智能菌（公眾號：Smartman163）為媒體智能工作室旗下專注人工智能熱點話題和產(chǎn)品的欄目，未來已來，還不上路？
本文為智能菌第10期：《中國詩詞大會》讓大眾重燃對詩詞的喜愛，清新脫俗的詩詞讓人心動，當(dāng)然，谷歌人工智能作的詩詞也能讓你怦然心跳！
央視一套的《中國詩詞大會》經(jīng)過了最后的巔峰對決，00后才女武藝姝奪得桂冠。北大博士陳更、詩刊編輯彭敏、8歲馮子一等無數(shù)才子佳人也都火的“一塌糊涂”。
近年來，良莠不齊的綜藝節(jié)目和歌唱節(jié)目充斥著大屏幕，而《中國詩詞大會》的出現(xiàn)猶如一股清流，沁人心脾，讓人如沐春風(fēng)。從寂寞空庭春欲晚，梨花滿地不開門到忽如一夜春風(fēng)來，千樹萬樹梨花開，我們品味著人生百態(tài)，你以為只有人才能寫詩做詩？谷歌AI也能寫詩了?？吹揭蝗壶B兒飛上天空時不用再說“哇，好多鳥”，可以說“鳥啄冰潭玉鏡開”等自創(chuàng)詩句了。
在嘗試?yán)斫饬_曼史小說之后，現(xiàn)在谷歌的一個人工智能項目居然也開始提筆寫詩。
“there is no>there is no>they were the>they were the>he had to be with me. she had to be with him.
i had to do this. i wanted to kill him.
i started to cry.
i turned to him.”
谷歌正在同美國馬薩諸塞州大學(xué)和斯坦福大學(xué)合作，提高一種被稱為“循環(huán)神經(jīng)網(wǎng)絡(luò)語言模型”（RNNLM）的自然語言技能。這種技術(shù)可用于機器翻譯和圖文識別等領(lǐng)域，通過逐一對每個單詞前面一個單詞的分析構(gòu)建出整個句子。
目前，RNNLM尚無法在其生成的語句中實現(xiàn)貫穿全篇的主題或特征，比如很好地圍繞一個給定主題展開談話。由算法產(chǎn)生的每個句子不能和下一句很好地銜接。這項工作的論文在康奈爾大學(xué)開放科學(xué)論文檔案arXix發(fā)表，詳細描述了研究人員為了實現(xiàn)全局貫穿主題所進行的努力。他們在研究中使用到了變分自編碼器，所產(chǎn)生的結(jié)果包括一些可以被稱作詩句的文字。
研究人員給系統(tǒng)展示開篇和結(jié)束的句子，然后要求機器填補空白。下面的例子中，首行和末行的粗體文本為研究人員給出的示例，中間的文本是AI生成的。研究人員稱，這些句子是基于AI對數(shù)千本羅曼史小說的理解產(chǎn)生，這使得一些主題“相當(dāng)戲劇化”。
“he was silent for a long moment.
he was silent for a moment.
it was quiet for a moment.
it was dark and cold.
there was a pause.
it was my turn.
this was the>it was the>it was her turn to blink.
it was hard to tell.
it was time to move>he had to do it again.
they all looked at each other.
they all turned to look back.
they both turned to face him.
they both turned and walked away.”
生成的句子合乎語法，并保持了主題的連貫，并且大部分與給定的詩句契合。其他的詩句在詩意上有所欠缺，但仍然很好地保持了主題的統(tǒng)一。
“no. he said.
“no,” he said.
“no,” i said.
“i know,” she said.
“thank you,” she said.
“come with me,” she said.
“talk to me,” she said.
“don’t worry about it,” she said.
i don’t like it, he said.
i waited for what had happened.
it was almost thirty years ago.
it was over thirty years ago.
that was six years ago.