Google spent most of 20 16 years redesigning its translation tool and making it driven by artificial intelligence. In this way, there is a disturbing and powerful thing. Google Translate was once famous for providing blunt but useful translations, and now it has begun to provide translated texts with fluent language and high accuracy. For people without professional translation training, this kind of text output is almost the same as manual translation. The New York Times published an article of 15000 words, calling it "the awakening of great artificial intelligence". Google Translation Engine soon began to learn new skills and figured out how to translate between two languages that had never been tried before: if it can translate English into Japanese and English into Korean, then it can translate Korean into Japanese. At the Google Pixel 2 mobile phone conference last month, Google took its ambitious plan one step further and launched a wireless headset that can translate 40 languages in real time.
Since IBM introduced the first machine translation system in 1954, the concept of perfect machine translation machine has occupied all the imaginations of programmers and the public. Science fiction writers seized on this idea and provided all kinds of utopian fantasies, from the universal translator of Star Trek to the Babel fish in the Hitchhiker's Guide to the Galaxy. Human translation can describe the meaning of the original text in fluent words, which is the holy grail of machine learning: a challenge to "complete artificial intelligence". If machine translation can reach the same level, it will mean that the machine has reached the level of human intelligence. The hype surrounding Google's progress in neuromachine translation shows that the "Holy Grail" is close at hand. Moreover, when this moment comes, human translation will also be eliminated.
However, translation has always been in the front line of employment panic caused by artificial intelligence, and they are not worried about it. In fact, some people are very happy. For those who have mastered the potential of artificial intelligence tools, their work efficiency and work demand are soaring.
Think of them as canaries in white-collar coal mines, and you can smell the wind and grass in any industry at the first time. At the moment, the canary is still singing, which proves that it is still safe. With the rapid development of machine deep learning, many industries have begun to realize that artificial intelligence can indeed accomplish tasks that were once thought to be completed by only humans. Unlike drivers and warehouse employees, knowledge workers are not in danger of being replaced immediately. But with artificial intelligence becoming an important part of their workflow, their work is also changing, and no one can guarantee that today's artificial intelligence tools will not become a threat in the future. This gives employees a choice: put aside conceit and embrace your new colleague in artificial intelligence, otherwise, you will be left behind. We are not living in the golden age of artificial intelligence, but in the golden age of artificial intelligence to improve productivity. It can be called the first passage era. Artificial intelligence is now powerful enough, and the first attempt in countless complex tasks is very reliable, but it is not strong enough to look threatening. We still need human beings to complete tasks that require more intensive thinking and subjective consciousness.
This kind of labor transfer is being carried out in various industries. Washington post's internal artificial intelligence Heliograf published about 850 articles last year, and human journalists and editors added analysis and rich details to these articles. In graphic design, artificial intelligence tools can now design drawings that can be initially passed for human designers to finally implement. In the field of film and publishing, new artificial intelligence tools can clear up piles of bad scripts to find the next explosive script and free editors from the endless submission queue. These artificial intelligence tools are like brave and strong young assistants: they are very capable and effective, but they still need an experienced manager to complete the heavy mental work. Of course, the manager must work with the machine to benefit from it.
Fennemore Craig is an Arizona law firm. Lawyers got on the train of artificial intelligence and tried a new technology from the startup ROSS Intelligence. With the combination of IBM's supercomputer "Watson" and proprietary algorithms, Ross is driven by artificial intelligence and inherits tools such as LexisNexis: it combs millions of pages of case law and records its findings in a draft memorandum. This process may take a human lawyer four days, while Ross spent about 24 hours. Ross won't be upset by fatigue and burnout: this tool can liberate people who work all night without feeling too hard.
Although Ross can also write, this is not its outstanding feature. Blake Atkinson has been a partner of Fennemore Craig for three years. According to him, Ross's writing level is "the level of a first-year law student". (Anthony Austin, a partner in the company, speaks highly of him: he says that in his opinion, Ross is as good as his colleagues in the first and second grades. ) This tool can generate neat memos. Although it is not a great writer Hemingway, it provides a practical first draft, which is full of abstracts of applicable case law, some basic analysis and a straightforward conclusion. Then, a human lawyer will add deeper analysis and modify the language to make people feel happier when reading, at least for lawyers. Austin said: "It can expose us to interesting and dry articles. When you say,' God, I don't care about the steam engine of 1885, what I really want to do is to write something interesting to make the judge or the opposing lawyer feel,' God, I'm finished. In the end, tools like ROSS will almost certainly reduce the need for human lawyers in evidence collection procedures.
It is not clear how this will change the employment of entry-level lawyers, who usually have to study old case law hard and work irregular hours. But the ability of in-depth analysis and excellent writing ability is far beyond Ross's ability. Lawyers are not worried that Ross will take their jobs, which is crucial to the success of this startup. After all, who wants to cultivate their own body double? Because of this, CEO Andrew Arruda touted Ross as a productivity tool, not an artificial intelligence lawyer; It allows lawyers to serve more clients and focus on the interesting parts of their work. Austin put it more succinctly: with Ross's help, he said, "You look like a rock star."
For many translators, the superhuman productivity brought by artificial intelligence is nothing new. When Alessandro Cartland started his translation career in 2003, he predicted that he could earn 175 USD by translating 2000 words every day. He used a computer-aided translation tool, which occasionally suggested translating individual phrases according to his previous translations, but translation was a very hands-on process. Cattland said that nowadays, translators working in artificial intelligence should translate 8,000 to10,000 words a day (adjusted for inflation) to earn the same money. This process is called "post-editing machine translation" (PEMT), which includes pre-translation by machines, and then human translators sort out the language and check inappropriate terms to ensure that the tone, context and cultural meaning of the translation are in line with the original text.
Cartland said: "You must figure out what parts of your work can be replaced by machines and what parts of your work as a human can bring value to yourself." He is now the vice president of operations of Translated, which develops translation tools based on artificial intelligence. In April this year, translation began to provide neural machine translation for post-editing machine translation, which greatly improved productivity, especially in the translation of German, Russian and other languages. Due to the complexity of grammar, the grammar of these languages is more complicated, and additional text adjustment was needed before.
Post-editing machine translation is not new. This niche market has been developing since at least the 1980s. However, with the emergence of neural machine translation, post-editing machine translation is more popular. According to the data of Common Sense Consulting, a market research company, in the next few years, the demand for post-editors is expected to exceed other areas of the language industry, and the growth rate of enterprise translation business may reach double digits. Kamenson Consulting warned, "Even if the language industry wants to add new translators at an unprecedented rate, the current method cannot keep up with this growth rate." Some people say that cooperation with machine translation is becoming more and more compulsory: Spencer Green, CEO of Lilt, said that machine translation "is now a requirement. For older translators, they don't even need to use translation memory software." Lierte Company is a machine translation platform.
Charlotte Brassler, a translator in Sydney, said that last year, machine translation tools have become very good, unless her use of machine translation tools will break the confidentiality agreement (which is an unusual obstacle), and she tends to welcome the development of machine translation tools. By cooperating with excellent artificial intelligence, she can take over more projects and make time to translate more creative texts, which are usually not translated by machines.
But this is also changing: Brasler said that in the past year, since joining the neural network, Google Translation has done a very good job in translating sales and marketing materials, in which translation involves using rich languages and explaining idioms. Of course, Google translation engine is not a poet, but it is rapidly improving the ability of human beings in areas that machines have long thought could not be conquered. For those translators who define themselves by their own translation skills, this is a bitter pill to swallow.
The technological leap will always overcome many obstacles. Some people can't stand the idea of working with machines. They prefer to bury themselves in imaginary magazines and pretend that nothing has changed. For these people, this gushing of "artificial intelligence" is a crisis of life and death. Of course, a computer can sift through data and even piece together a basic sentence-but can it write an article that makes you cry? Can it analyze the nuances of an idiom, or find the next best-selling novelist, or persuade the Supreme Court judge to change his mind?
Not yet, but it can help you achieve your goal. As some of the most creative industries begin to try artificial intelligence, they face resistance. In April this year, Blacklist (a network connecting filmmakers and screenwriters) announced that it would cooperate with an artificial intelligence company called ScriptBook to evaluate some scripts. Brian koppel, executive producer of the TV series Billions, called the tool "rude". The blacklist quickly canceled the cooperation with the script book, which will scan the script for personality analysis, target population statistics and box office success. Although the startup has successfully established partnerships with two major film companies, Nadella Aselmeyer, CEO of the startup, said that most filmmakers have not yet overcome their fear of this tool.
Asselmeyer said: "A few years ago, people thought that our creativity was not threatened, because artificial intelligence could not be as creative and unique as human beings. However, this is not true. " When people in the industry accused her of creating a tool to steal work, she told them that their work was indeed threatened, but it was not from artificial intelligence. Instead, she said to the opponents, "Only those who have learned how to cooperate with machines will take your job away. If you continue to turn a blind eye to this, you will lose your job. "
A similar tool is StoryFit, which provides services such as box office prediction, script structure and style analysis, and interpretation of emotional composition of stories. As TJ Barak explained, his studio Adaptive Studios will never adopt a script just because of what is seen in StoryFit's report, but his team may consider how to improve the script according to what they know. Barak said: "If this shows me that it may have trouble in the market on these specific issues, what can we do to improve this script?" We can adjust some plot points. "Can we add more emotional elements here or there?"
People have just begun to downplay the hype of artificial intelligence and begin to study how artificial intelligence tools can help their work. Monica Landers, CEO of StoryFit, said that she has recently begun to be cautious about her company's products. But she still needs to be cautious. She hesitated when I asked her about the company's next plan. She said, "If you start talking about future plans too early, it will still make people nervous."
However, at present, the work of translators, lawyers, doctors, journalists and literary agents is safe. Some people even say that their work is better than ever. But now we find ourselves in a strange situation. We must admit that artificial intelligence is quickly mastering tasks that we have long thought impossible for machines. We must accept the fact that embracing artificial intelligence is rapidly becoming a prerequisite for achieving excellent results in many fields. We must welcome these new artificial intelligence colleagues and correct them when they make mistakes. At the same time, we have to admit that at some point, we may have taught them enough to make their position in the company more important.