

  5Quora, How does Amazon use Deep Learning?: https://www.quora .com/How-does-Amazon-use-Deep-Learning

  6Wired, Inside Amazon's Artificial Intelligence Flywheel: https://www. wired.com/story/amazon-artificial-intelligence-flywheel/

  7Amazon, The Scalable Neural Architec
ture behind Alexa's Ability to Select Skills: https://developer.amazon.com/blogs/alexa/post/4e6db03f-6048-4b62-ba4b-6544da9ac440/the-scalable-neural-architecture-behind-alexa-s-ability-to-arbitrate-skills

  8Amazon, Machine Learning on AWS: https://aws.amazon.com/ machine-learning/

  9CBS, Amazon unveils futuristic plan: delivery by drone: https://www. cbsnews.com/news/amazon-unveils-futuristic-plan-delivery-by-drone/

  10Amazon, Machine Learning on AWS: https://aws.amazon.com/machine -learning/

  4

  Apple: Integrating AI Into Products And Protecting User Privacy

  Apple is the world's largest information technology company by revenue. The California-based company designs, develops and sells iconic smart tech products such as the iPhone, iPad, Macs, Apple Watch, Apple TV, as well as accompanying software and services. In 2018, Apple became the first-ever public company to be valued at over US$1 trillion.1

  Apple's artificial intelligence (AI) strategy centers around its devices, and in recent years the company has positioned itself as a pioneer of in-device AI technology, citing its superior security and potential for creating unique, user-engaging experiences.

  How Does Apple Use Artificial Intelligence?