等不及馬丁新作，人工智能續(xù)寫《冰與火之歌》！想看戳這里

從第一季拖到第七季，《權(quán)力的游戲》的凜冬終于來了，囧雪和龍母匯合，夜王也帶著他的異鬼大軍攻破了長(zhǎng)城，然而維斯特洛大陸上的生死大戰(zhàn)剛拉開帷幕，第七季就這么結(jié)束啦。據(jù)《權(quán)游》演員透露，劇終季第八季至少要等到明年秋季才能播出。

為了給苦苦等待的粉絲們找點(diǎn)樂趣，軟件工程師Zack Thoutt讓循環(huán)神經(jīng)網(wǎng)絡(luò)人工智能技術(shù)學(xué)習(xí)該劇原著《冰與火之歌》前五部的內(nèi)容，然后續(xù)寫五章劇情。這些人工智能創(chuàng)作的情節(jié)與粉絲早前的一些推測(cè)部分吻合，比如，詹姆最終殺死了瑟曦，囧雪成了龍騎士，而瓦里斯毒死了龍母。如果你感興趣，可以在GitHub的主頁上查看所有章節(jié)。

下面來了解一下人工智能是如何做到的：

After feeding a type of AI known as a recurrent neural network the roughly 5,000 pages of Martin's five previous books, software engineer Zack Thoutt has used the algorithm to predict what will happen next.

軟件工程師Zack Thoutt讓一種名為循環(huán)神經(jīng)網(wǎng)絡(luò)的人工智能技術(shù)學(xué)習(xí)了《冰與火之歌》前五部近5000頁的內(nèi)容，然后利用該算法預(yù)測(cè)接下來的情節(jié)。

According to the AI's predictions, some long-held fan theories do play out - in the five chapters generated by the algorithm so far, Jaime ends up killing Cersei, Jon rides a dragon, and Varys poisons Daenerys.

根據(jù)人工智能的預(yù)測(cè)，一些粉絲早前的推測(cè)的確出現(xiàn)了。在該算法目前撰寫的五章內(nèi)容中，詹姆最終殺死了瑟曦，囧雪成了龍騎士，而瓦里斯毒死了龍母。

如果你感興趣，可以在GitHub的主頁上查看所有章節(jié)。附上傳送門：

https://github.com/zackthoutt/got-book-6/tree/master/generated-book-v1

Each chapter starts with a character's name, just like Martin's actual books.

和馬丁本人撰寫的小說一樣，每章打頭的文字都是一個(gè)角色的名字。

But in addition to backing up what many of us already suspect will happen, the AI also introduces some fairly unexpected plot turns that we're pretty sure aren't going to be mirrored in either the TV show or Martin's books, so we wouldn't get too excited just yet.

不過，我們也不要太過興奮，因?yàn)槌舜嬖诤芏嗳艘呀?jīng)預(yù)測(cè)會(huì)發(fā)生的劇情外，這個(gè)人工智能算法還引入了一些令人意外的情節(jié)，它們絕對(duì)不會(huì)出現(xiàn)在電視劇或馬丁的小說中。

For example, in the algorithm's first chapter, written from Tyrion's perspective, Sansa turns out to be a Baratheon.

例如，算法編寫的第一章從小惡魔的視角寫道，珊莎其實(shí)屬于拜拉席恩家族。

There's also the introduction of a strange, pirate-like new character called Greenbeard.

書中還出現(xiàn)了一個(gè)名叫Greenbeard的怪咖，這個(gè)新角色的身份和海盜類似。

"It's obviously not perfect," Thoutt told Sam Hill over at Motherboard. "It isn't building a long-term story and the grammar isn't perfect. But the network is able to learn the basics of the English language and structure of George R.R. Martin's style on its own."

Thoutt在接受Motherboard采訪時(shí)告訴山姆•希爾，“這個(gè)算法顯然并不完美，它不能編寫長(zhǎng)篇故事，語法也有問題。但是神經(jīng)網(wǎng)絡(luò)可以自學(xué)英語的基本語言知識(shí)以及馬丁的文風(fēng)結(jié)構(gòu)。”

Neural networks are a type of machine learning algorithm that are inspired by the human brain's ability to not just memorize and follow instructions, but actually learn from past experiences.

神經(jīng)網(wǎng)絡(luò)是一種機(jī)器學(xué)習(xí)算法，設(shè)計(jì)靈感來自于人腦的記憶能力、遵循指令的能力以及從過去經(jīng)驗(yàn)學(xué)習(xí)的能力。

A recurrent neural network is a specific subclass, which works best when it comes to processing long sequences of data, such as lengthy text from five previous books.

一個(gè)循環(huán)神經(jīng)網(wǎng)絡(luò)是一個(gè)特定的子集，最擅長(zhǎng)處理長(zhǎng)的數(shù)據(jù)序列，比如《冰與火之歌》前5部冗長(zhǎng)的文本。

In theory, Thoutt's algorithm should be able to create a true sequel to Martin's existing work, based off things that have already happened in the novels.

理論上，Thoutt的算法應(yīng)該能基于書中已經(jīng)出現(xiàn)的劇情創(chuàng)作出《冰與火之歌》真正的續(xù)集。

But in practice, the writing is clumsy and, most of the time, nonsensical. And it also references characters that have already died.

但實(shí)際上，這個(gè)算法的寫作能力還很低級(jí)，大部分內(nèi)容都不知所云，還會(huì)提到已經(jīng)死掉的角色。

"A perfect model would take everything that has happened in the books into account and not write about characters being alive when they died two books ago," Thoutt told Motherboard.

Thoutt告訴Motherboard：“完美的算法模型能把書中的所有劇情考慮在內(nèi)，且不會(huì)再讓兩部以前去世的角色再次復(fù)活。”

"The reality, though, is that the model isn't good enough to do that. If the model were that good authors might be in trouble ... but it makes a lot of mistakes because the technology to train a perfect text generator that can remember complex plots over millions of words doesn't exist yet."

“然而，實(shí)際上這個(gè)算法現(xiàn)在還不夠完善。如果它有那么完美的話，作家們可能就要丟飯碗了……完美的文字創(chuàng)作機(jī)器可以記住數(shù)百萬字的復(fù)雜劇情，現(xiàn)在的技術(shù)還不能訓(xùn)練出這種功能，它會(huì)犯很多錯(cuò)誤。”

One of the main limitations here is the fact that the books just don't contain enough data for an algorithm.

最主要的局限之一是書中包含的數(shù)據(jù)對(duì)一個(gè)算法而言是不夠的。

Although anyone who's read them will testify that they're pretty damn long, they actually represent quite a small data set for a neural network to learn from.

雖然《冰與火之歌》的讀者都認(rèn)為這部小說太長(zhǎng)了，但是對(duì)于神經(jīng)網(wǎng)絡(luò)要學(xué)習(xí)的數(shù)據(jù)集來說，這些內(nèi)容太少了。

But at the same time they contain a whole lot of unique words, nouns, and adjectives which aren't reused, which makes it very hard for the neural network to learn patterns.|

此外，書中包含了許多獨(dú)特的詞匯、名詞和形容詞，它們不能重復(fù)出現(xiàn)，這使得神經(jīng)網(wǎng)絡(luò)很難學(xué)習(xí)到模式。

Thoutt told Hill that a better source would be a book 100 times longer, but with the level of vocabulary of a children's book.

Thoutt告訴希爾，更合適的數(shù)據(jù)源是一本比《冰與火之歌》長(zhǎng)100倍，且詞匯水平相當(dāng)于兒童圖書的書籍。