ARTICLE · 1392588
11.21 大学英语四六级阅读题源外刊模拟试题【难度7.0】
Section C
Directions: There are 2 passages in this section. Each passage is followed by some questions or unfinished statements. For each of them there are four choices marked A, B, C and D. You should decide on the best choice and mark the corresponding letter on Answer Sheet 2 with a single line through the centre.
Passage One
Questions 46 to 50 are based on the following passage.
How should one respond to a problem so overwhelming that it is hard even to know where to begin? That was the situation facing DQN, an artificial intelligence system from Google DeepMind that in 2013 was challenged to learn classic video games without being told the rules. Its only guide was the score, and through random moves it gradually discovered which actions made the number rise.
The method worked brilliantly, and at several games DQN soon surpassed the best human players, sometimes by more than ten times. Yet one game, Montezuma's Revenge, defeated it completely. Because the game awards no points until a long, complicated sequence of actions is complete, and the system learnt only by linking its actions to rises in the score, even ten correct moves gave it no sign of progress.
The obvious remedies, however, proved useless. Rewarding the character for staying alive merely taught the system to remain motionless, while offering extra points for intermediate targets, such as dodging a rolling skull, risked producing a player that dodged skulls for ever. Moreover, such supplementary rewards work for one game only, whereas the real aim was a machine able to master unfamiliar problems of every kind.
A better idea was to reward the system whenever it made something new and surprising happen. Driven by curiosity, it began to explore its surroundings, which usually earned points eventually and arguably served the real purpose of the game, and it soon fought its way through most of Montezuma's Revenge. Novelty-seeking behaviour is no cure-all, but in many circumstances it proves remarkably powerful.
It can also backfire: researchers have found that a curious system can be trapped by a screen showing an endless stream of fresh images, its appetite for novelty being satisfied so fully that it never moves on. Humans are not immune either, as slot machines and social media demonstrate.
Still, a taste for variety may be productive. Bernice Eiduson, a psychologist, set out in the 1950s to follow 40 promising scientists, and colleagues later analysed her data in search of clues to who would enjoy long and productive careers. Psychological tests and interviews told them little, but the scientists' early publications proved revealing: the most productive of them had moved between five areas on average, switching 43 times within their first hundred papers.
Variety may help for two reasons. Techniques learnt in one field may find unexpected applications in another, and dwelling on a single problem makes it easy to get stuck, as a wrong quiz answer stays in the mind and blocks the right one. The remedy is often to walk away and try something quite different. It worked for DQN; perhaps it can work for us.
46. Why did DQN fail so completely at the game Montezuma's Revenge?
A) It had not been told the rules of the game.
B) It was unable to judge whether it was getting anywhere.
C) It could not carry out such a long series of actions.
D) The game was too complicated for any machine to master.
47. According to the author, what was the real goal of those developing DQN?
A) To have it outplay the best human players at every game.
B) To make it explore the game world as fully as possible.
C) To design a separate reward system for each of the games.
D) To create a machine that can cope with new challenges.
48. What does the author think of novelty-seeking behaviour as a way of solving problems?
A) It is effective in many cases but has its drawbacks.
B) It guarantees success wherever the score gives no guidance.
C) It suits machines much better than it suits human beings.
D) It is harmful to machines but helpful to human beings.
49. What do we learn about the most productive scientists in Eiduson's study?
A) They stood out in psychological tests and interviews.
B) They began publishing earlier than other scientists did.
C) They repeatedly shifted from one research area to another.
D) They stuck with one research topic until it was exhausted.
50. Why does the author mention a wrong answer to a quiz question?
A) To show that people soon forget the mistakes they have made.
B) To show how sticking to one problem can hold us back.
C) To show that skills learnt in one field can serve another.
D) To warn that quick answers are more often wrong than right.
附注:根据历年英语六级真题阅读题源外刊等,摘选最新文章,模拟仿真出题。
参考答案见以下。
Quick look: BDACB
46.【正确答案】B
【解析】题型: 因果细节题
定位: 第二段第二、三句 “Yet one game, Montezuma's Revenge, defeated it completely. Because the game awards no points until a long, complicated sequence of actions is complete, and the system learnt only by linking its actions to rises in the score, even ten correct moves gave it no sign of progress.”(然而有一款游戏,《蒙特祖玛的复仇》,让它一败涂地。由于这款游戏要等到一长串复杂的动作全部完成才给分,而系统只靠把自己的动作与得分的上升联系起来进行学习,因此即使它连着做对了十步,也看不到任何进展的迹象。)
分析: 题干问 DQN 为何在 Montezuma's Revenge 上败得如此彻底,对应第二句的 defeated it completely;原因由第三句 Because 引导的从句给出:游戏在一长串动作完成之前一分不给,而 DQN 只能靠“动作与得分上升的关联”来学习,于是哪怕前十步都做对,得分也毫无变化,系统无从知道自己做得对不对。B 项 It was unable to judge whether it was getting anywhere 是对 gave it no sign of progress 的转述(unable to judge whether … getting anywhere 对应 no sign of progress)。本题需要把 Because 之后的两个条件(延迟给分、只靠得分学习)合起来,推出“无法判断自己是否在进步”这一结果。
干扰项: [A] 它没有被告知游戏规则,真实但无关:第一段确实说 DQN 学习游戏时 without being told the rules,但这是它面对所有游戏的共同条件,而它在其他游戏上表现出色(第二段首句),因此不能解释它为何偏偏败在这一款游戏上;[C] 它无法完成这样长的一系列动作,偷换概念:原文说的是游戏要等一长串动作完成后才给分(awards no points until … is complete),症结在于得不到反馈,并没有说 DQN 做不出这些动作;[D] 这款游戏对任何机器来说都太复杂、无法掌握,与原文相反:第四段指出,好奇心驱动的系统很快就闯过了这款游戏的大部分内容(it soon fought its way through most of Montezuma's Revenge),可见并非“任何机器都无法掌握”,any machine 过于绝对。
47.【正确答案】D
【解析】题型: 事实细节题
定位: 第三段末句 “Moreover, such supplementary rewards work for one game only, whereas the real aim was a machine able to master unfamiliar problems of every kind.”(而且,这类附加奖励只对一款游戏有效,而真正的目标是造出一台能够掌握各种陌生问题的机器。)
分析: 题干中的 real goal 对应末句的 the real aim。末句用 whereas 作对照:前半句是只对一款游戏有效的附加奖励,后半句才是研发者真正想要的——“能够掌握各类陌生问题的机器”。D 项 To create a machine that can cope with new challenges 是对 a machine able to master unfamiliar problems of every kind 的改写(cope with 对应 master,new challenges 对应 unfamiliar problems)。本题的难点在于前文出现了好几个看似“目标”的内容(超越人类玩家、让系统探索游戏世界、设计奖励机制),需要抓住 real aim 这一信号词,区分“手段、阶段性成果”与“真正的目的”。
干扰项: [A] 让它在每一款游戏上都胜过最优秀的人类玩家,扩大范围兼偷换概念:第二段只说 DQN 在“若干款(several)”游戏上超过了人类玩家,这是方法奏效的表现,并非作者所说的 real aim,而且它在 Montezuma's Revenge 上一败涂地,at every game 过于绝对;[B] 让它尽可能充分地探索游戏世界,把手段当目的:第四段中,探索是“好奇心奖励”带来的行为,是攻克 Montezuma's Revenge 的办法,不是研发者的最终目标;[C] 为每款游戏设计单独的奖励机制,与原文相反:第三段末句正是在指出这类附加奖励只对一款游戏有效(work for one game only),真正的目标恰恰要超越这种做法。
48.【正确答案】A
【解析】题型: 观点态度题
定位: 第四段末句 “Novelty-seeking behaviour is no cure-all, but in many circumstances it proves remarkably powerful.”(寻求新奇的行为并非万灵药,但在许多情况下它确实相当有效。);第五段首句 “It can also backfire: researchers have found that a curious system can be trapped by a screen showing an endless stream of fresh images…”(它也可能适得其反:研究人员发现,好奇的系统可能被一块不断播放新图像的屏幕困住……)
分析: 题干问作者如何看待“寻求新奇的行为”作为解决问题的方式。第四段末句先给出评价:no cure-all, but … remarkably powerful,即并非万能,但在许多情况下相当有效;第五段首句紧接着用 also backfire 补充它的另一面,并以系统被源源不断的新图像“困住”为例。两处合起来,作者的态度是褒贬兼顾:多数情况下有效,但也有弊端。A 项 It is effective in many cases but has its drawbacks 正是这一态度的概括(effective in many cases 对应 in many circumstances … powerful,has its drawbacks 对应 no cure-all / backfire)。
干扰项: [B] 只要得分毫无提示,它就能保证成功,过于绝对:第四段末句明确说 no cure-all,第五段还指出它可能适得其反,guarantees success 与之相反;[C] 它对机器比对人类更合适,无中生有:原文没有比较新奇追求对机器和人类“谁更适用”,第五段末句反而说人类同样不能幸免(Humans are not immune either);[D] 它对机器有害而对人类有益,张冠李戴兼无中生有:选项把第五段对机器的“适得其反”与第六段对科学家的“多样性可能有益”拼成机器与人类的对立,而原文并无此对比,第五段还说人类同样会被新奇事物困住。
49.【正确答案】C
【解析】题型: 事实细节题
定位: 第六段末句 “Psychological tests and interviews told them little, but the scientists' early publications proved revealing: the most productive of them had moved between five areas on average, switching 43 times within their first hundred papers.”(心理测试和访谈几乎没有提供什么线索,但这些科学家早期发表的论文却很能说明问题:其中最高产的人平均在五个领域之间转换,在最初的一百篇论文里转换了 43 次。)
分析: 题干问 Eiduson 研究中最高产的科学家有什么特点,信息在第六段末句冒号之后:the most productive of them had moved between five areas on average, switching 43 times …,即最高产者在不同研究领域之间频繁转换。C 项 They repeatedly shifted from one research area to another 是对 moved between five areas … switching 43 times 的概括(repeatedly 对应 43 times,shifted from one research area to another 对应 moved between … areas / switching)。冒号之前只是交代心理测试和访谈没有提供有用线索,属于排除性信息,不是答案。
干扰项: [A] 他们在心理测试和访谈中表现突出,与原文相反:原文说 Psychological tests and interviews told them little,即这两项并没有提供预测线索;[B] 他们比其他科学家更早开始发表论文,偷换概念:early publications 指的是研究对象“早期发表的论文”,说明的是研究方向的转换情况,原文没有比较谁开始发表得更早;[D] 他们固守一个课题直到把它做尽,与原文相反:最高产的科学家恰恰是频繁转换领域(switching 43 times),第七段也指出长时间纠缠于单一问题容易陷入僵局(dwelling on a single problem makes it easy to get stuck)。
50.【正确答案】B
【解析】题型: 写作意图题(例证作用)
定位: 第七段第二句 “Techniques learnt in one field may find unexpected applications in another, and dwelling on a single problem makes it easy to get stuck, as a wrong quiz answer stays in the mind and blocks the right one.”(在一个领域学到的技巧可能在另一个领域找到意想不到的用途,而长时间纠缠于单一问题则容易陷入僵局,就像一个错误的测验答案会留在脑海里、挡住正确答案一样。)
分析: 第七段首句交代多样性有益的“两个原因”,第二句用 and 把它们并列:其一,技巧可以跨领域迁移;其二,死盯一个问题容易卡住。quiz 的例子由 as 引出,位于第二个原因之后,作用是说明这第二个原因——错误答案占据头脑、挡住正确答案,正是“纠缠于单一问题会让人卡住”的具体表现。B 项 To show how sticking to one problem can hold us back 对应 dwelling on a single problem makes it easy to get stuck(sticking to 对应 dwelling on,hold us back 对应 get stuck)。本题的关键是分清 and 前后的两个原因,并把例子归给后者。
干扰项: [A] 说明人们很快会忘记自己犯过的错误,与原文相反:原文说错误答案 stays in the mind,会一直留在脑海中、挡住正确答案,而不是很快被忘记;[C] 说明在一个领域学到的技能可以用于另一个领域,张冠李戴:这是第二句前半句的第一个原因(Techniques learnt in one field …),而例子由 as 引出,说明的是 and 之后的第二个原因,二者不能混为一谈;[D] 提醒人们不要答题过快,无中生有:原文只是借答题时的体验作类比,没有谈到回答的快慢,也没有给出这样的告诫。
【参考译文】
面对一个极其棘手、让人不知从何下手的问题,该如何应对?谷歌DeepMind的人工智能系统DQN就曾面临这样的挑战:2013年,它被要求在不被告知游戏规则的情况下,学会玩各种经典电子游戏。它唯一的指引就是分数;通过随机尝试各种操作,它逐渐摸索出哪些动作能让分数增长。
这种方法效果极佳,DQN很快就在多款游戏中超越了人类顶尖玩家,有的甚至领先了十倍以上。然而,有一款名为《蒙特祖玛的复仇》(Montezuma's Revenge)的游戏却让它彻底束手无策。在这款游戏中,玩家必须完成一连串漫长而复杂的操作才能得分;由于该系统只能通过将“动作”与“分数增长”建立关联来学习,即便它连续做对了十步,也无法获得任何“取得进展”的信号。
然而,一些显而易见的补救措施却毫无用处。如果奖励角色“保持存活”,系统就会学会一动不动地待在原地;如果为达成中间目标(比如躲避滚动的骷髅头)设置额外奖励,则可能导致系统沉溺于“永远在躲避骷髅头”的行为中。此外,这类辅助性奖励往往只适用于特定游戏,而我们的真正目标是打造出一种能够掌握各种陌生问题的通用智能系统。
一个更好的思路是:每当系统引发了某种“新颖且令人惊奇”的结果时,就给予它奖励。在“好奇心”的驱动下,它开始探索周围环境——这种探索往往最终能带来分数,也确实契合了游戏的核心目标——很快,它就成功通关了《蒙特祖玛的复仇》的大部分内容。这种“寻求新奇”的行为虽非万能,但在许多情况下却展现出了惊人的效力。
当然,它也可能适得其反:研究人员发现,一个充满好奇心的系统可能会被困在某个画面前——如果屏幕上不断涌现出新鲜的图像,系统对“新奇感”的需求就会得到极大的满足,以至于它再也不会去探索其他地方了。人类也未能幸免,老虎机和社交媒体就是典型的例子。
尽管如此,对多样性的追求依然可能带来积极的成果。20世纪50年代,心理学家伯尼斯·艾杜森(Bernice Eiduson)开始对40位前途无量的科学家进行长期追踪研究;后来,她的同事们对这些数据进行了分析,试图从中寻找线索,探究究竟什么样的人能够拥有漫长且富有成效的职业生涯。心理测试和访谈提供的信息有限,但科学家们早期的学术发表记录却揭示了一些重要线索:其中科研产出最高的人,平均涉足过五个不同的研究领域,在他们发表的前一百篇论文中,研究方向累计转换了43次。
这种“多样性”之所以有益,原因有二:首先,在一个领域习得的技术,可能会在另一个领域产生意想不到的应用;其次,如果死磕同一个问题,人很容易陷入思维僵局——就像做题时错误的答案会占据脑海,从而阻碍你想到正确的答案一样。解决之道往往是暂时放下手头的工作,转而去尝试完全不同的事物。这一策略对DQN(深度Q网络)有效,或许对我们人类也同样适用。















































































