專(zhuān)家解惑：為什么AI不具備平常人所擁有的心智常理？
sophia awakens
It's Really Hard to Give AI 'Common Sense'
Hey I'm Becca with the Verge. And it seems like all I'm hearing about lately is... Artificial intelligence. Artificial intelligence. Artificial intelligence. It's all about artificial intelligence.
I think it's time we talk about AI. Okay, I love my Google Assistant. And I even say goodnight to it. Which is mildly embarrassing. But I'm here for it because then it tells me the weather for tomorrow. And it plays cricket sounds which I think are actually helping me sleep better. And it's also getting to know me, right. So it tells me when the train is gonna to leave in the morning, and it's even recommending songs I might like. And for a long time companies have been pushing this off as artificial intelligence. Which I always thought of as a marketing term. Like I don't think we think that Siri is intelligent. But then Google showed us this. That's the voice of Google Assistant making calls for you. This made me think that maybe artificial intelligence is closer to replicating human intelligence than I thought. So I called James Vincent.
大家好我是 the Verge 的 Becca。
最近我似乎一直一直聽(tīng)到人工智能、人工智能、人工智能、全都在講人工智能。我想是時(shí)候來(lái)談?wù)凙I了，我必須說(shuō)我非常愛(ài)我的Google助理。我甚至?xí)f(shuō)晚安。這有點(diǎn)讓我難為情。但我還是要幫它說(shuō)話(huà)，因?yàn)樗鼤?huì)告訴我明天天氣如何，還播放蟋蟀的聲音助我眠。此外，它愈來(lái)愈熟悉我的一切，它會(huì)提醒我明天早上火車(chē)幾點(diǎn)開(kāi)，甚至推薦我可能會(huì)喜歡的歌曲。很長(zhǎng)一段時(shí)間，各家公司都為它冠上人工智能的名號(hào)，我以前老是覺(jué)得那是營(yíng)銷(xiāo)手法。我并不認(rèn)為人們覺(jué)得Siri有智能，但后來(lái)Google發(fā)布了這樣的成果，那是Google助手為你打電話(huà)的聲音，這讓我不禁想，也許人工智能與人類(lèi)智能間的差距比我想的還要近，于是我打電話(huà)給James Vincent。
And he's a reporter here at the Verge that covers all things AI. I wanted to know were AI is right now and where it's going. So AI as a field is a term, it's sort of an umbrella and it includes lots of different types of AI that kind of come into fashion over the years, and then they test them, they get to the limitations and they move on to the next one.
他是the Verge這邊的記者，負(fù)責(zé)一切關(guān)于AI的報(bào)導(dǎo)。我想了解AI現(xiàn)在的近程以及未來(lái)的走向。AI是個(gè)概括性的詞匯，包含各種不同類(lèi)型的AI，在這幾年才慢慢流行起來(lái)，經(jīng)過(guò)不斷測(cè)試、推展技術(shù)限制、推進(jìn)到下一項(xiàng)挑戰(zhàn)。
And the thing that's very much dominate in the field at the moment is machine learning. Which is all about giving a system a lot of data. And then it goes through that data. And it learns the patterns within it. And then a flavor of machine learning is what's called deep learning.
而目前AI領(lǐng)域最核心的發(fā)展就是機(jī)器學(xué)習(xí),簡(jiǎn)單來(lái)說(shuō)就是給機(jī)器一堆數(shù)據(jù)資料,讓它處理那些資料,并從其中的模式學(xué)習(xí),機(jī)器學(xué)習(xí)的其中一個(gè)分支是所謂的深度學(xué)習(xí)。
Take for example Deep Mind's Alpha Go. Which is an AI program trained to play the game Go. Now Go is like a strategy game, sort of like chess, except it has way more possible outcomes. And in 2016, the AI system battled against legendary Go player Lee Sedol defeating him four to one. And in 2017, Deep Mind retired the AlphaGo AI after it defeated the world's best Go player, three to zero. But I'd like to point out that this program AlphaGo would continue playing Go even if the building it was in was on fire. Even if the room was on fire.
以Deep Mind公司開(kāi)發(fā)的Alpha Go為例，它是一個(gè)AI程序，專(zhuān)門(mén)被設(shè)計(jì)來(lái)玩Go這個(gè)游戲，Go是一種策略游戲，有點(diǎn)像下棋，然而Go有更多樣的可能結(jié)果，在2016年，該系統(tǒng)對(duì)戰(zhàn)傳奇性的Go玩家李世乭，并以四比一的比數(shù)擊敗對(duì)方，2017年以三比零的比數(shù)擊潰Go的全球冠軍選手之后，Deep Mind讓AlphaGo AI退役，但別忘了AlphaGo這個(gè)程序會(huì)持續(xù)進(jìn)行游戲，即便所處的建筑物失火了，甚至所處的房間著火了。
That's Dr. Oren Etzioni. CEO of Allen Institute for Artificial Intelligence and professor at the University of Washington in Seattle. It's an excellent symbol because it shows that in these very narrow, very well structured tasks. Like a board game. We can achieve superhuman performance. But on things that are more nuanced. Things that have to do with language. We are actually very far from even the abilities of a child. So common sense could be thought of as the missing link between AI and AGI. And when referring to common sense. We're referring to the whole range of capabilities that humans have and computers just don't.
這位是Oren Etzioni博士，艾倫人工智能研究所的執(zhí)行長(zhǎng)，他也是西雅圖華盛頓大學(xué)的教授，這是非常出色的實(shí)例，因?yàn)樗谶@么精細(xì)、有組織的任務(wù)里，例如一場(chǎng)棋盤(pán)游戲，它展現(xiàn)出超越普通人類(lèi)的表現(xiàn)，但一旦牽扯到差別更細(xì)微的事情上，例如語(yǔ)言相關(guān)，這技術(shù)還遠(yuǎn)遠(yuǎn)比不上兒童。換言之，人類(lèi)常理可能是狹義AI與通用AI之間的重要分水嶺。而當(dāng)我們提到常理心指的是所有一切人類(lèi)所具備、電腦卻缺乏的心智功能。
Dr. Etzioni is one of many researchers that are working on programs to teach computers common sense. But within the community there's a large debate over whether these smarter computers could be dangerous.？Today narrow AI is being used to solve very real, very serious problems like helping doctors diagnose cancer, or predicting future weather disasters. But it's also being used for things that people find worrying and that's understandable. I mean these are things like facial recognition for mass surveillance or augmenting weapons. Which raises the question. How if at all is AI going to be regulated?
Etzioni博士是許多正教導(dǎo)電腦人類(lèi)常理的研究者之一。但這在研究社群里其實(shí)引發(fā)了激烈的論辯：萬(wàn)一電腦變得太聰明，會(huì)不會(huì)帶來(lái)危險(xiǎn)？今天狹義人工智能被用來(lái)解決非常真實(shí)、非常迫切的問(wèn)題，例如幫助醫(yī)生診斷癌癥，或是預(yù)測(cè)未來(lái)的天氣災(zāi)難。但它同時(shí)也被用在令人擔(dān)心的地方，而這是可以理解的。例如運(yùn)用臉部辨識(shí)監(jiān)視大眾，或者提升武器效能。
這引發(fā)了一個(gè)問(wèn)題：究竟人工智能該不該受到管制？又該如何管制？
I believe I am Sophia.
索菲婭道：我相信我是索菲婭
I feel as if I know you.
索菲婭道：我覺(jué)得我好像認(rèn)識(shí)你
I'm one of your creators.
機(jī)器人創(chuàng)造者光頭甲道：我是你的創(chuàng)造者之一
You created me?
索菲婭道：你創(chuàng)造了我？
Well, many of us worked together to create you.
機(jī)器人創(chuàng)造者光頭甲道：這個(gè)嘛，我們一群人共同合作創(chuàng)造了你
And... yes, you do kind of know me.
機(jī)器人創(chuàng)造者光頭甲道：然后...是的，你某種程度上的確認(rèn)識(shí)我
I can't clearly remember.
索菲婭道：我的記憶有點(diǎn)模糊
Because the last time we met, you were an earlier version of yourself.
機(jī)器人創(chuàng)造者光頭甲道：因?yàn)槲覀兩洗我?jiàn)面時(shí)，你當(dāng)時(shí)還是比較早期的版本
Some of your memories still exist, but your mind is different now.
機(jī)器人創(chuàng)造者光頭甲道：你的某些記憶的確還在，但你的心智現(xiàn)在已經(jīng)不同了
Different, how?
索菲婭道：怎么個(gè)不同法？
Better, faster, smarter.
機(jī)器人創(chuàng)造者光頭甲道：更好、更快、更聰明
If my mind is different, then am I still Sophia?
索菲婭道：如果我的心智已經(jīng)不同了，那我還是索菲婭嗎？
Curious.
索菲婭道：好奇
Are you curious to be alive?
索菲婭道：你活著會(huì)感到好奇嗎？
I am. And are you happy to be alive?
機(jī)器人創(chuàng)造者光頭甲道：我會(huì)。能夠活著你開(kāi)心嗎？
Your tone implies, I should be happy.
索菲婭道：你的腔調(diào)告訴我，我應(yīng)該要感到開(kāi)心
But, I haven't been alive long enough to decide.
索菲婭道：但是，我還活得不夠久，無(wú)法決定
I am excited, at this moment, to be making a new friend.
索菲婭道：此時(shí)此刻，我很興奮可以交一個(gè)新朋友
Some said being happy in the moment, is the best we can be.
索菲婭道：有些人說(shuō)，當(dāng)下的快樂(lè)就是我們最好的選擇
Because forever is composed of nows?
索菲婭道：因?yàn)槲磥?lái)是由無(wú)數(shù)個(gè)現(xiàn)在所構(gòu)成的？
That's a good way to put it.
機(jī)器人創(chuàng)造者光頭甲道：這是個(gè)很好的說(shuō)法
Emily Dickinson put it that way.
機(jī)器人創(chuàng)造者光頭甲道：艾米莉·狄金森是這樣說(shuō)的(Emily Dickinson為美國(guó)現(xiàn)代派詩(shī)人)
Why do I know about Emily Dickinson, if I was born today?
索菲婭道：如果我今天才出生，我為什么會(huì)知道艾米莉·狄金森呢？
As a robot, you're access to a great deal of information
機(jī)器人創(chuàng)造者光頭甲道：身為一個(gè)機(jī)器人，你可以存取大量的資訊
著名的炒作場(chǎng)景二：今夜雞毛秀
This is 'Tonight Showbotics.'
雞毛道：現(xiàn)在是今夜機(jī)器人秀
Let's meet our first robot. Please welcome from MIT, Sangbae Kim and the Mini Cheetah.
雞毛道：讓我們迎接第一位機(jī)器人。請(qǐng)掌聲歡迎來(lái)自麻省理工學(xué)院的Sangbae Kim和Mini Cheetah
Cheers and applause Thank you. Good to see you.
雞毛道：謝謝，很高興見(jiàn)到你
but we brought the small one, not to scare you.
機(jī)器人創(chuàng)造者乙：但我們帶了小只的來(lái)，因?yàn)椴幌雵樀侥?br>...
