图灵测试的作用、价值和意义: “ 迄今为止,没有一台计算机通过了图灵测试。 虽然如此,在尝试通过图灵测试的漫漫征程上,模拟人类思维的愿望不断激励着我们。 这一原动力对计算机科学乃至认知科学的发展,产生了深远影响。” 具体的感受,请阅读附件! 附录 : 人工智能下一步,通过图灵测试 Drink-Me 发表于 2012-04-19 19:27 http://www.guokr.com/article/155296/ 图灵是 20 世纪最伟大的数学家之一。作为现代计算机概念的缔造者他的密码破译工作在第二次世界大战中起到了决定性作用。在那个创意无限的计算机黎明时代,图灵率先提出的测试,说来似乎很简单如果一台计算机通过对话能使人们认定它是人类,那么这台计算机便被认为是具有智能的。 图灵试验的标准模式: C 使用问题来判断 A 或 B 是人类还是机械。对象为:一个具有正常思维的人(代号 B )、一个是机器(代号 A )。如果经过若干询问以后, C 不能得出实质的区别来分辨 A 与 B 的不同,则此机器 A 通过图灵试验。 在现代社会,无论是 GPS 导航系统与 Google 搜索引擎,还是自动柜员机与苹果 Siri ,更遑论象棋大师深蓝( Deep Blue )和满腹经纶的沃森( Watson ),人工智能无处不在。但是, 迄今为止,没有一台计算机通过了图灵测试。 虽然如此,在尝试通过图灵测试的漫漫征程上,模拟人类思维的愿望不断激励着我们。 这一原动力对计算机科学乃至认知科学的发展,产生了深远影响。 而现在,我们有理由相信,一台内核代码已经写就的计算机, 拥有通过图灵测试的能力 。 “两项革命性的信息技术进步,可能 将重新给被闲置已久的图灵测试 ,带来新的任务,”法国国家科学研究中心的 认知科学家 罗伯特•弗兰茨( RobertFrench )在 4 月 12 日 的《科学》杂志上撰文称“第一步是准备数量巨大的原始数据: 输入的内容 包括视频资料和完备的声音环境信息,以及随意的谈话内容和关于各种各样事物的技术文档。第二步是能够 整理、收集、处理 这些丰富数据的复杂技术。” 这有没有可能创造出 相当于人类大脑认知水平的神经连结网络 ?它能够感知到我们所感知的吗? 图灵第一次进行他的测试是在一次聚会时。他巧妙的实验令人印象深刻:参加者努力让评判者相信他们的性别是伪装的(图灵本人由于其同性恋取向受到了严酷的迫害)。那时候,这种 创建等效于人类大脑认知方式的低水平神经网络 的想法还不存在。然而复制人类的思想似乎很有可能,相比之下似乎更简单。 我们一般认为,人类的思维是逻辑性的,而计算机能够运行逻辑性的命令。因此,我们的大脑应该是可计算的。 计算机科学家 由此认为,二十年之内或许不超过十年,我们就可以看到这样激动人心的事情:人们无法根据对话分辨出对方是计算机还是人类 。 这个过分简约的构想,被证明是建立在错误的理论基础上的。认知过程要远比 20 世纪中叶的计算机科学家及心理学家所设想的复杂得多 。并且令人沮丧的是,在运用逻辑学描述我们的思想过程时科学家遇到了非常大的困难。并且 我们越来越清楚:根据人类大脑所特有的,适应快速变化的外界环境、整合信息碎片等一系列特殊功能来看,模仿人类思维几乎是无法完成的任务。 “对于现实中众多不确定性而言,符号逻辑本身过于脆弱,”斯坦福大学研究 机器智能 模拟的 计算机科学家 诺亚•古德曼( Noah Goodman) 如是说。 尽管如此现在被我们认为已经失败的传统 AI 技术上依旧颇具启发性。因为,它们彻底改变了我们对于人类大脑运作方式的看法。挫折过后,不断涌出的是许许多多极其重要的 认知科学 新观点。 直到 20 世纪 80 年代中期, 图灵测试 , 一直都是被放弃闲置的探索领域 (尽管今天它衍生出了专为 虚拟聊天机器人 设置的年度 Loebner 奖,同时, 即时虚拟广告机器人 在我们的日常生活中也益发普遍) (康奈尔大学创新机器实验室的两部 虚拟谈话机器人 正在进行非常有趣的唠嗑 ) 与此同时衍生出的是现代 认知科学和人工智能 的两个主要研究方向: 1. 推算事件发生的概率,做出准确判断。(称为概率性) 2. 在与简单、微小的程序的互动过程中,得出复杂的行为模式。(称为连结性) 和那些,像深蓝( Deep Blue )(曾因击败国际象棋大师 Garry Kasparov 扬名 ) 一样,使用“蛮力”的电脑程序的计算特点不同,人们认为这些程序至少精确反映出了人类思维中产生的某些特有现象。 迄今为止,所谓“概率性”和“连结性”两大人工智能研究新思路指导开发出了一系列现实生活中被广泛使用的人工智能产品:自动驾驶汽车, Google 搜索引擎,自动机器翻译,以及 IBM 开发的能巧妙回答任何刁钻古怪问题的 Watson 电脑。 ( IBM 的 Watson 电脑击败 KenJennings ,他是 Jeopardy! 节目的人类最高水平玩家。) 但是美中不足的是,它们在某些方面能力仍旧有限——“如果你说:‘ Watson ,给我做晚饭好不好,’或者‘ Watson ,写首十四行诗吧 , ’他会憋炸的。”古德曼这样说。但是人们不断上涨的使用(或调戏)欲望使它们的性能得以飞快进步,数据库更详实。 “你所说过的、听到的、写下的、或者是读到的每一个字,每一句话,以及每一个看到的场景,每一段经历的声音片段 , 一并同其他成百上千、甚至成千上万的人们的相关数据,都被录制下来并可随时调用。久而久之,甚至触觉以及嗅觉传感器也可以被接入以全面丰富我们这个充满图像和声音的数据库。”作为对 MIT (麻省理工)研究员戴伯•罗伊( Deb Roy )的相关研究的延伸,弗兰茨在《科学》杂志上这样设想。戴伯曾经录制了 9 万小时的视频,内容是关于他襁褓中的儿子清醒状态下的认知发展过程。 假定我们拥有可编目、分析、串联和交叉全部海量信息的处理软件以及备有上述数据库和分析系统的程序,应该完全能够使得一台计算机回答现今的 AI 们无法回答的棘手问题。这最终意味着通过图灵测试。 密歇根大学 人工智能专家 赛汀德•辛( Satinder Singh) 对数据所显示的前景充满了信心:“ 大容量数据库终会造就一台极具灵活性的人工智能机器 。” 但这样来说,梳理所有曾经学习过的问题数据就显得重要了许多。计算机要懂得:什么更值得记住,什么更值得去预测。可是,如果 你把一个孩子领进屋内,让他自由自在、随心所愿,不交给他任何任务,他为什么会自发地去做他想做的事情呢?所有的这类问题都变得异常有趣。 “ 为了变得更渊博,更灵活,更有能力,一个人必须要被动力和好奇心所驱使,从而,提炼出重要的事情, ”辛说:“这些对计算机来说,都是巨大的挑战。” “一架机器要通过图灵测试,一定要充满着人类的情感与欲望吗?。就像是弗兰肯斯坦(玛莉•雪莱( Marry Shelley )笔下的人造人),或者有生命的泥人( Golem 希伯莱传说中的用粘土、石头或青铜制成的无生命的巨人,注入魔力后可行动)一样吗?”墨西哥国立自治大学的 计算机科学家 卡洛斯•格申森( CarlosGershenson) 充满了疑问。但是,这和更基本的问题一样,难以回答。 “这做起来一定很困难,可是我们这样做的目的是什么?”他充满了疑问。 Artificial Intelligence Could Be on Brinkof Passing Turing Test BY BRANDON KEIM 04.12.12 | 5:37 PM | PERMALINK Share on Facebook Tweet In Share One hundred years after Alan Turing wasborn, his eponymous test remains an elusive benchmark for artificialintelligence. Now, for the first time in decades, it’s possible to imagine amachine making the grade. Turing was one of the 20th century’s great mathematicians,a conceptual architect of modern computing whose code-breaking played adecisive part in World War II. His test, described in a seminal dawn-of-the-computer-agepaper, was deceptively simple: If a machine could pass for human inconversation, the machine could be considered intelligent. Artificial intelligences are nowubiquitous, from GPS navigation systems and Google algorithms to automatedcustomer service and Apple’s Siri, to say nothing of Deep Blue and Watson — but no machine hasmet Turing’s standard . The quest to do so,however, and the lines of research inspired by the general challenge ofmodeling human thought, have profoundly influenced both computer and cognitivescience. There is reason tobelieve that code kernels for the firstTuring-intelligent machine have already been written. “ Tworevolutionary advances in information technology may bring the Turing test out of retirement ,” wroteRobert French, a cognitive scientist at the French National Center forScientific Research, in an Apr. 12 Science essay. “ The first is the ready availability of vastamounts of raw data — from video feeds to complete soundenvironments, and from casual conversations to technical documents on everyconceivable subject. The second is the advent of sophisticatedtechniques for collecting, organizing, and processing this rich collection ofdata. ” ' Tworevolutionary advances in information technology may bring the Turing test outof retirement .' “Is it possible to recreate something similar to the subcognitivelow-level association network that we have? That’s experiencing largely whatwe’re experiencing? Would that be so impossible?” French said. When Turing firstproposed the test — poignantly modeled on a party gamein which participants tried to fool judges about their gender; Turing wascruelly persecuted for his homosexuality – the idea of “a subcognitive low-level association network” didn’t exist.The idea of replicating human thought, however, seemed quite possible, evenrelatively easy. The human mind wasthought to be logical. Computers run logical commands.Therefore our brains should be computable. Computer scientists thought thatwithin a decade, maybe two, a person engaged in dialogue with two hiddenconversants, one computer and one human, would be unable to reliably tell themapart. That simplistic ideaproved ill-founded. Cognition is far more complicated than mid-20th centurycomputer scientists or psychologists had imagined, and logic was woefullyinsufficient in describing our thoughts. Appearinghuman turned out to be an insurmountably difficult task, drawing on previouslyunappreciated human abilities to integrate disparate pieces of information in afast-changing environment. “ Symboliclogic by itself is too brittle to account for uncertainty,” said Noah Goodman,a computer scientist at Stanford University who models intelligence inmachines. Nevertheless, “the failure of what we now call old-fashioned AI wasvery instructive. It led to changes in how we thinkabout the human mind. Many of the most important things that have happened incognitive science” emerged from these struggles, he said. By the mid-1980s, the Turing test had been largely abandoned as a research goal (though it survives today in the annual Loebner prize for realistic chatbots,and momentarily realistic advertising bots are a regular feature of onlinelife.) However, it helped spawn the two dominantthemes of modern cognition and artificial intelligence: calculatingprobabilities and producing complex behavior from the interaction of manysmall, simple processes. Unlike the so-called brute forcecomputational approaches seen in programs like Deep Blue, the computer thatfamously defeated chess champion Garry Kasparov, these are considered accuratereflections of at least some of what occurs in human thought. As of now, so-called probabilistic and connectionist approachesinform many real-world artificial intelligences: autonomous cars, Googlesearches, automated language translation, the IBM-developed Watson program thatso thoroughly dominated at Jeopardy. They remain limited in scope — “If yousay, ‘Watson, make me dinner,’ or ‘Watson, write a sonnet,’ it explodes,” saidGoodman — but raise the alluring possibility of applying them tounprecedentedly large, detailed datasets. “ Suppose, for a moment, that all thewords you have ever spoken, heard, written, or read, as well as all the visualscenes and all the sounds you have ever experienced, were recorded andaccessible, along with similar data for hundreds of thousands, even millions,of other people. Ultimately, tactile, and olfactory sensors could also be addedto complete this record of sensory experience over time,” wrote French inScience, with a nod to MIT researcher Deb Roy’s recordings of 200,000 hours ofhis infant son ’ s wakingdevelopment. He continued, “Assume also that the software exists to catalog,analyze, correlate, and cross-link everything in this sea of data. These dataand the capacity to analyze them appropriately could allow a machine to answerheretofore computer-unanswerable questions” and even pass a Turing test. Artificial intelligence expert Satinder Singh of the University of Michigan was cautiously optimistic aboutthe prospects offered by data. “Are large volumes of data going to be thesource of building a flexibly competent intelligence? Maybe they will be,” hesaid. “ But all kinds of questions thathaven’t been studied much become important at this point. What is useful toremember? What is useful to predict? If you put a kid in a room, and let himwander without any task, why does he do what he does?” Singh continued. “Allthese sorts of questions become really interesting. “ In order to be broadly and flexiblycompetent, one needs to have motivations and curiosities and drives, and figureout what is important,” he said. “These are huge challenges.” Should a machine pass the Turing test, it would fulfill a humandesire that predates the computer age, dating back to Mary Shelley’sFrankenstein or even the golems of Middle Age folklore, said computer scientistCarlos Gershenson of the National Autonomous University of Mexico. But it won’tanswer a more fundamental question. “ It will be difficult to do — butwhat is the purpose?” he said. http://www.wired.com/2012/04/turing-test-revisited/ Citation: “Dusting Off the Turing Test.” By Robert M.French. Science, Vol. 336 No. 6088, April 13, 2012. “ BeyondTuring’s Machines.” By Andrew Hodges. Science, Vol. 336 No. 6088, April 13,2012. Images: 1) The Alan Turing memorial at Bletchley Park, the site of Turing’s codebreakingaccomplishments during World War II. (Jon Callas/Flickr) 2) IBM’s Watsoncomputer defeating Ken Jennings, the highest-ranking human Jeopardy! player.(IBM) Video: Two chatbots hold a conversation in the Cornell Creative MachinesLab. (Cornell University/YouTube)