今天，所謂的“智能系統(tǒng)”(smart system)，能夠直觀地處理自動化任務、整合資料中心的分析數(shù)據(jù)和嵌入式計算機，為企業(yè)和消費者創(chuàng)造價值。
我們賦予這些計算機人類的名字──Watson、Siri──告訴我們自己，他們‘有多們像我們’。今天的智能系統(tǒng)已經(jīng)能夠直觀地處理迄今仍無法實現(xiàn)實時自動化的任務。透過匯聚來自全球數(shù)十億的信息串流，這些智能系統(tǒng)提供的分析科學能力，可以創(chuàng)造甚至超越其系統(tǒng)本身的價值。
今年稍早，在Jeopardy游戲中戰(zhàn)勝人類冠軍的IBM Watson超級電腦獲得了廣泛關注。蘋果(Apple) iPhone-4S的Siri能用接近自然的英文來回答使用者問題，在一部任何人都能夠買得起的手機中，已經(jīng)搭載了如此驚人的智能系統(tǒng)。
雖然這些系統(tǒng)得到廣大贊譽，不過，在幾乎所有的電子產業(yè)──汽車、工業(yè)、通訊、計算機、交通、能源、醫(yī)療和個人保健──等領域中，或多或少都已經(jīng)開始運用智能系統(tǒng)了。事實上，根據(jù)美國總統(tǒng)的科學和技術顧問委員會表示，“網(wǎng)絡實體系統(tǒng)”(cyber-physical systems)最終將串連全球50%左右的電子產品。
美國國家標準與技術策員會(NIST)最近因而針對互操作性定義了一個接口，可作為智能系統(tǒng)之間的性能比較指針及標準化。現(xiàn)階段所做的努力，是協(xié)助美國企業(yè)自主開發(fā)設計智能系統(tǒng)，但這些設計會在海外進行低成本生產。
這個賭注非常巨大。市調公司IDC近日指出，每年的智能系統(tǒng)銷售量近20億部，市場規(guī)模達1兆美元，IDC預測到2015年該市場將成長一倍到40億部，規(guī)模達2兆美元。
而據(jù)IDC的分析，最具價值的智能系統(tǒng)，是能針對實時信息串流進行分析的服務。
“資料是全新型態(tài)的貨幣，”IDC半導體研究副總裁Mario Morales說?！跋馡BM、惠普(HP)、英特爾(Intel)、微軟(Microsoft)、德州儀器(TI)、飛思卡爾(Freescale)和甲骨文(Oracle)等公司，都早已明白資料的價值，而且致力于開發(fā)能進行資料分析以獲得最大價值的基礎設備?！?
“過去三年來，我們不僅看到計算機在轉型，網(wǎng)絡也是，甚至包括使用者與智能設備互動的方式也在轉變。企業(yè)尚未找出這些資料可以怎樣轉換成利潤，但其中確實蘊藏著巨大的機會，促使有遠見企業(yè)投資在分析軟件和服務領域，以便與其智能硬件結合。
10多年來，IDC一直將相關設備定義在嵌入式計算機領域，最近才開始將‘智能系統(tǒng)’定義為嵌入式領域的繼任者。IDC并不是唯一一家聲稱智能系統(tǒng)是未來大勢的市場研究公司。另一家位于紐約的Applied Business Intelligence Inc.,最近也激活了一項名為‘智能城市電網(wǎng)’的研究服務。
本文下一頁：IBM引領智能系統(tǒng)發(fā)展
{pagination}
IBM引領智能系統(tǒng)發(fā)展
IBM或許是對當前的智能系統(tǒng)有著最深刻理解的公司。在全球各地，已經(jīng)有數(shù)十個該公司稱之為‘智能行星’(smart planet)的系統(tǒng)用于解決各式各樣的基礎設施問題，如位于瑞典斯德哥爾摩的智能運輸系統(tǒng)；馬耳他的國家級智能電網(wǎng)(也是全球第一個)；以及設置在紐約大都會博物館，用于保護藝術品的無線傳感器網(wǎng)絡等。
“智能系統(tǒng)每天所產生的資料比美國所有圖書館的資料加起來還要多出8倍──而其中有85%都是非結構化的，”IBM院士暨技術長及系統(tǒng)部門技術策略副總裁Jai Menon說?！澳壳暗纳虡I(yè)情報仍難以對所有非結構化資料進行分析并進一步獲取價值，而Watson則是一個很好的例子，因為它能快速回答有關非結構化資料的問題?！?
傳統(tǒng)的IT分析都執(zhí)行結構化資料，透過數(shù)據(jù)庫整理所有資料，可以輕易地進行搜索、排序，并使用知名的數(shù)學公式進行分析。但Watson證實凌亂的、非結構化的資料，也能借著巧妙制作分析優(yōu)勢，在最佳化的系統(tǒng)架構上輕易地進行搜索。
“針對金融市場的分析──如商品價格預測──是透過已知周期性模式進行的。然而，要預測基礎設計的失效風險，如水管究竟有多長，而最后這就是我們所說的非結構化問題，”IBM研究中心工程師暨商業(yè)分析總監(jiān)Arun Hampapur說。
IBM最近運用智能系統(tǒng)解決非結構性問題的例子，主要是利用Watson為醫(yī)療保健、銀行和金融、零售、法律和政府監(jiān)控等領域建立自動化顧問?！拔覀兠刻於冀拥礁餍懈鳂I(yè)領導人的電話，他們都希望Watson能展開更多新應用，舉例來說，如何能更快速、更方便地訂機票等，”Hampapur說。
另一個例子，全美最大型的保健服務供貨商WellPoint公司最近宣布，將采用Watson所衍生的智能系統(tǒng)，透過從數(shù)百萬的醫(yī)療記錄、期刊文獻和最新醫(yī)學研究結果中，獲取和患者癥狀匹配的資料來簡化并加速醫(yī)療診斷。
Watson是以IBM所建構的技術為基礎，這些技術是為了解決智能城市開發(fā)項目中非結構化問題所開發(fā)的。IBM的探索行動是從運用其傳統(tǒng)資中心分析的長處開始，而后不斷朝智能系統(tǒng)的方向進行開發(fā)。這家公司不斷朝著可運用嵌入式處理器本身進行分析的邊緣連接網(wǎng)絡方向努力。Menon指出，芝加哥警察“在邊緣網(wǎng)絡使用了智能分析，能自動將保全攝影鏡頭轉向qiang響的方向，因此當接到911報案電話時，他們已經(jīng)能夠獲得qiang枝口徑讀數(shù)和攝影鏡頭所轉向的方位等信息了?！?
過去幾年內，IBM已經(jīng)花費超過150億美元，用于收購具備專業(yè)分析知識的公司，該公司希望為可融合來自多個感測輸入的智能系統(tǒng)，開發(fā)新一代的感知計算機(cognitive-computer)芯片。
本文下一頁：惠普──人人有機會
{pagination}
惠普：人人有機會
在此同時，惠普也正在轉變其業(yè)務模式，希望能運用無線傳感器網(wǎng)絡提供智能系統(tǒng)，讓惠普的云端服務器能與多種不同的實時信息串流通訊，以執(zhí)行各種資料分析工作，預測從公共電力中斷到個人心臟病發(fā)作等一切事物。
“我們并未試著模仿IBM，但我們發(fā)現(xiàn)，我們與他們面對著相同的機會，很明顯，對一個完整的系統(tǒng)而言，資料分析的價值愈來愈高，”HP資深院士Stanley Williams說?！拔覀兇罅客顿Y在資料分析上，因為這能將0與1轉變?yōu)橐饬x的東西，它能創(chuàng)造并提供知識和意識，讓人們能夠迅速對任何情況做出反應，以預防不良后果?！?
在建立第一代智能系統(tǒng)時，惠普僅側重在能源和保健領域，這家公司從頭開始建構系統(tǒng)，包括了傳感器芯片到執(zhí)行在云端的分析軟件。
“智能系統(tǒng)代表了極為龐大的發(fā)展過程，它涉及了信息技術的各個領域，但我們決定，僅運用我們首個垂直整合的平臺進入兩個領域，”Williams說。“我們希望至少進入兩個應用領域，如此我們就能比較和對照不同的應用，以了解何者是共通的，何者又是需要差異化的。而后，一旦我們在這兩個領域建立了基礎，我們便可以再轉向其它的垂直細分市場?！?
惠普是透過與殼牌石油(Shell Oil)這類客戶合作而首次將其原型擴展至大規(guī)模應用，殼牌石油已經(jīng)與HP簽署了無線傳感器網(wǎng)絡合約，可運用該技術進行智能地震成像。在HP服務器上執(zhí)行的分析工作，會將來自于數(shù)千個HP地震傳感器的資料串流轉化為實際可用的情報資料，告訴這些公司哪里可以進行石油開采。
HP同時負責制造專門針對地震感測應用的專用MEMS加速器，以及負責將資料串流送回服務器進行分析的無線傳感器節(jié)點。
本文下一頁：英特爾──先做好本地處理
{pagination}
英特爾：先做好本地處理
英特爾正透過增加本地分析能力來推動智能系統(tǒng)業(yè)務，隨著嵌入式系統(tǒng)的自然演進，該公司也更加專注在軟件領域，讓OEM能在Intel X86及Atom處理器上執(zhí)行分析，而不是直接將原始資料串流送到云端上。
“隨著傳感器日益普及，嵌入式系統(tǒng)已經(jīng)開始創(chuàng)造大量信息，這些信息都會匯流到云端，”英特爾副總裁暨嵌入式通訊集團總經(jīng)理Ton Steenman說?！拔覀冋J為一切都要送到云端上的想法并不合理，事實上，我們建議，應該在嵌入式處理器上執(zhí)行需要實時分析的任何問題?！?
英特爾去年收購了CognoVision Solutions Inc.的不記名視頻分析(anonymous video analytics, AVA)的技術，該技術可執(zhí)行在x86處理器上，英特爾將讓技術重新命名為Intel Audience Impression Metrics Suite (AIM)。AIM可執(zhí)行在本地數(shù)字看板系統(tǒng)上，并依照觀眾類型來更換播放的廣告內容。
“過去，數(shù)字看板和配備低階處理器的媒體播放器沒什么兩樣，”Steenman說?！暗F(xiàn)在數(shù)字看板已經(jīng)添加了相機，可在本地進行智能分析來識別觀眾的性別和年齡，然后配合觀眾來更改播放的廣告?！?
英特爾還收購了McAfee和嵌入式操作系統(tǒng)供貨商Wind River。這些收購而來的解決方案都與英特爾的遠程程管理工具搭配，將PC類的安全策略擴展到智能嵌入式系統(tǒng)中。
英特爾的數(shù)字看板提供了能與消費者互動的顯示器，如阿迪達斯的球鞋廣告，可對觀眾的性別、年齡及興趣進行本地分析，以確定要發(fā)送的資料。
本文下一頁：微軟──將所有的智能特性整合起來
{pagination}
微軟：將所有的智能特性整合起來
微軟公司也藉由擴展其智能系統(tǒng)的軟件兼容性，來擴展嵌入式業(yè)務，目前其Windows已經(jīng)能執(zhí)行在 ARM、MIPS和x86嵌入式處理器到更高階的Xeon服務器上。截至目前，該公司稱Windows平臺已有超過300萬個嵌入式系統(tǒng)，并希望運用此一成果進入智能系統(tǒng)。
“我們的策略優(yōu)勢，是將Windows環(huán)境從嵌入式智能系統(tǒng)再升級，用戶目前透過我們的系統(tǒng)收集資料，傳回云端Windows服務器，”微軟Windows Embedded行銷部資深總監(jiān)Barb Edson 表示。
舉例來說，微軟現(xiàn)在有一家大型企業(yè)客戶，在裝載食物的貨箱內設置了傳感器，來感測果實的成熟情況，因此當貨抵達碼頭時，不同熟成度的水果便可以分開運送到正確的目的地，可能被送到一個必須要人工熟成的倉庫，或是將已成熟的水果直接送到生產廠，從而節(jié)省了人工檢查步驟。
“在所有這些智能嵌入式系統(tǒng)中，情報是最重要的關鍵，”Edson說?！拔覀兿嘈?，一切都會被連接到網(wǎng)際網(wǎng)絡，最終演變成智能系統(tǒng)。”
編譯： Joy Teng
參考原文： In a smart-system world, data’s ‘the new currency’，by R. Colin Johnson
{pagination}
In a smart-system world, data’s ‘the new currency’
R. Colin Johnson
We give them human names—Watson, Siri—that suggest how much “l(fā)ike us” they are. Today’s smart systems can intuitively handle tasks that until now have been impossible to automate in real-time. And by mining the resultant sea of real-time data coming in from billions of streams worldwide, analytics science is creating services that have even more value than the smart systems themselves.
IBM’s Watson supercomputer captured the public’s attention earlier this year when it beat human champions at Jeopardy. Siri, the intelligent agent on the Apple iPhone-4s, answers users’ ad hoc questions about almost anything in natural, conversational English, putting a scary-smart system in the pocket of anyone who can afford the phone.
While those systems get the glory, there’s a seething mass of smart systems already at work in virtually every electronics sector: automotive, industrial, communications, computing, transportation, energy, medical and personal health maintenance. In fact, according to the U.S. President’s Council of Advisors on Science and Technology, such “cyber-physical systems” will eventually constitute 50 percent of all electronics worldwide, making them a U.S. strategic asset.
In response, the National Institute of Standards and Technology recently announced a standardization effort to define interfaces for interoperability, as well as metrics and methods for measuring and comparing performance among smart systems. Such efforts set the stage for U.S. entrepreneurs to build successful smart systems from homegrown designs, but to realize those designs with electronics that are manufactured at low cost overseas (see sidebar, final page).
The stakes are huge. Market watcher International Data Corp. (Framingham, Mass.) recently reported that nearly 2 billion smart systems per year are already being sold, making for a $1 trillion market that IDC predicts will grow to 4 billion units and $2 trillion by 2015.
The most valuable services performed by smart systems, according to IDC, result from the application of analytics to real-time data streams.
“Data is the new currency,” said Mario Morales, vice president of semiconductor research at IDC. “And the companies that understand this are the ones already developing the analytics and infrastructure to extract that value—companies like IBM, HP, Intel, Microsoft, TI, Freescale and Oracle.
“Over the last three years, we have seen a transformation not merely in computing, but also in networking, and even in the way users are interacting with smart devices. Enterprises have yet to figure out exactly how to monetize all this data, but there is a tremendous amount of opportunity there, which is prompting visionaries to make huge investments in the analytics software and services that will couple to their intelligent hardware.”
IDC has been covering embedded computers for over a decade but only recently started delineating “intelligent systems” as the successor to the embedded space. And IDC is not the only market forecaster claiming that smart systems are the future. Applied Business Intelligence Inc. (New York), for one, recently started a “smart cities and grids” research service.
IBM ahead of curve
IBM probably has the deepest understanding of smart systems today. Dozens of its so-called smarter-planet systems are already solving widespread infrastructure problems worldwide, including a smart transportation system in Stockholm, Sweden; a national smart grid—the world’s first—in Malta; and a smart wireless sensor network that protects paintings at the New York Metropolitan Museum of Art.
“Smart systems are generating eight times more data every day than there is in all the U.S. libraries combined—85 percent of which is unstructured,” said Jai Menon, IBM fellow, chief technology officer and vice president of technical strategy for the company’s systems group. “Business intelligence has the problem of using analytics to derive value from all that unstructured data, and Watson is a good example of how to answer questions about unstructured data very quickly.”
Traditional IT analytics were run on structured data that was carefully tailored by database experts into neat, isomorphic containers that could be easily searched, sorted and analyzed using well-known mathematical formulas. But Watson proved that messy, unstructured data can also be easily searched, by virtue of cleverly crafted analytics designed to run on optimized system architectures that preposition the technological capabilities needed to address specific unstructured problem domains.
“Analytics for the financial markets—such as predicting commodity prices—is a cyclical phenomenon driven by well-known patterns. But predicting the risk of failure in infrastructure—say, how long a water pipe will last—is what we call an unstructured problem,” said Arun Hampapur, distinguished engineer and director of business analytics at IBM Research. “And analytics for unstructured problems is best done by instrumenting a strategy that custom-tailors the analytics and architecture for a particular problem domain.”
IBM’s latest foray into addressing unstructured problem domains with smart systems is aimed at using Watson to create automated advisers for apps in health care, banking and finance, retailing, law and governmental regulation. “We get calls every day from industry leaders who want to repurpose Watson for new applications, such as helping to make airline reservations faster, better, easier,” said Hampapur.
For instance, Wellpoint Inc. (Indianapolis), the nation’s largest health-care provider, recently announced that it would use a Watson-derived smart system to simplify and speed medical diagnoses by matching patients’ symptom sets with data from millions of medical records, journal articles and late-breaking medical-research results.
Watson is based on technologies that IBM created to solve unstructured problems in smart-city projects. IBM started its quest for such smart systems by leveraging its strengths in the data center, where traditional analytics are run. But it has been steadily working out toward the edge of the connectivity network, where analytics can be run on the embedded processors themselves. Menon noted that Chicago police “use smart analytics at the edge to automatically turn surveillance cameras toward a gunshot, so that by the time the 911 call comes in they already have a readout of the caliber of gun that was used and a camera pointing at the location from which it was fired.”
IBM has spent more than $15 billion in the past few years acquiring companies with specialized analytics expertise, and it is building a new generation of cognitive-computer chips for smarter systems that can fuse the inputs from multiple sensors.
HP sees 'the same opportunities'
Hewlett-Packard, meanwhile, is transforming its business model to deliver smart systems that harness wireless sensor networks in order to communicate vast streams of real-time data to HP’s cloud-based servers, where analytics can be run to predict everything from public power outages to personal heart attacks.
“We are not explicitly trying to emulate IBM, but we are finding the same type of opportunities as they are, because it’s clear that more and more of the value is in the system as a whole,” said senior HP fellow Stanley Williams. “The money will be in the analytics, because analytics is what turns ones and zeros into something meaningful; it creates the knowledge and awareness that allows people to quickly react to situations, and to prevent undesirable outcomes.”
In creating its first generation of smart systems, HP is focusing on just two sectors—energy and health—for which it is building systems from scratch, from the sensor chips to the analytics software running in the cloud.
“Smart systems represent a huge development effort, involving every aspect of information technology, but we have consciously decided to only enter two sectors with our first vertically integrated platform,” said Williams. “We wanted to enter at least two so that we could compare and contrast the different applications in order to understand what
is common and what needs to be different. Then, once we have those two under our belt, we’ll look to the other vertical market segments.”
HP is prototyping its first widescale applications in cooperation with customers such as Shell Oil, which has contracted for a wireless sensor network that can perform smart seismic imaging. Analytics run on HP servers will turn data streams from thousands of HP seismic sensors into practical intelligence indicating where to drill.
HP manufactures both a specialized MEMS accelerometer for seismic sensing and a wireless sensor node that streams the sensor data back to its servers, where analytics are run.
Intel takes the local
Intel Corp., for its part, is pursuing a smart-systems business that adds local analytics as the natural evolution from the embedded model, focusing on software that lets OEMs perform analytics on Intel X86 and Atom processors instead of diverting raw data streams up to the cloud.
“As sensors become more pervasive, embedded systems have begun creating large volumes of data that today flow straight into the cloud,” said Ton Steenman, vice president and general manager of Intel’s Embedded Communications Group. “We believe that it is unreasonable to think that everything should move to the cloud; in fact, we advise that any problems needing real-time analytics should be run on the embedded processor itself.”
Intel last year acquired CognoVision Solutions Inc. (Toronto) for that company’s anonymous video analytics (AVA) technology, which runs on X86 processors, and branded the technology the Intel Audience Impression Metrics Suite. AIM runs locally on digital signage and changes the ads displayed depending on who is viewing them.
“Digital signs in the past were just media players with a low-end processor that passed through a media stream to the sign,” said Steenman. “But now that cameras have been added, smart analytics running locally can recognize gender and age, then change the advertisement on the sign to match the viewer.”
Intel also recently acquired security powerhouse McAfee and embedded OS specialist Wind River. Those acquisitions now are collaborating to use Intel’s remote management tools to extend PC-like security strategies down to smart embedded systems.
Intel's digital signs offer consumers interactive displays—here of Adidas shoes—while using local analytics to ascertain and send data about users’ gender, age and interests to cloud servers.
