Alexa, Siri, Cortana are some of the names that I know maybe more about us than our closest friends.How did the Silicon Valley giants come to know the most hidden secrets?
Ruthy Hope Slatis couldn't believe what she could hear.She was temporarily employed by a Boston recruitment agency for a post with a vague description: Transcript of AMASON audio files.com.For $ 12 per hour, she and her colleagues brought by contract or "data associates" listened to random conversations and noted each word on their laptops.Amazon would say that work was essential for a Top Secret Product of Speech Recognition.The files included records with the most intimate moments in people's homes, writes Bloomberg.
They were happening in the fall of 2014 when Amazon revealed the Echo Portable speaker by the virtual assistant activated by voice Alexa.Amazon says about Alexa that it is a miracle of artificial intelligence in its first advertisement for Echo - in the clip a family asked and received news, answers to trivial questions and help on children's themes.But Slatis soon began to understand the extent to which people were behind the robotic magic they saw in the advertisement."O God, I work for this," she remembers what she thought then.Amazon surprised each cloud vocal command and was based on data analysts like her to learn the system.Slatis initially believed that he listened to paying tests that offered their vocal models in exchange for a few dollars.He realized it wasn't that.
The recordings she listened to with her colleagues were often intense, strange or very strange.People who seemed alone confessed their secrets and intimate fears: a boy expressed his desire to rape;There were men who treated Alexa as Joaquin Phoenix in the movie SF "Her".And as the transcript program increased with Alexei's popularity, so did the private information revealed in recordings.Other data associates remember children who shared their home address and phone number, a man who was trying to order sex toys, a guest at a party who was wondering if I didn't look at them.at that time."They couldn't know they were listened to," says Slatis."These people did not agree with this."She gave up in 2016.Five years after Slatis felt for the first that he is made of hen's skin, a quarter of Americans have "smart speakers" such as Echo, Google Home and Apple Homepod.(There are a few who even bought the Facebook portal, a smart video screen.) Amazon wins so far sales battle, reporting that over 100 million Alexa animated devices have been purchased.
But now, among the largest companies in the world there is a war to bring Alexa, Siri of Apple, Google assistant, Microsoft's Cortana and on Facebook's equivalent service in people's lives..Microphones are incorporated into phones, smart watches, TVs, refrigerators, SUVs and everything that means smart device.The Juniper Research consulting company estimates that by 2023 the world market for smart speakers will reach $ 11 billion a year and that there will be approximately
7.4 billion devices controlled by voice.This is a device for every person on the ground.
These cars do not create audio files from every decibel - technological companies say that their smart speakers only record audio when users activate, but introduce permanent microphones in kitchens and bedrooms that could capture sounds that users do not intend to intend toshare them with nobody."Microphones listening all the time are something worrying.I discovered that the users of these devices close their eyes and trust that companies will do nothing wrong with the registered data, "says Florian Schaub, professor at the University of Michigan, who studies human behavior around the voice control software.."There is this creepy erosion of private life that continues without interruption.People don't know how to protect themselves."
However, the so-called smart devices depend on thousands of poorly paid people who gather sound fragments with which the technological companies improve their electronic ears;Our lightest whispers have become some of the most valuable data sets of Silicon Valley giants.Earlier this year, Bloomberg News wrote for the first time about the extent of use by the people's technology industry to listen to the sounds collected from their users without permission..Current and former data associates like Slatis clearly show that the disadvantages of general audio surveillance were obvious for those with smaller financial stakes. „Nu s-a simţit niciodată ca fiind ceva corect", spune un transcriptor pentru un rival al Alexa, care, la fel ca majoritatea asociaţilor de date, a semnat un acord care-i interzice să vorbească despre munca sa. „Ce vând ei cu adevărat clienţilor?"
For the process of learning of cars to identify and respond to spoken sentences, it was most often needed to match the audio files with the transcribed text.It is a slow and expensive process.The first pioneers bought or built massive registration libraries-people reading newspapers or other pre-written materials at the microphone.Apple became the first major company to change the model when in 2011 delivered iPhone 4s with Siri, purchased a year before a research center funded by Pentagon.There was no need for recordings to be written and suitable in laboratories.Apple has sold over 4 million 4s phones in a few days and soon began to accumulate a free and natural vocal data.In the early years, the company trusted specialists in vocal software to use data in order to improve Siri's skills, but Apple returned to control around 2014. „Munca era foarte migăloasă: după ce ascultai 15 sau 30 de minute, căpătai dureri de cap", spune Tao Ma, fost specialist de vorbire la Siri.Apple's team gave up a large part of this work of outsourcing companies in Europe, including Globetech, Ireland.
In recent years, Apple has become more aggressive in the process of collecting and analyzing people's voices, being worried that the power to understand Siri's speed remained behind Alexa and Google Assistant.Apple has treated Siri's development as a verbal search engine they had to prepare to answer users' endless questions and increased audio analysis to feed the assistant lexicon.
Former collaborators describe the system as a Babel tower or something in George Orwell's 1984 dystopia.Some tell how at a Globetech office near the Cork Airport, Ireland were silent at MacBookuri, wearing headphones, with the mission to transcribe 1.300 records per day, each of them can be a single sentence or a whole conversation.(This quota has been reduced from 2.500 records, others say, to improve the accuracy rates.) Când un asociat de date făcea clic pe butonul de redare pentru a asculta o înregistrare vocală, computerul umplea o casetă de text cu cuvintele despre care credea că Siri „le-a auzit", apoi solicita angajatului să aprobe sau să corecteze transcrierea şi să meargă mai departe.
A program that collaborators used, called Crowdcollect, included buttons that allowed them to overcome several reasons-accidental triggering, lack of audio, indecent language-but, they say, there was no specific mechanism toreport or delete an inadequate offensive audio file, such.Collaborators who asked managers if they can skip too intimate records were told that no registration is too intimate.
Data associations often resisted only a few months, and training in confidentiality issues was minimal. Un fost colaborator care nu avea nicio calificare cu privire la munca sa spune că ascultarea utilizatorilor din lumea reală a fost ceva „absolut hilar".
În 2015, acelaşi an în care directorul executiv al Apple, Tim Cook, a descris confidenţialitatea ca fiind un „drept fundamental al omului", maşinăriile Apple prelucrau mai mult de un miliard de cereri pe săptămână.Until then, users had been able to activate an application of the operating program so that they no longer have to press a button on the iPhone to activate the vocal assistant;He always listens.In the use agreement, says Apple, it is stipulated that the vocal data can be recorded and analyzed to improve, but nowhere is it mentioned that people could listen to. „M-am simţit jenat ascultând alţi oameni", spune unul dintre foştii colaboratori, mai ales având în vedere cât de dese erau înregistrările copiilor.
Ten former Apple directors in the Siri Division say they have not seen and still do not see this system as a violation of private life.These former directors say the recordings were disassembled by Apple user IDs and assumed that users understood that the company processes their audio recordings, so what mattered if people have helped processing?
Foşti şi actuali colaboratori spun că cele mai multe solicitări pentru Siri sunt banale – „cântă o melodie a lui Justin Bieber", „unde este cel mai apropiat McDonald's", dar îşi amintesc că au dat şi peste mesaje extrem de grafice, cu conţinut rasist sau homofob.A former data analyst who worked on transcripts from Siri says that the employees in Cork were exchanging horror stories during cigarette breaks..A current analyst, asked to tell the most scandalous record that came through Crowdcollect, says it was something like a scene in Fifty Shades of Gray.
Apple argues that less than 0.2% of the requests for Siri are subject to human analysis, and former managers reject the testimonies of contractors as exaggeration."Oh, I heard someone who has sex or so.But you also hear people who emit gases or sneezes.There are all kinds of noises there when you start a microphone, "says Tom Gruber, co -founder of Siri."It is not as if the device aims to register people who make certain types of sounds.Is like an incident."
Până în 2019, după ce Apple a pus Siri pe produse precum căştile sale fără fir şi boxa HomePod, numărul de comenzi vocale procesate a ajuns la 15 miliarde; 0,2% din 15 miliarde reprezintă 30 milioane de „întâmplări" pe lună, sau 360 milioane pe an.