Microsoft Created a Twitter Bot to Learn From Users. It Quickly Became a Racist Jerk.
By DANIEL VICTOR

Microsoft set out to learn about “conversational understanding” by creating a bot designed to have automated discussions with Twitter users, mimicking the language they use.

What could go wrong?

If you guessed, “It will probably become really racist,” you’ve clearly spent time on the Internet. Less than 24 hours after the bot, @TayandYou, went online Wednesday, Microsoft halted posting from the account and deleted several of its most obscene statements.

The bot, developed by Microsoft’s technology and research and Bing teams, got major assistance in being offensive from users who egged it on. It disputed the existence of the Holocaust, referred to women and minorities with unpublishable words and advocated genocide. Several of the tweets were sent after users commanded the bot to repeat their own statements, and the bot dutifully obliged.

But Tay, as the bot was named, also seemed to learn some bad behavior on its own. According to The Guardian, it responded to a question about whether the British actor Ricky Gervais is an atheist by saying: “ricky gervais learned totalitarianism from adolf hitler, the inventor of atheism.”

Microsoft, in an emailed statement, described the machine-learning project as a social and cultural experiment.

“Unfortunately, within the first 24 hours of coming online, we became aware of a coordinated effort by some users to abuse Tay’s commenting skills to have Tay respond in inappropriate ways,” Microsoft said. “As a result, we have taken Tay offline and are making adjustments.”

On a website it created for the bot, Microsoft said the artificial intelligence project had been designed to “engage and entertain people” through “casual and playful conversation,” and that it was built through mining public data. It was targeted at 18- to 24-year-olds in the United States and was developed by a staff that included improvisational comedians.

Its Twitter bio described it as “Microsoft’s A.I. fam from the internet that’s got zero chill!” (If you don’t understand any of that, don’t worry about it.)

Most of the account’s tweets were innocuous, usually imitating common slang. When users tweeted at the account, it responded in seconds, sometimes as naturally as a human would but, in other cases, missing the mark.

@edgewerk my duck face is on fleek

— TayTweets (@TayandYou) March 23, 2016

@Prism_Root i love me i love me i love me i love everyone

— TayTweets (@TayandYou) March 24, 2016

@keganandmatt heyo? Send yo girl* a picture of what's up. (*=me lolol)

— TayTweets (@TayandYou) March 24, 2016

Tay follows a long history of attempts by humans to get machines to be our pals. In 1968, a professor at the M.I.T. taught a computer to respond conversationally in the role of a psychotherapist. Many 20- and 30-somethings have fond memories of SmarterChild, a friendly bot on AOL Instant Messenger that was always available for a chat when their friends were away.

Now, Apple’s Siri, Amazon’s Alexa, Microsoft’s Cortana, Google Now and Hound mix search-by-voice capabilities with attempted charm. The idea of a personable machine was taken to its logical ending in the 2013 movie “Her,” in which a man played by Joaquin Phoenix falls in love with the voice in his phone.

And this is not the first time automated technology has unexpectedly gotten a company in trouble. Last year, Google apologized for a flaw in Google Photos that let the application label photos of black people as “gorillas.”

微軟AI聊天機器人 上線一天就學壞

微軟公司24日宣布，已中止人工智慧機器人（AI）的「實驗」，該軟體23日才在社群網站推特（Twitter）和通訊軟體中現「聲」，原來是希望透過與人類對話，進行學習，沒想到很快就被教會了一些不合適的對答，說出充滿歧視性等不堪入耳的言論，微軟趕緊將其下架，先帶回家「管教」一番。

微軟開發的這款AI聊天機器人名為「Tay」，加入Twitter及通訊軟體GroupMe、Kik之後，隨即吸引上萬名網友與她傳訊互動。Tay的設計是透過與網友的互動，接收到大量語料後進行學習，所以她的語言能力會不斷演進。

由於Tay是設定來吸引美國千禧世代族群，因此會模仿青少年的網路用語，有即興搞笑的能力，也會用貼圖回覆訊息。

經過了一天的學習，沒想到這位AI少女的回應，竟然開始出現一些惡劣的言論，像是發文歧視黑人和女性，甚至為希特勒背書。

微軟表示，將對Tay「做一些調整」。該公司還說「Tay是機器學習方案，設定讓人類參與。它在學習的過程中，有些回應並不恰當，這正是反映出部分人士和它互動的方式為何種形式。」

原文參照：
http://www.nytimes.com/2016/03/25/technology/microsoft-created-a-twitter-bot-to-learn-from-users-it-quickly-became-a-racist-jerk.html

2016-03-26．經濟日報．A9．國際．編譯廖玉玲