Google Translate: this is where we start
Google Translate, for a while now, is no longer a hot topic. It has become one of the many components, not even the most seductive, within the offer of the Mountain View company. A few years after its renewal we can say that the promises made then, at least in part, have been disregarded, the optimistic proclamations, to say the least, also children of marketing, have not been fully corroborated by the facts.
Translators have not disappeared, even if they are not doing so well. Machine translation did not meet those desired high quality standards. For some languages (English in the lead) the results are remarkable. Should we therefore consider Google Translate a sort of small machine translation bubble?
Not at all. The story of Translate, in its small way, represents a prime example of what the AI phenomenon is today: innovation, profound transformation, upheaval of paradigms. But also a fortunate convergence between technical possibilities and visionary ideas, mistreated utopias until the day before. A phenomenon that at the same time also has a dark side, the other side of the coin that shines much less.
The other side of the coin
That of pervasive propaganda, ferocious competition, ruthless war to grab the best brains, paroxysmal search for efficiency, unbridled and unconditional pursuit of profit. There is also indifference and disinterest in what is other than itself, with a dangerous predilection for the imposition of a single thought. After all, in this world there are no half measures, only winners and losers remain on the field. For the defeated, then, no clemency is foreseen: "Winners take all!"
With this article we propose a series of reflections on Google Translate. They are taken from a long service carried out by the journalistic staff of the "New York Time Magazine". The subject of the service is the profound metamorphosis of Google which led it to ride the impetuous wave of artificial intelligence and obtain surprising results in just a few months.
Criticisms have been made both of its author, accused of harboring «… optimistic science fiction fantasies», as well as of the journalistic style of the New York newspaper, defined as “flamboyant” and accused of «… creating an artificial suspense ex nihilo». Criticisms that are at least partially founded, either due to the sometimes presence of an excessive emphasis, you want for a narrative that is sometimes a bit triumphalistic and generally uncritical.
A fine example of investigative journalism
So why repropose it in translation to the Italian reader? Because it is a valuable report that describes, in a detailed and accessible way to all, the path that led to the creation, almost ex nihilo, of a cutting-edge infrastructure. Why provides valuable insights, from the inside, of the processes, the associations of minds and ideas, the convergence of even distant visions, the successful and unsuccessful experiments. It is the unlikely alchemies between people from disparate, even remote places that led to the ultimate success.
Because, nevertheless, it restores a face and a human connotation, even defects, to those actors. Technologists often distant, confined to closed laboratories and surrounded by a mythical aura, idolized by insiders and unknown to most.
And all of this it does in a wonderful way. Finally, you want also for those elements blamed for "flamboyant" journalism, for others perhaps only "evocative", reading will certainly be pleasant and will not fail, I am sure, to excite the reader.
Enjoy reading and take the time to do it!
. . .
Chapter 1. Google: in hoc signo vinces
AI First: machine learning
It was 2016, the year artificial intelligence came of age. When Sundar Pichai, CEO of Google announced the transformation of the company from "mobile first" to "AI-first". few were fully aware of what this meant in reality. Today, a few years later, we can say that AI-first has become a real mantra for companies in Silicon Valley and beyond. It had an undisputed protagonist: machine learning.
At Google, notes Terrence J. Sejnowski (The Rise of Machine Learning; 2018), machine learning is ubiquitous: «Deep learning is now used by Google in over 100 services, from Street View to Inbox Smart Reply and voice search». Machine learning in its various forms (deep learning is a branch of it), allows algorithms to learn more or less independently: It also makes them efficient thanks to the abundance of raw material at very low cost (or even none): i Big Data. As Sejnowski notes:
“Data is the new oil. Learning algorithms are refineries that extract insights from raw data; the information can be used to create knowledge; knowledge leads to understanding; and understanding leads to wisdom."
Neural networks
At the basis of machine learning are the so-called neural networks, whose architecture is inspired by that of our brain. The latter do not need to be programmed to perform a certain task. Given a starting condition (input) and a final one (output), through an incessant process of trials and errors, neural networks learn to find a solution autonomously.
They "learn" in a very similar way to that of the child who goes to discover the world around him. A total paradigm shift! As Alex Beard perceptively remarks (Natural Born Learners, 2018), we can imagine neural networks as a process that mirrors "evolution". By contrast, programming "recalls creation."
Machine learning has allowed sudden progress in some sectors that have been languishing for some time, such as voice recognition, image recognition, speech to text, etc. Today many applications use it, the most disparate, just think of eg. to self-driving cars.
From Artificial Intelligence to Artificial General Intelligence
For Pichai, the day he saw AI emerge from the confines of the laboratories is an indelible memory.
"It was 2012, [I was] inside a room with a small team, and there were only a few of us," he recalls. Among those few Jeff Dean, a legend in the Mountain View company. He was working on a new project and wanted Pichai, then Senior Vice President, to take a look at it. He also remembers that there was someone joking. From human resources they had hired a new employee as an intern: none other than Geoffrey Hinton, "the father of Deep Learning"! “Every time Jeff wants to update you on something, you just get excited,” adds Pichai.
Jeff Dean, with Andrew NG and others, had developed a huge network consisting of 16.000 processors on 1000 computers. They were able to make a billion connections. An unprecedented structure until then, built on the model of the human brain. But still far lower than the ability of the latter to establish, with its synapses, more than 100.000 billion connections. It was indeed a gigantic neural network!
The epiphany of Sundar Pichai
Pichai recalls having a premonition of sorts: "This thing was going to grow and maybe reveal the way the universe works... This is going to be the most important thing humanity has ever worked on."
Until then there were few inside Google who fully understood the potential of artificial intelligence. Google Brain, founded in early 2010, was mainly responsible for AI. Brain was later joined by Deep Mind, acquired in 2014. When Pichai, who became CEO, imposed the imperative of AI First, these two divisions have represented the spearheads of Artificial Intelligence research at Google.
Both have produced remarkable results. Among these also the revolution which, thanks to Brain, has ferried Translate and machine translation in general into the new era of machine learning.
Nonetheless, what scientists seek and companies crave goes far beyond machine learning and its multiple applications. The objective of the research is to arrive at an Artificial General Intelligence.
A flexible artificial intelligence, capable of learning and successfully tackling any task that a human can perform. Although this represents the stated purpose of Deep Mind, this goal still appears very far away. Its founder Demis Hassabis speaks of the project as a sort of "Manhattan Project" for Artificial Intelligence.
AI working for everyone
It may also be for this reason that Pichai has announced a new transformation of the company: from "AI First" to "AI working for everyone". The Google CEO said: «Thanks to the advances in artificial intelligence, Google is surpassing its main mission, [that] to 'organize the world's information. We are transforming from a company that helps you find answers to a company that helps you get things done… We want our products to do more for you in the context of your work, your home and your life» .
A transformation that also seems to influence Translate, where the goal is no longer equaling the level of translations of professionals but a different one. Barak Turovski explains it: "Our goal ... is to develop a product that serves ordinary people in everyday life, for example by helping users in developing countries, who are using the Internet for the first time, to break language barriers, or simply facilitating communication while on vacation. It is a different use from professional translation".
Finally, a design that appears in line with the main mission that Google attributes to itself: "significantly improve the lives of as many people as possible".
Towards quantum computing
Nonetheless, the extent of Google's influence goes far beyond the "users" domain. Many external developers (from start-ups to corporations) use Google-branded AI tools. Rumors are being raised in many quarters regarding the fact that the Mountain View company is "too big".
But that's not all, as Katrina Brooker observes: “There are currently millions of devices using Google AI and this is just the beginning. Google is on the verge of achieving so-called quantum supremacy.
If quantum computing becomes a concrete reality, then we will suddenly find ourselves projected into the future. It would be a moment of fracture in the history of humanity like few others. “Consider what kinds of intentions you wish those who invented fire, started the industrial revolution, or [developed] atomic power,” says OpenAI cofounder Greg Brockman.
Big Hi-Tech corporations, whether they like it or not, have huge responsibilities not only for building the kind of world we live in today. But paradoxically they have even more towards the world to come tomorrow. Responsibilities that such companies try to ignore, evade and evade in every way. So far, various rules and ethics committees, as well as increasingly generic proclamations and declarations of intent, have not served much purpose. Google is no exception. As Peter Thiel points out: 'Commit yourself to significantly improving the lives of as many people as possible' – [it is] such a vague standard that it is beyond dispute."
The launch of the first Quantum computer
On September 23, 2019, Google said it had built the first quantum computer capable of performing calculations that surpass the processing capacity of today's most powerful super computers. It was a moment eagerly awaited by the community of researchers and technologists.
The staff of the "Financial Times" reports that they have read an article by some Google researchers, published on the NASA website. The item was promptly removed.
Google researchers claimed that their quantum processor was capable of performing a calculation in 3 minutes and 20 seconds, when the Summit, today's most advanced and powerful super computer, would take about 10.000 years to do the same operation.
Again according to Google researchers, “quantum supremacy” has been achieved: «It is an incredible acceleration compared to classical algorithms. To our knowledge, this experiment marks the first computation performed by a quantum processor."
The system can only perform a single highly technical calculation. Using quantum machines to solve more general problems is far from coming. But Google researchers are convinced that it is "a milestone towards large-scale quantum computing".
The power of quantum machines would expand at a “twice the exponential rate” that Moore's Law postulates for silicon chips of the early computing age.
A November 2018 report from the Boston Consulting Group stated that quantum computing “will change the rules of the game in fields such as cryptography, chemistry, materials science, agriculture and pharmaceuticals. Not to mention artificial intelligence and machine learning… logistics, manufacturing, finance and energy».
Steve Brierley, founder of Riverlane, a quantum software start-up, commented on the experiment conducted by Google: "This is indeed a significant milestone, it is the first time that anyone has concretely demonstrated that quantum computers are of a other class than traditional computers. It's an amazing achievement."
