One of Google's most heavily used products is Google Translate, its computer-translation service. Its 500 million monthly users enter 140 billion words in nearly any language and almost instantaneously get a translation of the word or phrase into another language.
As you would expect, the computer translations were competent, but were clearly not up to the translations that could be provided by a skilled human.
Then one weekend in November of 2016 that all changed, according to an article by Gideon Lewis-Kraus (Lewis-Kraus, 2016).
He illustrates using examples that are translations of the opening of Ernest Hemingway's short story "The Snows of Kilamanjaro" as rendered from the Japanese translation back into English.
For reference, here is Hemingway's original:
Kilimanjaro is a snow-covered mountain 19,710 feet high, and is said to be the highest mountain in Africa. Its western summit is called the Masai "Ngaje Ngai," the House of God. Close to the western summit there is the dried and frozen carcass of a leopard. No one has explained what the leopard was seeking at that altitude.
First, here is the translation as done by the "old" version of Google Translate:
Kilimanjaro is 19,710 feet of the mountain covered with snow, and it is said that the highest mountain in Africa. Top of the west, "Ngaje Ngai" in the Maasai language, has been referred to as the house of God. The top close to the west, there is a dry, frozen carcass of a leopard. Whether the leopard had what the demand at that altitude, there is no that nobody explained.
Here is the same passage done by the improved Google Translate:
Kilimanjaro is a mountain of 19,710 feet covered with snow and is said to be the highest mountain in Africa. The summit of the west is called "Ngaje Ngai" in Masai, the house of God. Near the top of the west there is a dry and frozen dead body of leopard. No one has ever explained what leopard wanted at that altitude.
The only evidence that the "new" Google Translate version was not done by a human are the missing articles before each occurrence of the word "leopard" in the passage. (Lewis-Kraus, 2016)
The improvement of Google Translate was the result of a decision made by Google to move its products from traditional programming to Artificial Intelligence (A.I.) or "machine learning."
The difference is between strategies used to solve the problem of how to emulate human intelligent behavior with a computer.
The original version of Translate was based on the strategy that if you want to emulate the human ability to translate from English to Japanese, "you would program into the computer all of the grammatical rules of English, and then the entirety of definitions contained in the Oford English Dictionary, and then all of the grammatical rules of Japanese, as well as all the words in the Japanese dictionary, and only after all of that feed it a sentence in a source language and ask it to tabulate a corresponding sentence in the target language." (Lewis-Kraus, 2016)
This strategy is a cumbersome (program all of the English grammatical rules!) Further, rule-based programming works best in contexts like mathematics or chess, where the rules are well-defined. The dictionary meaning of words and how they are used in actual writing and speech are often mis-matched. For example, the phrase "the minister of agriculture" might be rendered as "the priest of farmers," for example. (Lewis-Kraus, 2016)
Machine learning, another strategy for achieving A.I. takes its inspiration from how the brain works. Our brains were built from the bottom up. The well-studied roundworm C. elegans has 304 neurons and these are adequate to operate the organism. To operate the human organism it takes around 100 billion neurons with each neuron having around 10,000 connections, or perhaps 1,000 billion connections.
What these neurons do, whether in C. elegans or H. sapiens, is exactly the same: "they pass along an electrical charge to their neighbors or they don't. What's important are less the individual neurons themselves than the manifold connections among them. This structure, in its simplicity, has afforded the brain a wealth of adaptive advantages." It can withstand damage; it can store copius amounts of information; and it can identify patterns, even messy and inconsistent ones and so can deal with ambiguity. (Lewis-Kraus, 2016)
The electrical charges being passed are of course signals generated to respond to and to act in response to signals from the organism's internal and external environment. Whether the charges are strengthened or weakened depends on whether they are used more often and become stronger, or not used and weaken and fade away. The process of strengthening by use or weakening by disuse is what happens when we learn. Modern computers with very fast processors and huge numbers of artificial neurons (circuits) can be trained to recognize whether there is a cat in a picture using the same process; "the reason that the network requires so many neurons and so much data is "that it functions, in a way, like a sort of giant machine democracy. Imagine you want to train a computer to differentiate among five difference items. Your network is made up of millions and millions of neuronal "voters," each of whom has been given five different cards: one for cat, one for dog, one for spider monkey, one for spoon and one for defibrillator. You show your electorate a photo and ask, 'Is this a cat, a dog, a spider monkey, a spoon or a defibrillator?' All the neurons that voted the same way collect in groups, and the network foreman peers down from above and identifies the majority classification: 'A dog?' You say: 'No, maestro, it's a cat. Try again.'" (Lewis-Kraus, 2016)
The new Translate is not hard-wired with English grammar rules. Instead, it uses machine learning precisely how a baby learns language.
The more people interact with Translate, the better it becomes at using language. Its developers "instructed the networks on enormous banks of "labeled" data--speech files with correct transcriptions, for example--and the computers improved their responses to better match reality." (Lewis-Kraus, 2016)
This was how the AlphaGo computer was able to master the game of Go and to defeat a highly skilled human competitor in a game that requires making choices from among many thousands of possible moves (unlike chess).
Machine learning is highly flexible. Lewis-Kraus reports that a Translate engineer took a network he designed to judge artwork and used it to drive an autonomous radio-controlled car.
A network designed to recognize specific animals can be transformed into one that works with a CT scanner. Samsung has announced that it has developed a program that can reliably recognize tumors well before trained humans can.
Much of school practice is based on teaching the rules: grammar, syntax, vocabulary. It, like its computer counterpart is laborious both for teachers and students and doesn't work very well where the rules are fuzzy.
The human brain is the model that the improved Translate is built on. What can we learn from Translate to improve student learning?
Lewis-Kraus, Gideon (2016). The Great A.I. Awakening: How Google used artificial intelligence to transform Google Translate, one of its more popular services — and how machine learning is poised to reinvent computing itself. Retrieved from http://nyti.ms/2hMtKOn
Dr. John Holton
Dr. John Holton joined the S²TEM Centers SC in July of 2013, as a research associate with an emphasis on the STEM literature including state and local STEM plans from around the nation.