How Amazon became AI-obsessed

  • Post by Rachel
  • Sep 01, 2020

Amazon’s long reach is everywhere - the e-commerce giant is also a film and TV producer, cloud computing provider, fashion designer, wind-farm backer, innovative grocer, maker of smart speakers, builder of fulfilment centres on the moon and maker of cutting edge robotics (and there’s more on Amazon’s DayOne blog). While Amazon’s indeed customer-obsessed, it’s also Artificial Intelligence obsessed.

A decade ago Amazon was still lagging behind other tech giants on the AI front. Today’s Amazon is far more than just the US’ biggest online retailer. Amazon is a leader in consumer-facing AI and enterprise cloud services as well as a force to be reckoned with in retail, logistics and media and entertainment. It’s now seeking to become the central provider for AI-as-a-service (alongside Microsoft, Google, Apple, IBM and to a lesser extent Salesforce, SAP, Oracle, Alibaba and Baidu) by doubling down on AI for AWS and the ecosystem around its AI assistant, Alexa. Whoever manages to dominate the business of providing artificial-intelligence services through cloud computing may have the operating system of the future and reign over what may be the most lucrative technology mega-trend yet.

This post explores how Amazon transformed itself from a deep learning wannabe to a formidable AI powerhouse in less than a decade.

“The prize will be to become the operating system of the next era of tech. The leaders in the AI cloud will become the most powerful companies in history.”

- MIT Technology Review, “_How the AI cloud could produce the richest companies ever”

A love-hate relationship

But first, a bit of context. As some argue, Amazon can be simultaneously described as both wonderful and terrible and its story has many contradictions. Amazon is famous for its relentless focus on customers; there’s no question that it has transformed how we live, work and shop in only 25 years. averages more than 200 million unique visitors per month and has over 100 million US based Prime subscribers. Nearly half (46.7%) of US internet users start product searches on Amazon (compared with 34.6% who go to Google first).

Consumers benefit from innovations like Amazon Prime’s same day free delivery, cheap prices, massive product choice and seamless e-commerce experience. Small business is also profiting with almost half of all US businesses generating between 81% to 100% of their revenue from Amazon sales. More than 20,000 small and medium-sized businesses worldwide on Amazon surpassed $1 million in sales in 2017. Amazon employs 750,000 employees worldwide and has generated more jobs overall than Facebook, Microsoft, Apple and Alphabet.

But there’s no doubt that Amazon has amassed unprecedented wealth and power, with some unintended consequences. Alleged anti-competitive practices have recently led to calls to break up the tech giants. Amazon has also been accused of crushing small businesses who operate on Amazon Marketplace as third party sellers, prompted backlash when it announced plans to open fulfilment centres in some communities and was entangled in an Alexa eavesdropping scandal raising consumer privacy concerns. The majority of its customers are worried about Amazon’s impact on the environment. Bernie Sanders criticised Amazon for its wages policy (later praising it for raising its minimum wage).

Love or hate Amazon, it’s hard not to be in awe of its unabating grit, drive and ingenuity. From humble beginnings as a bootstrapped Seattle-based online bookstore in 1995, it grew by roughly 24% a year for the past decade and currently expected to continue to grow at 16% annually by 2024, faster than any other company in history. In 2020 alone, Amazon has grown its market capitalisation by more than $600 billion and is currently valued at more than $1.5 trillion, with shares currently trading around $3,500. Amazon’s success has rattled companies across industries - companies fretting about Amazon’s impact on their business were an eclectic mix ranging from CVS and Macy’s to Netflix and Oracle.

How Amazon’s AI flywheel works

Amazon tends to be quite secretive about its strategy but here’s what we know: if you take a close look at Amazon’s key moves over the years, it’s clear that AI has been a continuous focus for the company since its inception and will be one of its main growth pillars in the future.

Amazon has cleverly weaved AI technologies into all aspects of its business, with deep learning powering Alexa, Amazon Web Services, autonomous robots at its warehouses to supercharge the order fulfillment process and make same-day delivery a reality, and in nearly every other division of the company.

Wired’s 2018 article, “Inside Amazon’s Artificial Intelligence Flywheel” provides a rare glimpse into how various parts of Amazon’s enourmous business work as a single perpetual motion machine:

“Amazon now has a powerful AI flywheel, where machine-learning innovations in one part of the company fuel the efforts of other teams, who in turn can build products or offer services to affect other groups, or even the company at large”.

Amazon’s AI flywheel was built on 5 key principles:

1/ Generate bold ideas - Since the early days, Amazon’s recommendations engine, shipping schedules and warehouse fulfillment robots have been infused with AI. For the past eight years, it’s rumoured that Bezos’ leadership team has been encouraged to trek to Bezos with a (now famous) six-pager memo in hand, to pitch ideas that would transform various Amazon departments or products, ideally by using machine learning. Additionally, every business planning document at Amazon has had to include an explanation of how the unit would make use of machine learning, and Amazon leaders were told that “None is not a good answer”.

Some of the resulting ideas involved entirely new businesses/products (like the Echo) while others reimagined existing businesses like AWS. A decade ago, the company faced challenges at every front: the talent, tools and algorithms required for many of these projects were either scarce or yet to be invented. But regardless of the constraints, the more audacious the idea, the more likely it was to gain Bezos’ commitment.

2/ Invest heavily in R&D - Since 2012, Amazon has invested a total of over $100bn in R&D, more than Alphabet, Apple, Intel and Microsoft. Lab126 is the secretive R&D lab based in Sunnyvale, California that designs and engineers consumer electronic devices for Amazon. Known for Amazon’s most popular devices: the Kindle e-reader and Echo smart speaker (as well as Fire TV and Fire tablets), Lab126 teams use high-performance compute (HPC) capacity and machine learning capabilities to scale design environments to accelerate product development, gain efficiencies, and speed time-to-market. Lab126 is also known as Amazon’s “reliability lab” where engineers perform “research and destruction” simulations ensuring devices’ resiliency in extreme weather conditions and other kinds of impact.

3/ Connect teams & disseminate knowledge - The results of Amazon’s highly experimental and entrepreneurial approach to AI have had an impact far beyond individual projects. Amazon transitioned from AI ‘talent islands’ that operated in isolated teams to cross-company collaboration, spreading the machine learning nous far and wide and creating an ambitious AI culture across the company’s various business units, spinning off new products and services that draw on past innovations (see below).

4/ Snap up the best talent - The increasing complexity of its AI projects (such as the far-field speech recognition problem solved for Alexa) and Amazon’s push to improve its reputation as a research-friendly organisation, drew top AI talent that Amazon had previously struggled to attract. Amazon also went on an acquisition spree and bolstered its AI capabilities through buying a host of AI startups in a bid to secure the best talent and technology on the market.

5/ Spin off new products & services - Amazon’s AI transformation is now visible in every part of its services and products. Its AI culture came into its own when Alexa’s voice technology was extended beyond the Echo, bringing together data and technology to power other products where a voice service would be useful (such as accessing Amazon Music, Prime Video, personal recommendations from the main Amazon shopping website, and other services directly from your Echo device). Then the technology began spreading through other Amazon domains like Fire TV, voice shopping, the Dash wand for Amazon fresh, and, ultimately, AWS.

  • In 2016, AWS released new machine-learning services that drew on innovations from Alexa: Polly, a cloud-based service that converts text into lifelike speech, enabling AWS customers to create applications that talk and build entirely new categories of speech-enabled products and a natural language processing engine called Lex for quickly and easily building natural language conversational bots. Earlier this year, Amazon launched Brand Voice, a fully managed service within Amazon Polly, that pairs customers with Amazon engineers to build AI-generated voices representing certain brand personas.nThese offerings allowed AWS customers, from giants like Pinterest and Netflix to tiny startups, to build their own mini Alexas without having to worry about voice recognition infrastructure.
  • A third service involving vision, Rekognition makes it easy to add image and video analysis to applications using proven, highly scalable, deep learning technology that requires no machine learning expertise to use. From content moderation to face verification and celebrity recognition, Rekognition is used by media giants like the NFL and CBS as well as financial services companies and dating apps.
  • In 2017 Amazon launched SageMaker, a fully managed machine learning service, enabling data scientists and developers to easily build and train machine learning models at any scale, and then directly deploy them into a production-ready hosted environment.

These machine-learning services are key to Amazon’s AI flywheel. As companies build their vital machine-learning tools inside AWS, they’re highly unlikely to move to competing cloud providers. Also, AI startups lean towards building their machine learning products on AWS given that’s where most of their target customers are already managing their data.

And so the classic data flywheel clicked into place as millions of customers started using the Echo and other Alexa-powered devices (like Sonos speakers) which helped Amazon amass data that not only made those third party systems perform better (and hence acquire more customers) but supercharged Amazon’s own array of machine-learning tools and platforms and made the company a hotter destination for machine-learning scientists.

Amazon’s AI mindset

What distinguishes Amazon’s AI success is its wide-ranging, ferociously ambitious AI strategy. Interestingly, Amazon’s remarkable rise in the machine learning arena started out as an early commitment to deeply embed technology in every aspect of the company’s operation, at a time when few companies outside tech had the vision to do so. In his 2010 letter to shareholders, Bezos notes:

“All the effort we put into technology might not matter that much if we kept technology off to the side in some sort of R&D department, but we don’t take that approach. Technology infuses all of our teams, all of our processes, our decision-making, and our approach to innovation in each of our businesses. It is deeply integrated into everything we do.”

Fast forward to 2017, and Bezos’ letter to shareholders that year confirmed AI as a strategic priority (in the context of maintaining relevance and edge over the competition). It highlights how machine learning has guided the company for many years and permeates many Amazon customer facing devices and services as well as back-end capabilities. Bezos specifically calls out AWS as a platform set to lower the costs and barriers to machine learning and AI so organisations of all sizes can leverage these advanced technologies and develop powerful systems, even when they have no machine learning expertise. Bezos also mentioned the machine learning APIs launched by Amazon (Amazon Lex, Amazon Polly and Amazon Rekognition) which can remove the heavy lifting from natural language understanding, speech generation, and image analysis making these technologies accessible to anyone.

“Over the past decades computers have broadly automated tasks that programmers could describe with clear rules and algorithms. Modern machine learning techniques now allow us to do the same for tasks where describing the precise rules is much harder. At Amazon, we’ve been engaged in the practical application of machine learning for many years now. Some of this work is highly visible: our autonomous Prime Air delivery drones; the Amazon Go convenience store that uses machine vision to eliminate checkout lines; and Alexa,1 our cloud-based AI assistant…But much of what we do with machine learning happens beneath the surface.

Machine learning drives our algorithms for demand forecasting, product search ranking, product and deals recommendations, merchandising placements, fraud detection, translations, and much more. Though less visible, much of the impact of machine learning will be of this type – quietly but meaningfully improving core operations.”

-Jeff Bezos’ letter to shareholders 2016

After leveraging the power of AI to dominate e-commerce, bringing an AI-powered personal assistant to 100 million homes, enabling more than 28,000 smart home devices that work with Alexa made by more than 4,500 different manufacturers and becoming the dominant cloud provider, in early 2019 Amazon announced it was launching the **re:Mars **event. This was an offshoot of its invite-only robotics and AI conference (previously only open for billionaires and tech elite). The new conference gathered experts in machine learning, robotics, automation and space in Las Vegas in June 2019, with speakers from Amazon, MIT, UC Berkeley, NASA and Harvard (and even a guest appearance from Iron Man himself, actor Robert Downey Jr.).

Re:Mars has some down-to-earth goals, like teaching its attendees - business leaders, technical builders and astronauts (who can attend for free) - how to apply AI and machine learning to their own businesses. But it’s also designed to generate excitement and optimism about AI, with sessions on topics like colonizing space and remarks from Bezos and AI leaders like Andrew Ng, acknowledging the challenges and pitfalls of AI but emphasising the enormous opportunities ahead.

The AI race is heating up

Unlike with AWS, where Amazon caught its big tech competitors slightly off guard and had a first mover advantage, it’s certainly up against tough rivals in the AI race and competitors aren’t sitting idly by. Amazon has 3 key challenges ahead:

1/ On the voice front, while Echo is currently the leading platform, Google’s rival offering, Home and Apple’s HomePod are both popular competitors. Google benefits from its extensive AI research, while Apple benefits from seamless integration across its macOS and iOS devices. Deepening Amazon’s moat in voice is important, especially as switching costs are low and Alexa could lose lustre and market share if Google or Apple manage to deliver a quality edge to consumers.

2/ Snapping up promising talent and technology is another challenge. When it comes to tech acquisitions, the usual suspects - Facebook, Amazon, Microsoft, Google and Apple - have all been aggressively acquiring AI startups in the last decade. Apple leads the way, with 20 total AI acquisitions since 2010, followed by Google and Microsoft. Amazon, however, has been the most aggressive acquirer of AI startups, spending more than twice as much as Google and four times as much as Microsoft in the past few years. Acquisitions played a pivotal role in helping these companies scale their AI capabilities - many prominent products and services came out of AI acquisitions (e.g. Apple’s Siri, or Google’s contributions to healthcare through DeepMind). But tech giants aren’t the only ones snapping up promising AI startups - today, traditional insurance, retail, and healthcare incumbents have joined the fray.

3/ On the AI-as-a-service front, Google and Microsoft, Alibaba and Baidu (and to a lesser extent Apple, IBM, Oracle, Salesforce, and SAP) are also vying for dominance. They all have huge amounts of valuable data, have been making AI research breakthroughs for decades and possess the massive computing resources and armies of talent required to build this AI utility, as well as the business imperative to get a foothold in what may be the most lucrative technology mega-trend ever. While Amazon will undoubtedly continue to focus on voice, virtual assistants and natural language processing, AI-as-a-service will certainly be a priority for Amazon, who is, more than ever, aspiring to become a platform company and lead the way in making AI more accessible to machine learning novices.

So far, none of this activity has resulted in much in the way of revenue; none of AI’s biggest players bother to break out sales of their commercial AI services in their earnings calls. But that would quickly change for the company that creates the underlying technologies and developer tools to support the widespread commercialisation of machine learning. That’s what Microsoft did for the PC, by creating a Windows platform that millions of developers used to build PC programs. Apple did the same with iOS, which spawned the mobile-app era.”

- Peter Burrows, MIT Technology Review, “How the AI cloud could produce the richest companies ever”

Final thought

Since founding Amazon in 1994, Bezos has followed a set of unconventional rules: don’t worry about competitors; don’t worry about making money for shareholders; and don’t worry about the short-term. He’s always argued that if you focus on the customers, everything else will fall into place. Amazon has proven time and again that this approach pays off in grand ways. And it’s not one to shy away from ferocious competition. It will be interesting to see who the AI winners are in the coming years. But one thing’s for sure - Amazon will make it a fun race to watch.

*Cover Photo: Jeff Bezos via Twitter