天上的小石子

飞吧！飞吧！飞越一切！我在梦中飞起来了……

飞过高架桥的视野

飞过城市的风景线

飞过皑皑的雪峰

飞过广袤的荒原

飞过一切阻隔

我们在山间相见

千丘万壑在我胸

山高水长可相逢

FreeBSD对我的Compaq Evo N800v的ACPI支持一直不是很好，可是没想到最近出现了惊心动魄的问题。不知道是不是同连续几个昼夜没有关机有关，那天上午醒来发现硬盘灯狂闪，机器停在屏保（就是那个著名的xmatrix）死机了。风扇也在声嘶力竭地狂转，以至于我都怀疑是不是它持续发出的噪音让我醒来的。也没怎么在意，拔电源重启了。哪里想到重启后那个鲜红的“COMPAQ”刚刚一闪，硬盘就发出了“啪”的一声，自动关机了。又试了一次还是如此，心头一阵慌乱。一摸机器底部，烫得灼手！莫非是开机时间过长主板和CPU烧了？以前听说过这种事情，但是很匪夷的是我的第一个念头是终于可以把这台用了三年的机器换掉了！然后才担心起里面的数据和可能对我造成的种种混乱……

让机器凉了一阵子，底下已经不烫了。再开机，硬盘发出了一声以前没有听到过的诡异的呻吟，然后成功启动。可是在mount硬盘的时候出现了/root不能读写的情况。用fsck扫描发现数个逻辑错误和貌似坏道的错误，数次试图修复未果。重新启动还是出现类似错误。于是想到了大概是FreeBSD的分区出现了坏道？重启能够顺利进入Windows，但还是不能保证那里就没有坏道。放在以前我对这种问题是不能容忍的，抱着死马当活马医的心态，让机器又凉了一阵子，进入FreeBSD。虽然还是出现mount的读写错误，经过fsck后竟然全修好了。如果是坏道的话，断没有自动修复的可能，所以只能归结于硬盘还没有彻底凉下来……

我这个硬盘不是原配的，是后来买的5400转（原本是一块4200转的），看样子吃不消我昼夜连续的折磨暂时罢工了。看样子COMPAQ的东西还是很牢啊，要是换过爆炸过两回的Dell话（看这里和这里），怕是小命也要难保。

update: 恐怖的Dell炸弹第三弹在新加坡现身，看这里。

一年中日照最多、气温最高的时期，却也是盛夏的最后一个节气。今年八月七日立秋。

今日本地天气（来自BBC）：

日出：05:00 (BST)

日落：21:39 (BST)

昼长：16小时39分钟

夜长：7小时21分钟

温度：14-24摄氏度

天气：晴间多云

气压：1015毫巴

能见度：中等

相对湿度：46%

今日黄历（来自新浪星座算命）：

丙戌年 乙未月 癸丑日（六月廿八日）

宜：求医 治病 破屋 坏垣 馀事勿取

讳：嫁娶 出行

冲煞：冲羊、煞东

总结：盛夏的高峰，三伏之中伏。北半球共热，地湿上蒸，天势下迫，交蒸之气乱于肠胃之间，机体难免内外不和。故需调理自身，安然度夏。行事莫贪，点到为止。大热天，呆在家里便好，出行恐有祸事。属羊的同学莫向东行，亦有祸事。

Compiled from Wired News and OSNews discussions:

Artificial intelligence is 50 years old this summer, and while computers can beat the world’s best chess players, we still can’t get them to think like a 4-year-old.

When the term “artificial intelligence” was coined at a Dartmouth workshop, the idea was to explore human-level intelligence as computation. But it was the term itself that quickly captured the public’s imagination. The new buzzword engrained itself in popular culture, while scientists set for themselves the ambitious goal of creating a computer that would ace the Turing test, i.e., pass as a human in prolonged conversation with one. Expectations were high: The same year the Turing test was proposed, Isaac Asimov wrote the book I, Robot, which depicted a world where machines had achieved the ability to interact seamlessly and intelligently with humans.

Today, AI still hasn’t produced a robot with enough common sense to describe what’s happening in a photograph, let alone hold a conversation. But in 1997 an IBM supercomputer Deep Blue defeated the world’s chess champion Garry Kasparov for the first time. The victory over Kasparov represented a shift in the field of AI from trying to replicate general human intelligence to perfecting deep expertise in specific fields. Today these types of programs have become pervasive in the modern economy - by banks to police transactions for fraud, by cell phone companies for voice recognition, and by search engines to scour the web and organize data. Beyond business, programs like AI in Medicine help doctors diagnose and treat patients, while vision-recognition programs scan beaches and pools and alert lifeguards to signs of drowning. Most science fields today depend on some form of AI. Take biology, which uses programs to make sense of the large sets of data reaped from mapping the human genome. The tools of AI are now powerful enough to become a scientist’s apprentice and help us understand data sets, whereas we are coming to a time where we are gathering large amounts of data, like on the web, but the bottleneck is analysis, not getting the data.

Most recently, AI rose to meet DARPA’s Grand Challenge of creating a robot car that could drive itself along a desert road to a specific destination. The success of this challenge envisions a future where cars will drive themselves, eliminating crashes and freeing up their passengers to pursue more productive activities than road rage. So while AI hasn’t put a robot in every household, the field has made strides, which makes room for speculation about the future. If the rate of computational power grows exponentially, the possibilities of true AI could be possible very soon - a world where humans and machines have merged, enhancing our cognitive abilities and keeping our bodies healthy from the inside.

In fact such real AI is facing with many difficulties. The initial goal of AI has not happened yet because the brain does not work like a CPU. To simulate the brain with an AI machine, it should not execute instructions, but like what the brain does, pattern matching on experiences (sight, sound, smell, touch etc). Meanwhile this is a huge parallel machine as the brain’s responses are calculated with parallelism. The human brain has over 10^11 neurons and each in average has 10^3 magnitude synapses with other neurons. The neural networks simulations proved that artificial brains are possible, but the technology to simulate a system so vast like the human brain has not been invented yet. Actually the biggest constraint is we do not even know what it is that we are simulating. Brain is not a binary system. Every synapse reacts in a different way dependent on environmental variables (chemical composition, stress etc.). This is termed ‘plasticity’. We do not know how such plasticity enables the brain decodes/encodes information (i.e. the elusive neural code) in terms of signal rate, spike timings, noise, chaos, etc., and so that produce different functions. Since we do not even know the very fundamentals of how the brain processes information, simulating all those billions of synapses isn’t going to help us in anyway.

Apart from these complexities, many other critiques arise on the idea of artificial brain. The human brain evolved over a few hundred millions of years. The moment the first life was formed, development of our brain started out. It may be questionable that we will be able to squeeze that process into 50 years? Also it is worth thinking whether or not intelligence has a mathematical model. This is why, despite its fifty year history of overpromising and under delivering, AI is still popular.

All in all, today, AI is still in its infancy, making it difficult to tell just what to expect in the future. In fifty years, AI has given us heuristic search. What it hasn’t given us is any insight at all into either intelligence or the working of the brain. Actually these are also insights into how the brain doesn’t work, which is valuable. Perhaps the first 50 years of AI is just the prologue and 200 years from now we are going to smile back and think of this era as blind and stumbling people who were trying to make progress but didn’t know where to poke.

编译自Wired News的报道和OSNews的讨论:

这个夏天人工智能(artificial intelligence, AI)满50岁了。然而尽管计算机能够打败世界上最好的国际象棋选手，我们仍然无法令一台机器像一个四岁小孩那样思考。

当“人工智能”这个词在一个达特茅斯研讨会上(Dartmouth workshop)被创造出来之时，当时的想法是用计算的方法来探索近似人的智能。但是这个词本身很快地引发了公众的想象，使得这个术语深深植根于流行文化中。当时的科学家们雄心勃勃地想要制造出一台能够通过图灵测试(Turing test)的机器，即能够长时间地与人类交谈而无法被认出是机器。图灵测试被提出的当年，艾萨克·阿西莫夫(Isaac Asimov)写出了著名的《我，机器人》，描绘了一个机器能够同人无障碍并且智能地互动的世界。

直到今天，AI机器甚至还没有足够判断力来描述一张照片的内容，更不要说与人交谈了。但是1997年IBM的超级计算机“深蓝”打败了世界象棋冠军卡斯帕罗夫，这个机器的胜利标志了AI界的研究从复制人类智能转向了专门领域的深层技术。今天这些成果已经遍布现代经济的各个角落——银行的防交易诈骗系统、手机语音识别、搜索引擎的数据检索，等等。在商业领域之外，AI的应用包括帮助医生诊断疾病、利用视觉识别系统监视海滩和泳池防溺水，等等。如今大多数的科学领域依赖某种形式的AI。比如在生物学领域，AI用来理解从人类基因中得到的大量数据的涵义。AI工具已经成为科学家的强大工具用来理解数据集，因为我们已经进入一个时代，能够轻易获得大量数据（比如互联网），但是分析它们却很困难。

最近，AI在DARPA的一项机器车穿越沙漠的挑战赛中发挥了巨大作用。这项挑战的成功预示着未来机车能够自动驾驶，避免相撞，让人从驾驶的疲劳中解放出来。尽管家家有一台机器人的时代还没来临，AI在这方面已经前进了一大步。如果未来的计算能力能够以指数级增长，真正的人工智能也许不远了——那可能是一个人和机器结合的时代，我们的认知能力将大大增强，身体将更加健康。

事实上“真正的人工智能”仍然困难重重。AI最初的目标没有实现是因为以前人们认为大脑像一个CPU那样工作是错误的。用AI机器来模拟大脑，不能像CPU那样来执行命令，而应当像大脑那样基于经验（光、声、味、触）的模式匹配。同时这台机器还应当像大脑那样并行处理大量信息。人类大脑有超过10^11个神经细胞，平均每个细胞有10^3数量级的突触与其它神经细胞相连。因而尽管神经网络模拟证明人工大脑是可能的，但是模拟规模如此之大的系统仍然难以想象。事实上最大的障碍在于我们甚至不知道我们要模拟的是什么。大脑不是一个二进制系统，每一个神经突触根据环境变量（化学元素、压力等）发生不同的变化（称为可塑性，plasticity）。我们不知道这种可塑性如何使大脑编码解码诸如神经信号的变化率、峰值、噪声、混乱等信息（难以捉摸的神经编码），从而具备不同功能。因为我们不知道这些大脑的最基本原理，模拟数以万亿计的神经就无从谈起。

除了以上的复杂性之外，还有对人工大脑的其它种种质疑。自从第一个生命诞生起，人脑经历了数亿年的进化。我们真能把这数亿年的漫长时间压缩到五十年吗？也有人怀疑智能究竟能否用数学模型来描述，抑或只能用造物的语言？这也许是为什么过了五十年的过度乐观和产出不足，AI仍然很受欢迎。

无论如何，现如今AI仍然处在婴儿期，很难预见将来的发展。五十年，AI没有告诉我们究竟智能是什么，大脑如何工作，但同时它也告诉我们大脑不是如何工作的（例如启发式搜索、符号逻辑等），这些都很有价值。也许AI的这第一个50年只是序章，再经过200年的发展我们回过头来就可以微笑着看这个时代的人们，如何懵懂蹒跚地寻找答案，茫然不知该向那个方向走。

太热了……谁说苏格兰没有夏天的来着？今天烈日当头，阳光灿烂，天气预报说30度。没有空调也没有电扇，窗户只能打开一条小缝，生活在缺少通风条件的房间里真是一种煎熬。好在岛子上时常风云变幻，说不准啥时候飘来一片云天气预报就得实时修改。不过想想去年在炽热缺水的CC都熬过了一个夏天，顿时很知足，顺便再次钦佩一下自己。

wikipedia上的解释：

Neophilia is defined as a love of novelty and new things. A neophile is an individual who is unusually accepting of new things and excited by novelty.

The word has particular significance in Internet and hacker culture. The New Hacker’s Dictionary gave the following definition to neophilia -

The trait of being excited and pleased by novelty. Common among most hackers, SF fans, and members of several other connected leading-edge subcultures, including the pro-technology ‘Whole Earth’ wing of the ecology movement, space activists, many members of Mensa, and the Discordian/neo-pagan underground (see geek). All these groups overlap heavily and (where evidence is available) seem to share characteristic hacker tropisms for science fiction, music, and oriental food. The opposite tendency is neophobia.

我喜欢这个词

ZDNet reported that a Finnish company Nexstim has developed a method enabling mapping of the activities of the human brain in real time.

The new technology is a non-invasive brain scanning and stimulation system called navigated brain stimulation (NBS). This system “guides the precise delivery of targeted transcranial magnetic stimulation (TMS) pulses to discrete brain areas.” It is today the only device available for accurate prediction of the TMS stimulus location and dose within the human brain. This system thus can help diagnose and treat the human brain diseases, trauma, and dysfunctions.

NBS involves using short magnetic field pulses to stimulate precisely defined points in the cerebral cortex and then measure how a specific site or the whole cortex reacts with a high definition electroencephalogram (EEG). Impaired responses caused by disease or injury can then be detected by the equipment. While providing a region-based stimulation for EEG analysis of brain responses and function, another advantage of NBS is its ease of deployment and use. This emerging new technology is promising as a tool to discover the brain function in neuron population level via EEG results in constraint small cortical area. More precise EEG result can be expected to help model the neural mapping in cortex.

ZDNet消息：一家芬兰公司Nexstim开发了一种新技术使得实时脑成像变得更准确便捷。

这种非侵入性的脑扫描新技术叫做导航脑刺激(NBS)。这个系统能够“将经颅磁刺激(TMS)脉冲信号精确地发送到分散的大脑区域中”。这是目前为止唯一能够精确预测人脑中TMS刺激部位和辐射量的仪器。这个系统因而能够很大地帮助诊断和治疗大脑疾病、肿瘤和官能障碍。

NBS系统使用短脉冲磁场来刺激大脑皮层中确定的点，然后用脑电波(EEG)来测量某个部位或者整个大脑皮层如何反应这种刺激。由于疾病或者损伤而导致的受损的反应能够通过这种方式被检测到。NBS不仅仅能够使用EEG来分析对精确部位刺激的反应和功能，而且易于部署和安装。NBS作为一种能够在小面积皮层上集合神经元级别（经由EEG分析）的新技术来探索大脑功能是很有潜力的。更加精确的EEG结果能够帮助大脑皮层神经系统建模。

Wired news: Researchers at Columbia University has developed a “cortically coupled computer vision system,” which harnesses human visual processing power to identify faces and objects in images. Sponsored by DARPA, this new brain-computer-interface technology could turn our brains into automatic image-identifying machines that operate ten times faster than human consciousness. The potential applications are in military intelligence, face-recognition, anti-terrorism, etc.

Existing machine visual systems are generally of narrow range of purpose - they are built to specific tasks and limited in their ability to recognize suspicious activities or events. On the contrary, human brain is the best visual processor ever known, only limited by its processing speed. The proposed system’s advantage lies in combining the strengths of traditional computer vision with human cortical vision. When the human user sees an interesting through retina, the brain emits a signal detectable by an electroencephalogram, or EEG cap. The system analyses the signal and tag the causal object in the image or video stream, and ranks them in order of the strength of the neural signatures. In this way the human user only need to examine the information that their brains identified as important, instead of wading through thousands of images.

Wired消息，美国哥伦比亚大学的科学家研究的“脑机结合视觉系统”利用人脑视觉处理能力来识别人脸等物体，这项由DARPA资助的脑机界面技术把人脑变成一部自动图像识别机，识别速度可达到一般人的10倍。潜在的应用可以在军事情报、脸部识别、反恐等。

现有的机器视觉系统总得来讲应用面窄－它们多是为专有的目的建造，而且识别能力有限。相反，人脑是已知最好的视觉处理器，只是速度稍慢。这个新系统把传统的机器视觉技术同大脑视觉结合起来。当使用者看到感兴趣的物体时，大脑会发送特殊的信号，能够被脑电图机检测到。计算机分析这种信号，找到引发兴趣的物体，并在图像上根据大脑认为重要性的不同等级标记出来。这样使用者仅仅需要查看重要的信息，而不必在无数幅图像中寻找了。

在《金融时报》中文网上看到的一篇文章，转载一下。文章中说到的理想的生活模式，大约是我曾梦想过的。但是生活的经验告诉我，“想干什么就干什么”，或者说彻底闲适的生活，未必会带来快乐。相反地，往往这种心态引发的惰性非常强大，有可能导致对许多事情失去兴趣，令生活变得黯淡。我所理想中的生活，当有一条主轴，能够一生为之奋斗；然而无须终日操劳，主轴的生活可以被分割成一个个片段，其间可以填满各种创造的、探索的、漂泊的、或者所有“有趣”的事情。人生若能在创造一些价值的同时，体验诸般美好，是我目前所能想象的最理想的模式了。

理想的生活模式：35岁退休？

作者：英国《金融时报》中文网专栏作家 谁谁谁

2006年7月12日 星期三

现在人人都想在35岁之前赚足余生的生活费，然后退休。仿佛现在的日子不是生活，赚足了钱，告老还乡之后才能够开始生活。

这种理想的生活模式大概是这样的：醒了就起来，困了就睡觉，想干什么就干什么，旅行、美食、兴趣爱好、友情、慈善，等等……。

可是，且慢！

仔细想一想，就会发现这种生活有一个致命的特点：从纷繁芜杂的社会关系中淡出了。

社会地位没了。谁会记得有你这样一个吃饱了玩，玩累了睡的纯粹生活家呢？除非你爱出风头，经常召开新闻发布会宣传自己的事迹。但这岂不又落入了另一种事业的俗套？与那些每天东奔西走的小老板有什么区别呢？

友情没了。你闲着，不代表别人闲着。你的钱，只能保证自己行动的自由，却不能给你号令别人的自由。除非你筹办一个公司，专门养一批朋友陪着你玩。

话题没了。就像现在和父母没了话题一样，独自逍遥的你再也无法和依然忙碌着的老朋友找到共同语言。

期待没了。吃喝玩乐，当所有的一切享乐排着队等你的垂幸时，你依然觉得那是享乐吗？

有个房车俱全的朋友，去年终于实现了自己的心愿，辞了职，背起背包游历全国。按照想像中的情景，他走了全国20多个省市，并且坚持不剃须。四个月后回到上海，已经像个野人一样。略微休整了几天后，他得出一条全新的人生经验：“还是家里舒服。”

在亲身证明过4个月是在外晃荡的时间极限后，他决定呆在家里写一部小说。写了半年只得3000字，并越来越觉得这是件比每天朝九晚五还要痛苦的事。于是彻底放弃了这项追求。

以前是朝九晚五束缚了他，没时间完成这部巨作；现在是太多的时间包围了他，让他得以充分地考虑：写作，有这个必要吗？

几天前再见到他，他已经流露出想重新加入上班族的愿望。用他的话说，上班的时候，至少每天都能和活人说说话吧。

若是真的在35岁就赚足了钱、退了休，就能快乐吗？是因为天天能睡到太阳落山而快乐吗？是因为天天旅游而快乐吗？是因为可以设计点卖不出去的东西自娱自乐吗？望尽脚下的路，还有几十年那么长，每天的事业就是让自己快乐，这绝对是比朝九晚五还要艰巨的事。

作者谁谁谁是英国《金融时报》中文网生活时尚版编辑、专栏作家。本文摘自她的《白领极限生存》(Lost in Office)，此书于 5月1日由上海人民出版社出版。

Engadget: Researchers at University of Glasgow are working on an electronic optical implant will help blind people to regain their vision. By using microelectronics techonology, a small device containing an imaging detector (as CCD in an digital camera) is to be implanted on the retina. If light forms an image on the detector, then the result will be electrical stimulation of the retina in the shape of this image. The stimulated cells then send the information via the optic nerve to the brain. The implant prototype has 100 pixels but the team hope that number will increase significantly to 500 pixels as their work progresses, which would allow people to walk down the street and recognise faces. Such technology could be in use within a decade, and would assist people with age-related macular degeneration or retinitis pigmentosa. Team leader Dr Mathieson also mentioned that beyond where they are today it might be possible to make smart chips which have memory in them which would allow action replay and slow motion.

I would expect daunting human vision with implanted chip embracing more powerful image processing speed, much bigger pixel resolution, tele and macro capability, wireless control, etc

Engadget报道，英国格拉斯哥大学(University of Glasgow)的科学家正在研究一种电子光学移植技术，使得盲人能够恢复视力。利用微电子技术，科学家在视网膜上植入一个类似数码相机CCD的图像感应器，当光在这个感应器上成像之后，图像被转化成电刺激传到视网膜上。之后视网膜上的色感细胞就像原生的那样将视觉信号通过视神经传递给大脑。目前这种植入感应器的原型只有100个像素大小，但研究小组希望将来能够进步到500像素，那样可以让人进行类似走路或者脸识别的活动。这种技术可以在未来十年内投入使用，将会给因为老年性黄斑病变或视网膜色素变性而失去视觉的患者带来福音。研究者Mathieson博士还表示，将来更智能的感应芯片可以加入记忆体，实现诸如回放或者慢镜头等功能。

要是能实现更强大的图像处理、巨量像素、变焦微距、无线遥控等功能，将视网膜变成数码相机可以带来完全不同的视觉体验。