Alexa Amazon: This is the artificial intelligence that puts voice to Alexa and finds your purchases at Amazon | Technology

A large speaker cubicle about three meters high and one wide attracts the attention of those who enter the London headquarters of Amazon for the first time. Inside, several blue stickers on the wall invite you to interact with Alexa, the virtual assistant of the company. They suggest that you ask for relaxed music, a song from the 80s or that you wonder about time.

- "Alexa, tell me a joke".

- "How does an electrician know that he is in love? Because there are sparks. "

In addition to telling jokes, Alexa also checks calendar events, searches for recipes, controls television, regulates the light or temperature of a room or puts the music more in line to read or be at a party. Specifically, the virtual assistant, who will arrive in Spain before the end of the year, has 50,000 skills.

Alexa is one of Amazon's big bets, which tries to compete with other virtual assistants like Siri or Cortana. While in 2017 it had 4,000 compatible devices, it now has more than 20,000. To develop this device and carry out other projects based on artificial intelligence, the giant of electronic commerce has 25 development centers throughout Europe in which some 5,500 researchers work. The profiles include from scientists specialized in machine learning even experts in robotics and software, hardware and aerospace engineers.

Since 2010, Amazon has invested 27,000 million euros in all its European operations. And next year he plans to open a new development center in Manchester, as announced last week by Doug Gurr, Amazon's UK manager, at an organized event. at the headquarters of the technological giant in the London East End to which EL PAÍS was invited. There, the majority of the workers are under 40 years of age and move through the 15 steel and glass plants without any dress code. There are bars on every corner, with food to snack at any time, giant footballs and even live music.

enlarge photo

In addition to the United Kingdom, the company also has development centers that work together in Austria, France, Germany, Ireland, Italy, Luxembourg, the Netherlands, Poland, Romania and Spain - one in Madrid and the other in Barcelona. "Alexa found her voice thanks to the teams in Poland and learned to answer questions with the help of teams in the UK," said Jeff Wilke, Amazon's global consumer CEO.

"Achievements in automatic voice recognition have dramatically increased in the last five years," says Rafal Kuklinski, director of a development center in Poland where voice synthesis is studied. Working in this field is not only useful for Alexa to understand what a user asks for, but also to understand the product searches on the electronic giant's website.

"In addition, for a computer it is difficult to understand human language", explains Hugo Zaragoza, head of machine learning in the Amazon center of Barcelona. One of the most complicated challenges is that the algorithm understands what the user's search intent is. For example, when writing "mac" in the search engine, very different products appear. Among them, a computer, makeup or even a jacket. "There is a semantic problem. If someone searches for something and can not find it, they will look for it in a different way. We have to learn from user behavior, "says Zaragoza.

The technological giant has also been reinvented to offer companies the possibility to customize the searches of their employees. Amazon business allows companies to organize guides to determine which products they prefer their employees buy, according to Jaime Vallori, head of software development in Madrid. In this way, when for example a hospital worker searches for "alcohol", Amazon will not show him a drink but medicinal alcohol.

There are still many challenges in this regard, according to the experts consulted. How much importance do we give to clicks and how much to buy? When typing "iPhone" in the Amazon search engine, the most clicked product is the iPhone 7, but the best seller is the charger. Which should appear first in the search engine?

The key, the experts point out, is to teach the algorithm that has to show in each search and how to analyze the product from different points of view. For example, to establish filters. For the search "women's shoes", an artificial intelligence system would have to analyze if they have a zipper, laces or Velcro.

The goal of the development centers is to respond to all these challenges, in addition to being able to "offer more options, lower prices and a better shopping experience to hundreds of millions of customers around the world." But in the company they are aware that the technological sector is advancing at a dizzying pace and new challenges will soon emerge. "Now we buy with the voice, but who knows if in 10 years, we will not do it with gestures," says Ralf Herbrich, head of machine learning in the center of Berlin.