Resolvemos por meio deste blog estabelecer um canal direto de comunicação do grupo executivo da ODATA com nossos clientes, prospects e parceiros. É também uma forma de conhecer um pouco melhor nossa visão, forma de pensar e um pouco da cultura ODATA.
A great article on Artificial Intelligence (A.I.) has been published in the New York Times in December. After reading it, it became clear to me that computing and the “consciousness” of machines has accelerated so much in the past few years that one cannot but be concerned with the Hollywood doomsday stereotype of machines becoming more intelligent than humans (and then ruling us all). According to Times contributor Gideon Lewis-Krauss, this specific risk at this point in time is still science fiction. But a small group of researcher at Google and at other tech powerhouses have had a very significant breakthrough in the method at which they are training computer to do tasks that toddlers take for granted. These tasks include teaching computers to “see” and search images, such as identifying a cat or a chair in a photo. Any toddler will quickly point out a cat in a picture shown to her. But for a computer, this is significantly more daunting.
Mr. Krauss´ piece takes us inside the evolution of Google Brain, which started as a skunkworks effort at Google and today is permeating all the company. As Google CEO Sundar Pichai said it himself, “In the long run, we´re evolving in computing from a mobile-first to an AI-first world.”
Since the emergence of computers, researchers and engineers have been trying to teach machines to behave more like the human brain. Remember when IBM´s Deep Blue computer beat Gary Kasparov in a Chess match? That was in 1997. Since then IBM´s Watson smoked any living human in Jeopardy, a much-complicated game to teach a computer. But the way they went about it was by feeding large amounts of information and then defining parameters for search. For example, tagging millions of different photos of cats and then expecting the computer to find slight similarities among all of them and correctly interpreting a new picture that presents similar traits. This approach has limitations but it has taken us to today´s slightly intelligent computers, evidenced by a slew of personal assistants such as Siri (Apple), Alexa (Amazon Echo), Cortana (Microsoft) and Google´s own Google Assistant (name creativity not featured here).
This method is very time consuming as developers need to tag every piece of content to help computers search and interpret data. The breakthrough was in the way information is fed and how it is analyzed. These new algorithms search through large volume of unstructured data (not tagged), and through millions of iterations it “learns” the best way to arrive at the answer, or in our example, point out a cat in an image. Once its successful, it betters the method that generated a positive result, and so on.
The approach is not new, so why is it finally working? Besides throwing a huge amount of brain power to solve this problem by hiring the most prominent researchers in the field, this method only works if you have a large enough data pool (Google does!), and a very, very large computing power. Also, not a problem considering the large Data Center resources and server farms implemented by Google globally.
In the tech world, discoveries such as these generally take a long time to revert into something useful or palpable for our everyday lives, as they are tinkered with in the R&D labs for years. However, this was not the case as you can now test it for yourself the revamped Google Translate. Translation is a complicated task for a machine to perform, since language, unlike math, many times have multiple interpretations, depending on context and dialects.
Although the changes to Translate were launched without fanfare, the evolution is breathtaking. One of the first person to realize the improvement was Prof. Jun Rekimoto, from the University of Tokyo. He ran a few tests comparing Japanese-English translations, including his own interpretation of a Hemingway passage and the Japanese translation of F. Scott Fitzgerald´s “ The Great Gatsby” . In both cases the machine translation was as close as possible to his and to the professional translation of Fitzgerald´s masterpiece by Haruki Murakami. The results were almost indistinguishable.
I was impressed but not convinced, so I devised my own little test of Google Brain´s new code. Portuguese is a not so common language in the tech world. So, I decided to throw it a curve ball. I fed Google with an excerpt of Machado de Assis´s “The Posthumous Memoirs of Brás Cubas”. For the uninitiated, Machado de Assis is considered by many the greatest modernist novelist in Portuguese language. I thought that this would be difficult because the text is from 1881, so it still utilizes archaic Portuguese words and phrase constructions. Not only that, but Mr. Assis was many years ahead of his time, and his prose is full of wit and in most cases, hard even for a natural born Brazilian to fully understand. Additionally, Mr. Assis´ work never achieved large sales in English language, so the text is somewhat “obscure”. Below you can see the result, one of them is the passage from Gregory Rabassa´s translation, the other is from Google Translate. Mr. Rabassa is considered to have been one of the best translators of Spanish and Portuguese texts and books, ever.
Original Text: “ Vejo-a assomar à porta da alcova, pálida, comovida, trajada de preto, e ali ficar durante um minuto, sem ânimo de entrar, ou detida pela presença de um homem que estava comigo. Da cama, onde jazia, contemplei-a durante esse tempo, esquecido de lhe dizer nada ou de fazer nenhum gesto. ”
“I see her appear at the bedroom door, pale, shaken, dressed in black, and there to stand for a minute, unwilling to enter, or stopped by the presence of a man who was with me. From the bed where I lay I looked at her during this time, forgetting to tell her anything or make no gesture.”
“I see her appear in the door of my bedroom-pale, upset, dressed in black-and remain there for a minute without the courage to come in, or held back by the presence of the man who was with me. From the bed where I was lying I contemplated her all that time, neglecting to say to her or make any gesture. “
A.I. is sure to revolutionize many sectors and perhaps whole industries in the very near future. As usual, it will consume a ridiculous amount of energy and Data Center space. I don´t know if you agree, but I surely think that translation services (by humans) should not be considered as a prosperous career. And I also believe that Siri and Echo will soon evolve into Jarvis, HAL 9000 or the like. Let´s all hope they don´t become Skynet or the sentient machines of The Matrix.
If you haven´t figured out, the first passage was translated by Google.