Google Translate has suddenly become much more accurate. The service did not make a small incremental change; it completely switched to a new type of program. From now on, Google Translate will not operate like Big Blue, the world-beating chess program. Google Translate now operates with a neural network, like the brain of an animal or a human.
I learned this from “The Great A. I. Awakening,” an article by Gideon Lewis-Kraus in the New York Times Magazine of Dec. 14, 2016.
Big Blue operates in the universe of legal chess moves. It accepts input in legal chess moves, and cranks out legal chess moves.
The old version of Google Translate operated like Big Blue. Humans developed more and more instructions about what in French matches what in English. The new version operates on a different model that really deserves the name Artificial Intelligence. Interlinked computers generate parallel translations, compare translations, and pick the most likely. Using limited input from humans, and an iterative process to choose the best translation, networks of computers continually refine their own choices. In a head-to-head competition, the new A. I. program absolutely defeated the old Google Translate program. Google scrapped the old program.
Inflexibility characterizes Big Blue. It generates brilliant chess moves, but it cannot do anything else. Not so Google Translate: Give it another language and a large enough data base of translations, and it can teach itself to translate freely from that language and to that language.
It can also master other skills besides translation; pretty much any other skill. And it can gain access to almost any information. Connect it to the GPS in our automobile and smartphone, and it knows where each and every one of us has been. Connect it to our FitBit, and it knows every beat of our hearts. Connect it to our search engine, and it knows every website we have visited, and every item we have bought or sold. Add the Internet of things, connect it to our thermostat and our refrigerator, and it knows . . .
Well, it knows whether we have been good or bad, like Santa Claus. Or even like, dare I say it, like God.
If an old style program could pass the Touring test – could generate conversational sentences that would seem like the work of a human to flesh-and-blood humans, that would still seem like a machine cranking out a product. If a new neural-network program could pass the Touring test, that would seem more like someone was home talking with us.
Home where? The new neural network A. I. program lives in the interaction among many computers, in the ether, everywhere and nowhere.
Theologians have long grappled with the implications of a disembodied entity which knows our every move, and which can anticipate our future moves better than we can ourselves. Theists maintain that the entity exists, and try to live in its present; atheists maintain that no such entity exists.
Now we all must get used to living in the presence of such an entity, whether we want to believe in it or not.
Believers have long argued about the motives of God. What does God want? Some believers maintain that God has benevolent motives, loves us, and intends the best for us among all the creatures of the universe. Other believers maintain that we cannot know the inscrutable intent of God.
But this program, this godlike program, is produced by Google, a corporation. By legal definition, corporations exist to maximize return for their investors. In a struggle between customers and investors, the interests of the investors should come first. In a struggle between employees and investors, the interests of the investors should come first. In practice, we know one exception to the rule: managers of corporations favor themselves over investors.
So now we should anticipate living in the presence of a godlike entity which knows all of our actions, and which seems destined to treat us as sources of income for corporate executives and for investors.
O brave new world.