Amazon's empire rests on the low-key approach to AI
A MAZON's SIX-PAGE notes are famous. Managers need to write one every year and set up the business plan. Less well-known is that these missives always have to answer one question: How should you use machine learning? Answers like "not much" are, according to Amazon leaders, discouraged.
Machine learning is a form of artificial intelligence ( AI ) that mines data for patterns that can be used to make predictions. It took root on Amazon in 1999 when Jeff Wilke joined the company. Mr Wilke, who is currently Commander to Jeff Bezos, set up a team of researchers to study the Amazon's internal processes to improve efficiency. He woven his boffins into business units, turning a cycle of self-assessment and improvement into the standard pattern. Soon, the cycle was machine learning algorithms; The first recommended books that customers might like. As Mr. Bezos' ambitions grew, so did the importance of automated insights.
Nevertheless, while its other technological titans flaunt their professionalism at every opportunity, Facebook's face recognition software, Apple's Siri digital assistant, or Alphabet's self-propelled cars and master go player-Amazon have adopted a lower key to machine learning. Yes, Alexa competes with Siri, and the company offers predictive services in its cloud. But the algorithms that are most critical to the company's success are those who use it to streamline their own business continuously. The feedback loop looks like the one in the consumer perspective AI : build a service, attract customers, collect data and let computers learn from this data, all on a scale that human work could not imitate. [1[ads1]9659007] Mr Porter's algorithms
Consider Amazon's fulfillment centers. These large department stores, more than 100 in North America and 60-odd around the world, are the striking heart of the $ 207 billion online store. They store and ship items to Amazon seller. Inside one on the outskirts of Seattle, packs along conveyor belts pack at speed to a moped. The noise is deafening – and the facility is apparently deprived of people. Instead, inside a fenced area, lies a soccer field in thousands of yellow, cubic shelves, every six feet (1.8 meters) high. Amazon calls them pods. Hundreds of robots mix them in and out of neat rows, glide beneath them, and drag them around. Toothpaste, books and socks are stacked in a way that happens randomly to a human observer. However, through the lens of the algorithms that control the process, it understands the very highest.
Human workers, or "co-workers" in the company's vernacular man stations in holes in the fence surrounding this "robot field". Some pick items out of pods brought to them by a robot; others pack items into empty buckets, are shrouded away and stored. When they select or place an object, they scan the product and the appropriate shelf with a barcode reader so that the software can keep track of it.
The man responsible for developing these algorithms is Brad Porter, the Amazon's chief robot. His team is Mr Wilke's optimization match for fulfillment centers. Mr Porter notes "pod gaps" or how long human workers have to wait before a robot pulls a pod to the station. Fewer and shorter holes mean less downtime for the human worker, faster flow of goods through the department store, and ultimately faster Amazon delivery to your doorstep. Mr Porter's team is constantly experimenting with new optimizations, but rolls them out with caution. Traficylene in the robot field can be hell.
Amazon Web Services ( AWS ) is the second core infrastructure. It supports Amazon's $ 26bn cloud computing business, enabling businesses to host websites and applications without their own servers.
The main use of machine learning by AWS is to forecast calculation requirements. Insufficient computing power that Internet users flock to the customer's service can cause errors and lost sales that users encounter wrong pages. "We can't say we're out of stock," says Andy Jassy, AWS s chief. To make sure they never have to, Mr. Jassy's team customer data. Amazon can't see what hosts the servers, but it can monitor how much traffic each customer gets, how long the connections last and how solid they are. As in their fulfillment centers, these learning metadata machines predict when and where AWS will see demand.
One of AWS 's biggest customers is Amazon itself. And one of the most important things other Amazon companies want is predictions. The demand is so high that AWS has designed a new piece, called Inferentia, to handle these tasks. Mr Jassy says Inferentia will save Amazon money on all machine learning tasks it needs to run to keep the lights on and attract customers to cloud services. "We believe that there can be at least one order-of-magnitude improvement in cost and efficiency," he says. Algorithms that recognize voices and understand human language in Alexa will be a great recipient.
The company's latest algorithmic venture is Amazon Go, a cashless grocery store. A bank of hundreds of cameras monitors buyers from above, and converts visual data into a profile used to track hands and arms when handling a product. [3459005] D The system sees which goods customers pick up and calculate them to their Amazon account when they leave the store. Dilip Kumar, Amazon Go's boss, emphasizes that the system tracks the movements of the buyer's body. It doesn't use face recognition to identify them and link them to their Amazon account, he says. Instead, this is done by swiping a bar code at the door. The system attributes the subsequent actions of the 3 D profile of the swiped Amazon account. It's an ode to machine learning, crunching data from hundreds of cameras to determine what a shopper is taking. Try as possible, your correspondent could not fool the system and pilfer an object.
Suitable for Purpose
AI body tracking also pops up in the filling centers. The company has a pilot project, internally called the "Nike Intent Detection" system, which makes the associate associate what Amazon Go does for customers: it tracks what they choose and settles on shelves. The idea is to get rid of the handheld barcode reader. Such manual scanning takes time and is a nuisance to workers. Ideally, they can place objects on a shelf they like while the system looks and keeps track of. As always, the goal is efficiency, maximizing the speed at which the products flow. "It feels very natural for those employees," says Porter.
Amazon's cautious approach to data collection has isolated it from some recent research by Facebook and Google by governments. Amazon collects and processes customer data for the sole purpose of improving the customer experience. It does not work in the gray area between satisfying users and customers. The two are often different: people get social media or search for free because advertisers pay Facebook and Google for access to users. For Amazon, they are mostly the same (although it is toying with ad sales). Where regulators rise, is over Amazon's dominance in the core business of online store and cloud computing. This power is based on machine learning. It shows no signs of draining.
