登录以后才能看到帖子详情哦！
您需要 登录 才可以下载或查看，没有账号？立即注册
×
Teaching Robots To Be Moral 教导机器人明辨是非
A scene from “Chappie.” 《超能查派》的一幕
COLUMBIA PICTURES/EVERETT
“Chappie,” the highest-grossing movie in Americalast weekend, is, to put it mildly, not a great film; the critics have given ita twenty-nine on Rotten Tomatoes, and it is nowhere near as original as“District 9,” an earlier effort by the director, Neill Blomkamp. “Chappie” doesnot have the philosophical depth of “The Matrix” or the remade“BattlestarGalactica” series. Nor does it have the visual panache of “Interstellar” or “2001.” From its opening scene, the filmcomes across as little more than a warmed-over “RoboCop” remake, relocated toJohannesburg. There’s an evil company man, droids that menace the population,and a whole lot of blood, shooting, and broken glass.About the only thing that seems new is the intermittently adorable androidprotagonist, Chappie. 客气地说，上周美国票房最高的电影《超能查派》不是一部太好的影片；评论家们在烂番茄影评网上给出29分，一点都不如导演尼尔·布洛姆坎普早期的影片《第九区》。《超能查派》并没有《黑客帝国》和《银河战星》重制版的哲学深度，也没有《星际穿越》和《2001太空漫游》那样炫目的视觉效果。在首映礼上，影片仅仅给人翻炒“重置版《机械战警》“冷饭的印象，只是把情景搬到了约翰内斯堡。同样地，有一个邪恶的公司职员、几个对居民造成威胁的机器人、大量的鲜血、枪击场景和破碎的玻璃。唯一新鲜的点子是间接性卖萌的主角机器人查派。
Meanwhile, most of the science fiction in themovie utterly fails as science. The plot, for example, revolves around aportable scanner that reads the contents of both human and robot brains. Itmakes sense that someone would be excited to study a human’s brain in a magicelectronic skullcap. But if you were analyzing a robot, wouldn’t be it easierjust to download the software? SpikeJonze’s “Her”presented a future that seemed utterlybelievable; Blomkamp’s film presents a mishmash. And yet, as bad as most of itis, I walked away with a newfound respect for the need for science fiction in aworld of rapidly changing technologies. Even bad films can raise profoundquestions. 同时，影片里大部分的科幻元素一点都不科幻。比如说，情节围绕着一台可以读出人类和机器人思想的可移动扫描仪而展开。如果你能凭一顶神奇的电子便帽就能读出人类的思想，那肯定是让人超级兴奋的一件事。而如果你是在分析一台机器人，下载软件不就容易多了吗？斯派克·琼斯在《她·云端情人》中展示的未来就显得非常可信，而布鲁姆坎普的影片描述的未来场景却是混乱一团。可最糟糕的是，我发现在一个科技快速进步的世界里，科幻小说有其存在的必要性，所以我不得不怀着敬意走开。即使烂片也能引出深邃的问题。
The interesting partsof the film stem almost entirely from Chappie. Whereas most science-fiction androids enter thescene as a fully functioning “adults,” Chappie enters as a virtual newborn.Initially, it doesn’t know a word of English, nor even the most basic factsabout the world. 影片中的闪光点基本都来自于查派。尽管大部分科幻片里的机器人都是以全能“成年人”形象进入镜头，查派却是从一个虚拟的新生儿开始成长的。刚开始它连一个英文单词都不会，对世界上最基础的事情也一无所知。
“Chappie” takes its inspiration from a real (ifsmall) field of artificial intelligence known as developmental robotics, inwhich simple robots like iCublearn by doing. For now, that field has producedfew results of any significance. Most robots today are largely preprogrammed;Sony’s late AIBO robot came out of the factory walking, and the only thing thatyour top-of-the-line vacuuming robot ever learns is the layout of your home.The adorable commercial robot Baxter does better. It can learn new tasks as itworks, such as how to pick up objects of certain sizes and move them toparticular places. But, at least for now, Baxter’s learning abilities mainlyconsist of variations on themes; a certain basic set of skills is installed inthe factory, and it seems doubtful that Baxter could ever learn on its own todo something novel, like knit or juggle. 《超能查派》的灵感来自于被称为“自主心智发育机器人“的人工智能领域（可能很小），简单的机器人能像iCub那样在实践中学习。眼下这一领域还没有显著的成绩。今天大多数机器人还是程序预设的；索尼最后的AIBO机器狗淡出工厂流水线，即使说起你家最高端的吸尘机器人能认知的也就只有家里的布局了。最受欢迎的商业机器人Baxter的学习能力强一些。它能在工作中学习新任务，比如”如何拾取既定体积的物件并放到特定的位置上”。但至少到目前，Baxter的学习能力主要由不同的模式构成；一系列既定的技能都在工厂里安装好，因此Baxter是否可以自己学习像织毛衣或变戏法这些新奇的事情不得不让人怀疑。
Chappie is thus every roboticist’s dream—acombination of hardware and software that can learn pretty much anything. In asingle week, Chappie grows from infant to toddler to surly teen-ager. Itmasters everything from the English language to the fine art of throwing knivesat moving targets. The film never tells us how Chappie learns so fast, butpretty much every software engineer and developmental psychologist I know wouldlove a machine that could match Chappie’s skills. Ultimately, though, the movieis about something different: not the cognitive scientist’s question about howintelligent creatures manage to learn about the world but the educator’squestion about how human beings can raise moral robots. Chappie doesn’t justlearn a set of facts; Chappie learns a set of values. 查派是每个机器人专家的梦想——硬件和软件的结合能让机器人学会所有的事情。在一个星期内，查派从一个婴儿长成幼儿、再长成脾气粗暴的青少年。它掌握了从英语到精确瞄准移动的目标扔飞刀的所有事情。影片从没告诉我们为何查派学得这么快，但所有我认识的软件工程师和发展心理学家会爱死一台能与查派的学习技巧匹敌的机器。当然影片最后还是有亮点的：不是认知学家对智能生物如何学会认知世界的疑问，而是教育家对于人类如何教导机器人明辨是非。查派不仅学到了一系列的事情，还学到了一套价值观。
As any parent knows, teaching values is hard.Early in the movie, Chappie’s creator tries valiantly to teach the robot thedifference between right and wrong, but even the most basic lessons fall flat;the robot understands that killing people is wrong, but is seduced by a ratherless savory character into believing that merely cutting people—to “help themgo to sleep”—is acceptable. Children often learn more from their peers thantheir parents, within minutes of his activation, Chappie, surrounded by a badcrowd, is unable to sort right from wrong. 父母们都知道，给孩子传递价值观是一件困难的事情。在影片开头，查派的制造者勇敢地尝试教它分辨对错，可是即使最基础的课程也完全无效；查派理解杀人是错的，但是在一个不那么正派的人物唆使下，它相信仅仅把人砍开——以“帮助他们入睡“——是可以接受的。小孩在与同龄小伙伴玩耍时学到的通常比父母的言传身教多。查派被激活的时候，周围是一帮坏家伙，它并没有能力分辨对错。
None of us should want that. In a future worldever more populated with robots, we’ll want them—whether driving cars or takingcare of the elderly—to have some sort of moral compass. Elon Musk recently warned that artificial intelligence is thegreatest existential threat to mankind. He may be overstating things, but heisn’t wrong to be concerned that, in building robots and artificialintelligence, we could potentially unleash a demon. We do, as I argued here, in2012, need to learn to buildmoral machines. Robots and advanced A.I. could truly transformthe world for the better—helping to cure cancer, reduce hunger, slow climatechange, and give all of us more leisure time. But they could also make thingsvastly worse, starting with the displacement of jobs and then growing intosomething closer to what we see in dystopian films. When we think about ourfuture, it is vital that we try to understand how to make robots a force forgood rather than evil. 我们都不想那样。在未来会有很多机器人，我们需要它们有道德的指南针——不论是驾车还是在照顾老年人上。最近埃隆·马斯克警告说，人工智能是人类最大的威胁。他也许言过其实，但他的担忧并非无理，我们制造机器人和人工智能，有可能就是在制造着恶魔。像我之前在2012年争辩所言，我们需要制造有道德观的机器。机器人和高端人工智能确实会让世界变得更好——帮助治愈癌症、减少饥饿、减缓气候变化，让我们所有人有更多的闲暇。但是它们也会把事情变得很糟糕，首先是人类的工作被取代，然后整个社会慢慢发展成类似于我们所看的反乌托邦电影。设想一下未来，我们要学会让机器人成为有益的工具而非恶魔是多么的重要。
“Chappie” isn’t, of course, the first bit of science fiction to pointthat out. But thefilm raises the question in a different way. In Isaac Asimov’s imagination,robots came straight from the factory programmed to obey three laws, startingwith “A robot may not injure a human being or, through inaction, allow anotherhuman being to come to harm.” In Blomkamp’s imagination, robots don’t come fromthe factory with any laws; they learn from their makers, companions, and whatthey see in the world around them. 当然，《超能查派》不是第一部指出这个问题的科幻影片，但它从另外一个角度提了出来。在艾萨克·阿西莫夫的想象中，机器人出厂时就被程序化要遵守三条法则，第一条就是“一个机器人不得伤害人类，也不能坐视人类受到伤害。“但是在布洛姆坎普的想象中，机器人出厂时没有安装任何准则；它们从制造者、伙伴和它们所看到的周遭的世界学习。
Asimov’s factory-installed laws seem fine for a short story,but inadequate for the real world. Even if they were the right laws, wecertainly don’t yet know how to turn them into computer code. How, for example,do you even translate the concept of harm into the language of zeroes and ones?But, and this is Blomkamp’s point, learning morality is fraught with problems,too. Human children learn their values in at least two ways—through explicitinstruction (“Stealing is wrong”) and through observation (“What do my parentsdo? What are my friends getting away with?”). As the Yale psychologist PaulBloom has argued, we also seem to be born with the innate underpinnings of amoral sense. Future robots will likely be similar. Their decisions willinevitably be guided by a mixture of what is preprogrammed and what they learnthrough observation. How will we make that work in a way that providescertainty about our own safety? Robots may someday serve as police officers andbe pressed into deciding whom to protect and arrest; even in the home, dilemmasmay arise. (What happens if an intruder breaks in and threatens the robot’sowner?) In its best moments, “Chappie” can be seen an impassioned plea formoral education, not just for humans but for our future silicon-basedcompanions. 阿西莫夫的工厂程序化法则对于一个短篇故事似乎足够了，可对于一个真实的世界是不够的。即使那些法则是正确的，我们都不知道怎样把它们转换成计算机代码。例如，你怎样把伤害的概念转换成程序语言0和1呢？但这也是布洛姆坎普的观点，学习道德法则问题多多。人类的儿童至少通过两种途径习得价值观——一是通过明确的指引（“偷窃是不对的”），二是通过观察（“我的父母是怎样做的？我的小伙伴们是怎样做的？”）。像耶鲁大学心理学家保尔·布鲁姆所提出的，我们也似乎有一种与生俱来的道德感。未来机器人很可能也会这样。他们的形式不可避免地收到预设程序以及通过资深观察学习到的知识的混合体所指引。我们如何让它们以一种为人类自身安全提供确然保障的方式运作？也许有一天机器人会充当警官并得决定保护谁和逮捕谁；即使是在家里，也有两难的场面发生。（如果一个不速之客强行进入而且威胁到机器人的主人呢？）《超能查派》也有其优势面，它是对于道德教育激情洋溢的呼吁，不仅仅是为我们人类自身，也为我们将来的硅基伙伴。
How can we keep ourselves safe ina world in which we will be surrounded by autonomous steel contraptions thatmay someday be as smart as us, or even smarter? Blomkamp offers no real answers. That’s fine, and is hisprerogative as a filmmaker. But society needs to think about these questionssooner rather than later. Contemporary robots are neither as dexterous asChappie nor as quick to learn, but that’s just for now. No one knows what’scoming next. 在周遭都是自主钢铁机器的世界里，它们有朝一日会像我们人类一样聪明、甚至更加聪明，我们如何保证自己的安全呢？布鲁姆坎普没有给出真正的答案。好吧，那是他作为制片人的特权。但是我们的社会需要思考这些问题，越早越好。现代机器人既不如查派灵巧，也不像它那样有快速学习的能力，但仅眼下而已。没有人知道将来怎样。
翻译by山 校对by gabriellaz 终校by 小郭 树屋字幕组-文翻组 翻译仅供学习交流，严禁用于商业用途 PDF文档下载：
| 
发表于 2015-9-17 14:09:35
