是時候靠譜地談一談人工智能技術(shù)的未來了。之所以說這話，是因為坊間對人工智能技術(shù)的探討往往很不靠譜。很多人談到人工智能，往往會走向兩個相反的極端：要么精準(zhǔn)地預(yù)測人工智能未來幾年會催生或摧毀多少個就業(yè)機會，要么斷言人工智能會將我們的世界變成天堂（或者地獄）。因此，在本文中，我們決定靠譜地討論人工智能會怎樣改變商業(yè)的面貌。在這個過程中，我們會盡量減少戲劇化的猜想。

“人工智能焦慮癥”的一大癥狀，就是擔(dān)心人工智能會造成大面積失業(yè)。然而實際上，未來會發(fā)生什么，沒有人知道，也不可能有人知道。全球有千百萬企業(yè)家和經(jīng)理人在應(yīng)用著各種快速革新的技術(shù)，他們會創(chuàng)造出怎樣的奇跡，我們永遠(yuǎn)也猜想不到。舉個例子，美國郵政部長亞瑟·薩默菲爾德1959年曾自信地預(yù)言稱，過不了多久，郵件就會用導(dǎo)彈來運送了，因為導(dǎo)彈是當(dāng)時人們能想象出來的最神奇的技術(shù)。隨著二戰(zhàn)后經(jīng)濟的發(fā)展，郵政部門要寄送的信件越來越多，在當(dāng)時看來，郵政工人無疑是一份好工作。當(dāng)時薩默菲爾德絕對想不到，有一天信件將不再被寫在紙上——雖然當(dāng)時電子郵件、短信和無線通信網(wǎng)絡(luò)的雛形已經(jīng)誕生，或至少已在研發(fā)階段。在人工智能的問題上，我們可能也會犯同樣的認(rèn)知錯誤。

第二個重要事實，是人工智能的最終應(yīng)用情況，很大程度上將由市場的力量決定。很多人以為人工智能將把世界變成一個烏托邦式的理想國，他們都忽視了市場這只“看不見的手”的作用。美國無線電公司老板大衛(wèi)·薩諾夫曾經(jīng)預(yù)言，在彩色電視普及后，大家就可以在家里欣賞藝術(shù)作品了。這聽起來當(dāng)然是極好的，但是沒人想把它用在這么高雅的用途上。人工智能也是一樣的，它會被廣大企業(yè)和消費者用在數(shù)不清的實際用途上，其中大多數(shù)用途談不上好也談不上不好，但其累積效果是無法預(yù)見的。在我們思考人工智能的未來時，關(guān)鍵是不要把自己放在一個道德的高地上，而是要像現(xiàn)實世界的每一個趨利避害的人（包括好人和壞人）一樣思考。

不過本文不會討論壞人的例子。我們將在文中討論25個人工智能的有益案例，其中有些案例非常具有啟發(fā)性——而且它們都是真實的。

人工智能如何改變你的工作方式

讓所有人都說同一種語言

從《神秘博士》（Doctor Who）和《星際迷航》（Star Trek）開始，科幻作品中就出現(xiàn)了能自動翻譯語言的機器，有了它，人類和外星人就算不學(xué)習(xí)對方的語言也能無障礙地交流。現(xiàn)在，地球上的一些公司已經(jīng)在制造這種設(shè)備了。谷歌近日發(fā)布的Pixel Buds智能耳機就是一例，有了它，美國公司的高管就可以給說葡萄牙語的同行打電話，談?wù)効鐕献鞯氖聝海豢鐕髽I(yè)的員工可以更順暢地與其他國家的同事溝通，哪怕他們并不會說同一種語言；銷售人員可以給一個陌生地區(qū)的客戶打推銷電話，不必?fù)?dān)心對方聽不懂自己在說什么，說不定這一通電話就是公司咸魚翻身的機會。雖然很多跨國公司都將英語定為公司的官方語言，不過基于人工智能的同聲傳譯技術(shù)卻使非英語母語者可以繼續(xù)說自己的語言，保持自己的文化特點——在全球化的時代，這顯然是一個優(yōu)點。

讀心術(shù)

語音控制是一項挺“酷”的技術(shù)。然而無論是亞馬遜的Alexa、蘋果的Siri還是微軟的Cortana，你在公開場合跟它對話，都顯得特別尷尬，也比較打擾別人。然而麻省理工學(xué)院的研究人員已經(jīng)發(fā)明出了一種名叫AlterEgo的神奇裝置，它是一種非侵入性的可穿戴設(shè)備，它可以在你開口說話前就知道你要說什么。AlterEgo可以在幾秒鐘內(nèi)回答很多問題，也可以發(fā)送私人信息，或者內(nèi)部記錄信息流以留待稍后處理，這些都無需任何外部可觀察到的操作。當(dāng)然，AlterEgo并非真的有讀心術(shù)，不過它可以解讀人體下頜骨的電脈沖，這個位置正是人體的發(fā)聲器官，從而使AlterEgo能做到“聲未發(fā)而先知”。目前，麻省理工學(xué)院的研究人員仍在繼續(xù)收集數(shù)據(jù)，并對該系統(tǒng)進行訓(xùn)練。以后，該系統(tǒng)可以作為高噪音環(huán)境下的一個溝通平臺使用，也可以用于患有語言障礙的人士。不過AlterEgo雖然具有大大加速書寫、計劃和溝通的能力，但人類終究還是要花不少時間閱讀這些文字。

34%

根據(jù)Pegasystems公司的一項調(diào)查，34%的人表示，他們曾經(jīng)有過與人工智能互動的經(jīng)歷。（實際上，84%的人都曾與人工智能互動過。）

更聰明地招聘

千里馬常有，而伯樂不常有。招聘是一個特別容易受個別人主觀因素影響的過程。一個人很容易因為某個求職者的姓名、畢業(yè)院校甚至簡歷上的字體大小就對他產(chǎn)生好惡。所以現(xiàn)在有些公司已經(jīng)開始在招聘中尋求人工智能技術(shù)的幫助了。

比如沃達豐、尼爾森和聯(lián)合利華等公司在招聘時，會讓求職者先玩一款由AI創(chuàng)業(yè)公司Pymetrics開發(fā)的手游，這款手游能夠評估求職者的認(rèn)知和情商水平，同時在設(shè)計中有避免了所有種族、性別等其他因素的影響。當(dāng)軟件篩選出表現(xiàn)最好的一批求職者后，聯(lián)合利華會要求他們在HireVue網(wǎng)站上錄一段視頻，他們在視頻中要回答一些問題，比如如何解決工作中遇到的各類挑戰(zhàn)。該網(wǎng)站的人工智能算法不僅會分析求職者說了什么，還會觀察他們的反應(yīng)有多快，以及他們的面部表情透露了什么情緒線索。通過了這些初步測試的求職者便會得到真人面試的機會。

聯(lián)合利華表示，啟用該系統(tǒng)后，人才對該公司工作邀約的接受率提高了；從種族、民族和社會經(jīng)濟狀況等多個指標(biāo)上來看，公司人才的多元化也提高了。以錄取新員工的畢業(yè)院校數(shù)量來看，該公司新員工教育背景的多元化程度達到了以往的三倍。

20%

據(jù)英國Computerlove公司的調(diào)查顯示，有20%的人希望語音助手程序能使他們變得“更幽默或更有魅力”。

打造終極經(jīng)理人

很多人根深蒂固地認(rèn)為，只有人類才有資格評價人類的行為。不過那是過去了。現(xiàn)在，計算機算法已經(jīng)在越來越多地評估我們的行為甚至意圖，并得出結(jié)論。尤其是在職場中，為了了解潛在的人員流失風(fēng)險、高績效員工的特質(zhì)，以及哪些因素有助于保持團隊的活力，人力資源部門很多時候都會求助于人工智能程序。比如波士頓的Humanyze公司就在試驗一種智能身份徽章，它可以持續(xù)追蹤員工之間的溝通情況，使雇主能夠找到其中的模式，分析公司的工作實際上是怎樣完成的。

西雅圖的創(chuàng)業(yè)公司Textio則使用人工智能技術(shù)幫助企業(yè)撰寫合適的招聘廣告（該公司的“增強寫作平臺”在通過堆砌語言以吸引多元化的求職者方面特別有效）。很多大公司在人力資源領(lǐng)域也引入了AI技術(shù)。比如英特爾公司正在研究使用人工智能技術(shù)開發(fā)一項新的內(nèi)部工具，以將員工與公司內(nèi)部的其他工作機會進行匹配，以更好地保留人才。

人工智能的這些新功能可以幫助企業(yè)吸引和留住他們所需要的人才，待這些流程實現(xiàn)自動化后，企業(yè)的招聘成本也將有所下降。那么它有沒有什么缺點呢？一個潛在的風(fēng)險，就是有可能造成企業(yè)與員工的疏遠(yuǎn)——畢竟員工都不喜歡雇主越來越多地侵入他們的生活。

人工智能將如何顛覆華爾街和銀行業(yè)

你的按揭貸款經(jīng)理是機器人

次貸危機爆發(fā)后冒出了一種新觀點：機器可能比人類更知道如何正確地發(fā)放住房貸款。最近，房利美對抵押貸款機構(gòu)進行的一項調(diào)查發(fā)現(xiàn)，美國有40%的抵押貸款銀行已經(jīng)采用了人工智能技術(shù)來處理手續(xù)繁瑣的申請流程，檢測客戶可能的欺詐行為，以及預(yù)判借款者的違約風(fēng)險。比如舊金山的Blend公司已經(jīng)為包括富國銀行在內(nèi)的114家銀行提供了在線抵押貸款申請程序，使貸款審批過程縮短了至少一星期。假設(shè)當(dāng)年有這種人工智能技術(shù)，那么次貸危機還會出現(xiàn)嗎？我認(rèn)為即便不能完全避免，至少也能減輕次貸危機的烈度，因為機器會比人類更早發(fā)出預(yù)警信號。Blend公司聯(lián)合創(chuàng)始人、CEO尼瑪·甘沙里表示：“關(guān)于數(shù)據(jù)的錯誤決策可以在瞬間被發(fā)現(xiàn)和修正。”雖然銀行尚未開始基于人工智能的評估結(jié)果來審批貸款的發(fā)放，但很多銀行已經(jīng)發(fā)現(xiàn)了人工智能程序的另一個好處——能讓更多美國人獲得住房貸款。Blend公司定義的“低收入群體”一向不愿申請抵押貸款。但現(xiàn)在，該群體通過Blend的移動應(yīng)用申請房屋貸款的可能性是其他階層的三倍。富國銀行的消費銀行業(yè)務(wù)主管瑪麗·馬克表示： “它消除了人們的恐懼。”

為專業(yè)投資者帶來新優(yōu)勢

過去十年間，金融行業(yè)的數(shù)據(jù)量呈爆炸式增長，即使那些20來歲的分析師不眠不休地干，也不可能處理完所有數(shù)據(jù)。人力雖然做不到，機器卻可以。因此，彭博、FactSet研究系統(tǒng)和湯森路透等金融研究機構(gòu)都開發(fā)了一系列數(shù)據(jù)分析工具和技術(shù)，包括機器學(xué)習(xí)、深度學(xué)習(xí)和自然語言處理（NLP）技術(shù)等，以方便成千上萬的專業(yè)金融人士迅速從海量信息中挖掘出有價值的見解。

彭博就是使用情感分析（亦是自然語言處理技術(shù)的一種）技術(shù)的先驅(qū)，彭博從10年前就開始研發(fā)這項技術(shù)了。簡單說來，它所使用的機器學(xué)習(xí)技術(shù)會識別出某條新聞或某篇網(wǎng)文與一只股票有關(guān)，并賦予它一個情感分?jǐn)?shù)。除了用于股市分析，人工智能技術(shù)也在向財富管理領(lǐng)域拓展。過去五年間，由于整個行業(yè)都在爭相發(fā)掘包含在網(wǎng)站文章、語音分析、信用卡購買數(shù)據(jù)和衛(wèi)星數(shù)據(jù)中的交易信號，各大投資集團里的所謂“另類數(shù)據(jù)”分析師的人數(shù)增加了四倍有余。包括貝萊德、富達、景順、施羅德和普信集團在內(nèi)的行業(yè)頂級研究機構(gòu)都在使用人工智能技術(shù)。全球最大的資產(chǎn)管理公司貝萊德集團也是應(yīng)用人工智能技術(shù)的先行者，它還建立了一個“貝萊德實驗室”，專門用于開發(fā)人工智能技術(shù)。

72%

根據(jù)皮尤研究公司的調(diào)查，有72%的人擔(dān)心機器人會搶走他們的工作。

業(yè)余投資者也能受益

由Betterment等創(chuàng)業(yè)公司和嘉信理財?shù)葌鹘y(tǒng)經(jīng)紀(jì)公司推出的“機器人理財顧問”服務(wù)，就是利用人工智能技術(shù)服務(wù)普通投資者的例子。這些理財工具的費用較低，它們基于你的風(fēng)險偏好，用計算機算法來決定你的資產(chǎn)應(yīng)該如何在股票、債券和其他資產(chǎn)上進行分配。這些公司的AI技術(shù)可以自動調(diào)整你的投資組合。當(dāng)人工智能程序判定你需要合理避稅或者需要遺產(chǎn)規(guī)劃方面的幫助時，它還會讓一名顧問（非機器人）打電話給你。

與此同時，一些金融機構(gòu)也在研發(fā)能夠幫助投資者做出明智的長期決策的投資工具。美銀美林和摩根士丹利就是“量化基本面分析”這一新興領(lǐng)域的兩個大玩家。他們的目標(biāo)是在基本的AI定量分析（也就是從海量數(shù)據(jù)中識別出模式）的基礎(chǔ)上，結(jié)合由最頂尖的人類分析師的復(fù)雜分析訓(xùn)練的額外算法，用以進行基本面投資的評估，比如評估一個行業(yè)的潛長潛力，或是一家公司管理層的戰(zhàn)略敏銳性等等。有了機器學(xué)習(xí)技術(shù)的加成，量化基本面分析系統(tǒng)將能夠從失誤中學(xué)習(xí)，進而不斷完善。最終，普通投資者只需要花很少的錢，就擁了股神巴菲特般的長線投資智慧。到時，這套系統(tǒng)也可能有了一個比“量化基本面分析系統(tǒng)”更潮的名字。

人工智能如何改變我們制造事物的方法

更高效的設(shè)計

有人說，人工智能雖然對科技、醫(yī)療等行業(yè)沖擊很大，但我們這些搞藝術(shù)創(chuàng)作的總歸是安全的吧？并不完全是。美國有一家叫Autodesk的軟件公司開發(fā)了一款叫Dreamcatcher的軟件，它可以利用人工智能技術(shù)輔助人類設(shè)計師進行創(chuàng)作。這款軟件已經(jīng)被空中客車、安德瑪和史丹利百得等多家知名企業(yè)采用。這款軟件充分展示了機器也能創(chuàng)造出嘆為觀止的設(shè)計。人類設(shè)計師只需要輸入需求、限制以及其他品質(zhì)要求（甚至包括材料的總成本），軟件就會自動生成幾百甚至幾千種設(shè)計方案。設(shè)計人員可以對這些方案進行篩選，在篩選的過程中，系統(tǒng)會自動判斷設(shè)計者的偏好，并且給出更符合你偏好的迭代方案。空中客車?yán)迷撥浖320客機的內(nèi)飾隔板進行了重新設(shè)計，這種新設(shè)計的重量只有66磅，比之前的設(shè)計整整輕了45%。

人機融合

過去幾十年間，機器人已經(jīng)承擔(dān)了各種各樣的制造業(yè)工種。不過最近，有些機器人添加了一個新零件——人類。這種類型的機器人又叫“人機合作機器人”，形態(tài)不一而足，有的類似一個機器人助手，可以在人類工人勞動時將各種工具準(zhǔn)確地遞給他；有的則是像“鋼鐵俠”一樣的外骨骼套裝，人們穿了他，就會獲得額外的力量以及AI軟件的指導(dǎo)。比如寶馬公司的斯帕坦堡工廠里就有一款昵稱為“夏洛特小姐”的人機合作機器人，它主要用來安裝車門。梅塞德斯奔馳公司也在開發(fā)人機合作機器人技術(shù)，以使該公司組裝的部分奢華車型每一臺都能更加個性化。比如在使用人機合作機器人取代了體積更大的自動化系統(tǒng)后，人類工人在機器人助手的幫助下，就能更快地在大量零件中找出定制版S級轎車所需的特殊零件。麻省理工學(xué)院的教授朱莉·肖正在開發(fā)一種特殊的軟件算法，它能教會機器人解讀人類發(fā)出的信號，并使它們知道何時以及如何與人類進行溝通。有些研究人員甚至正在研究如何將人機合作機器人與人的腦電波相連。到了這一步，究竟是機器在為人服務(wù)，還是人類已經(jīng)成了機器的一部分呢？

48%

根據(jù)Mindshare公司的調(diào)查，48%的人認(rèn)為偽裝成人類的機器人“令人毛骨悚然”。

提供清潔能源

如果風(fēng)能的利用成本想要降到化石能源以下，那么風(fēng)能轉(zhuǎn)化為電能的過程就必須變得更加高效。西門子公司開發(fā)的一種機器學(xué)習(xí)技術(shù)恰好能起到這個效果。研究人員意識到，大型風(fēng)力渦輪機可以利用天氣和零部件振動等數(shù)據(jù)，不斷對自身進行調(diào)整——比如調(diào)整風(fēng)車葉片的角度等等。研究人員沃克瑪·斯特金指出，靠分析計算，是實現(xiàn)不了這樣的目標(biāo)的。

然而這對于人工智能和機器學(xué)習(xí)技術(shù)卻不是難事。斯特金指出，風(fēng)力發(fā)電機的傳感器其實早已生成了所需的參數(shù)，只不過“以前這些參數(shù)只用于遠(yuǎn)程和服務(wù)診斷，但現(xiàn)在它們也在幫助風(fēng)力渦輪機發(fā)出更多電力。”這項技術(shù)甚至可以對風(fēng)力輪機進行相應(yīng)調(diào)整，使其適應(yīng)因穿過前面其它風(fēng)力輪機而變得不可預(yù)測的氣流。

去年，西門子的風(fēng)能部門與西班牙歌美颯公司的風(fēng)能業(yè)務(wù)合并，成立了一家名叫西門子歌美颯可再生能源的獨立公司。這項人工智能技術(shù)的廣泛運用，也為該公司帶來了新的機會。

守護人類的安全和健康

很多人并不了解自己的極限。很多人要么吃得太多，要么睡得太少，或者高估了自己在一段時間里能達到的目標(biāo)。在一些小事上，這倒也無傷大雅。但在某些專業(yè)領(lǐng)域——比如長途開貨車，或是操作重型設(shè)備，如果你不知道自己的極限，后果可能相當(dāng)危險，甚至?xí)冻鰢?yán)重代價。

有鑒于此，現(xiàn)在很多公司都使用人工智能程序作為“守護天使”，以保護高風(fēng)險工種的安全。商業(yè)軟件公司SAP的高級副總裁麥克·弗拉納根介紹道，這些人工智能系統(tǒng)受過幾百小時的員工傳感器數(shù)據(jù)的訓(xùn)練，可以實時監(jiān)測工作人員的心率、體溫以及疲勞和緊張水平的指標(biāo)數(shù)據(jù)，當(dāng)員工需要休息時，系統(tǒng)就會發(fā)出警報信號，提醒員工休息。（SAP有一款安全產(chǎn)品就是這樣工作的。）

那么對于我們這些普通群眾呢？如無意外，很快我們也會在自己的汽車上看到這項技術(shù)了。目前，各大汽車廠商也在爭相研發(fā)疲勞檢測技術(shù)。目前市面上只有少輛車型搭載了防疲勞駕駛功能，功能本身也非常簡單——車子的儀表盤上會亮起一個咖啡杯狀的圖標(biāo)，提醒你應(yīng)該休息了。Nuance Communications公司的汽車創(chuàng)新管理總監(jiān)尼爾斯·蘭克表示，過不了多久，疲勞檢測以及語音和面部識別技術(shù)就將成為新車市場上的標(biāo)配。該公司目前已與多數(shù)主流汽車廠商展開了合作。

人工智能保護個人安全的三種方式

制造能夠自動選擇其目標(biāo)的武器

殺人機器人能夠識別并消滅敵方勢力，這個曾經(jīng)只出現(xiàn)在災(zāi)難科幻小說里的場景如今已不再遙遠(yuǎn)，前提是各大公司和五角大樓下決心開展這一合作。國防官員到目前為止已經(jīng)叫停了致命自主武器系統(tǒng)（政府的官方稱法）的研發(fā)。在理論上，這一系統(tǒng)能夠在沒有任何人命令的情況下發(fā)動攻擊，就像Facebook在沒有征得人們同意的情況下標(biāo)記照片中你的朋友那么簡單。

然而，用于支持進行類似攻擊的人工智能技術(shù)已經(jīng)處于研發(fā)中。五角大樓最知名的人工智能計劃Project Maven旨在使用機器學(xué)習(xí)算法，從無人機拍攝的視頻中發(fā)現(xiàn)恐怖分子目標(biāo)，協(xié)助軍方的ISIS打擊行動（據(jù)稱涉及20家技術(shù)和國防承包商，但并未對外宣布名單）。雖然開發(fā)戰(zhàn)爭用物資對于國防行業(yè)來說并不是什么新鮮事，但五角大樓正越來越多地采用硅谷在人工智能和面部識別方面的專長。雙方日趨緊密的關(guān)系最近引發(fā)了爭議。在多名員工因抗議而辭職之后，谷歌已于今夏宣布退出Project Maven。未來，各大公司是否能攬獲利潤豐厚的新人工智能國防合約，唯一的障礙可能便是其自身的意愿。

2022

牛津和耶魯大學(xué)研究人員稱，到2022年，人工智能在疊衣方面的能力要超過人類。

規(guī)避威脅

一旦預(yù)防網(wǎng)絡(luò)和現(xiàn)實生活中的攻擊以失敗告終，其代價是異常慘痛的。2017年，個人數(shù)據(jù)泄露的平均成本達到了近400萬美元。但最近攻擊的激增也并非都是壞事：它也意味著可供挖掘的數(shù)據(jù)增多了。機器學(xué)習(xí)技術(shù)數(shù)十年來一直被用于檢測攻擊模式，并過濾郵件，但諸如Barracuda Networks這樣的供應(yīng)商所提供的新系統(tǒng)居然能夠使用人工智能來學(xué)習(xí)特定公司和其高管獨特的溝通模式，從而確定可能的釣魚詐騙和其他黑客攻擊行為。在現(xiàn)實安全領(lǐng)域，連攝像頭都采用了人工智能技術(shù)，以發(fā)現(xiàn)并嘗試阻止威脅。來自于初創(chuàng)企業(yè)Athena Security的新攝像頭可識別拔槍動作，甚至自動報警。總之：我們掌握的數(shù)據(jù)越多，我們便可以更多地利用人工智能來打擊犯罪。

侵占公款的人，注意了！

如何抓住金融罪犯？像匯豐、丹斯克這樣的國際性銀行并沒有聘請負(fù)責(zé)合規(guī)業(yè)務(wù)的員工，并讓他們通過查閱上萬筆交易來尋找可疑的活動，而是更多地依靠人工智能來發(fā)現(xiàn)金融詐騙、洗錢和欺詐活動。（這一舉措最近大有升溫的趨勢，因為多家銀行因未能發(fā)現(xiàn)流經(jīng)其賬戶的非法資金而遭到了巨額罰款。）匯豐攜手人工智能初創(chuàng)企業(yè)Ayasdi實現(xiàn)其合規(guī)的自動化。在匯豐為期12周的試運行過程中，Ayasdi的人工智能技術(shù)讓正誤識（看起來可疑但卻是合法的交易）減少了20%，同時其可疑活動報告的數(shù)量與人工查驗的數(shù)量一致。

人工智能改變?nèi)藗冑徫铩⒉惋嫼蜕畹?個方式

無需親自駕駛的汽車

NBC電視劇《The Office》的主角邁克·斯考特在將一輛租來的福特金牛座（Taurus）推進賓夕法尼亞州斯克蘭頓附近的一個湖中時叫到：“這是它自愿的！”從技術(shù)方面來看，我們很久之前便已經(jīng)可以讓無人駕駛汽車在理想的路況下安全行駛，但在現(xiàn)實世界中，汽車應(yīng)多學(xué)點人類開車的方式。這是初創(chuàng)企業(yè)Comma.ai的主攻方向，該公司由臭名昭著的iPhone黑客喬治·霍茨創(chuàng)建。Comma.ai的Openpilot技術(shù)并沒有教授計算機系統(tǒng)如何辨別樹木或停止標(biāo)識，而是分析了普通駕駛員的駕駛模式，并以此來培訓(xùn)自動駕駛模型。公司從名為Chffr的行車記錄儀應(yīng)用以及一個名為Panda的插件模塊中調(diào)取了數(shù)百萬英里的駕駛數(shù)據(jù)，然后對數(shù)據(jù)進行累積，以打造能夠模擬人類司機的自動駕駛系統(tǒng)。公司的技術(shù)目前正用于本田、風(fēng)投和現(xiàn)代的部分車型上，公司將自己稱之為自動駕駛界的安卓，而將對手特斯拉的Autopilot視為iPhone。Autopilot是一個開源系統(tǒng)，聲稱自身的成功之處在于：用戶將讓其變得更好。但愿特色拉所說的用戶并不包括邁克斯·考特吧。

16%

說自己無懼乘坐無人駕駛汽車的女性的比例，而男性的這一比例為38%，來自于路透社/IPSOS的調(diào)查。

你的新旅行伴侶

事實證明，埃亞菲亞德拉冰蓋已經(jīng)伴隨我們很長一段時間了，它在火山灰褪去后便已經(jīng)存在。這座2010年噴發(fā)的冰島火山影響了數(shù)百萬的飛行員，而且它的噴發(fā)也讓旅行通信進入了新時代。在信息流功能受限的情況下，航空公司發(fā)現(xiàn)社交媒體可以作為一個有效、實時的乘客溝通方式。Accenture Interactive社交媒體和新興渠道負(fù)責(zé)人羅伯·哈勒斯表示：“一旦發(fā)生這種情況，這類通訊模式成為了一種無法阻止的力量。”然而自那之后，旅行者的數(shù)量出現(xiàn)了激增，2016年的游客數(shù)達到了12.5億人，增長30%。以人為基礎(chǔ)的社交媒體互動要達到如此規(guī)模是“不可能的”，哈勒斯說道。

讓我們來問問能夠回答旅行者基礎(chǔ)性問題的客服聊天機器人：我的航班有延誤嗎？我的酒店的退房日期是什么時候？例如，Booking.com便擁有這樣一臺機器，公司稱它可以自動回答60%的客戶問詢。該技術(shù)的下一個目標(biāo)是讓機器人了解旅客旅行的性質(zhì)，是商務(wù)還是休閑，然后再根據(jù)旅客的喜好圍繞整個旅程進行推薦，從建議航班升艙到預(yù)留最好的素食餐廳的座位，例如匹茲堡的餐廳。因此，當(dāng)前我們所說的這些聊天機器人可能很快會成為功能齊全的自動禮賓接待員。

升級呼叫中心

“需要什么幫助嗎？”到2020年，IBM預(yù)計85%的客戶服務(wù)互動在無需人工介入的情況下便可以完成。機器學(xué)習(xí)和自然語言處理讓聊天機器人、改良后的電話支持和自助服務(wù)界面能夠完成大多數(shù)人工代表可以提供的功能。

那270萬從事客戶服務(wù)代表工作的美國人怎么辦？一些可能會被部署到那些機器人無法從事的工作崗位（例如應(yīng)對怒氣沖天的客戶）。依靠這一技術(shù)的公司表示，這項技術(shù)能夠幫助消除人為失誤，大幅提升數(shù)據(jù)獲取速度，并杜絕客戶服務(wù)互動中的偏見。

不要以為這項技術(shù)的終點是機器人。瑞士投行瑞銀集團最近攜手新西蘭人工智能專家Faceme，對經(jīng)濟學(xué)家丹尼爾·卡爾特進行克隆，以便讓機器人能夠以他本人的方式與客戶進行互動。瑞銀集團表示，這個化身使用了IBM Watson人工智能技術(shù)，并由卡爾特本人親自培訓(xùn)，是該銀行探索提供“人類數(shù)字融合服務(wù)”的一部分。

IT’S TIME TO GET REAL ABOUT A.I.’S FUTURE, a subject in desperate need of discipline. The technology’s mind-blowing possibilities have apparently inebriated various seers, who take two routes to fantasyland: propagating boldly precise forecasts of jobs to be spawned and destroyed years hence, or spinning tales of A.I. transforming our world into a heaven (or hell). Instead, we wanted to confront the realities of how A.I. is changing business—minus the melodrama.

On the chief source of A.I.-induced anxiety—employment effects—the reality is that no one knows or can know what’s ahead, not even approximately. The reason is that we can never foresee human ingenuity, all the ways in which millions of motivated entrepreneurs and managers worldwide will apply rapidly improving technology. Postmaster General Arthur Summerfield predicted confidently in 1959 that mail would soon be delivered by packing letters into guided missiles, the wonder tech of the day. A growing economy meant more letters, and the future for postal workers seemed bright. It was, for a while. The possibility that mail would cease to be written on paper never occurred to Summerfield, though the necessary technologies for email, texting, and the cell network existed in rudimentary form or were being developed. We risk missing the boat in the same way with A.I.

The second reality to remember is that A.I.’s eventual uses will be determined largely by market forces. Earnest discussions of how A.I. can be directed to make the world a utopia miss that point. They recall RCA chief David Sarnoff’s long-ago prediction that the coming of color TV would enable people to see fine art in their homes. That sounded wonderful, but nobody wanted it for such high-minded uses. A.I. will be used by companies and consumers for countless practical purposes, most of them modest, and the cumulative effect can’t be foreseen. As we try to guess A.I.’s future, the key will be to think like self-interested people (including both good and bad guys) in the real world.

No bad guys here, though. These 25 examples of A.I. at work are beneficial, even inspiring—and they’re real.

How AI Is Changing the Way You Work

Getting Us All To Speak The Same Language

EVER SINCE THE GOLDEN age of the original Doctor Who and Star Trek, science fiction has highlighted devices that can automatically translate languages so that humans can talk to aliens without needing to study far-out dialects. It turns out that companies here on Earth, like Google, are using artificial intelligence technologies to create devices that can translate conversations from one language to another. While Google’s recently released Pixel Buds is a promising start, consider the ways businesses could use the technology when it works seamlessly. American executives could call up their Portuguese-speaking counterparts and brainstorm global partnerships on the fly. Businesses with international offices could more effectively communicate with employees, who could work in tandem with colleagues in other countries who don’t speak the same language. Salespeople could scout for potential leads in new regions and make cold calls that could bring about their next game-changing deal. Although many companies have instituted an English only policy as a way to keep employees speaking the same lingo, on-the-fly translation technology lets non-U.S. employees speak their mother tongues and retain aspects of their cultures—a benefit in this era of globalization.

Reading Your Mind

MOVE OVER, ALEXA. Voice control is cool, but consulting Alexa, Siri, or Cortana can be awkward and disruptive in public. Enter AlterEgo—a noninvasive, wearable device created by MIT researchers that knows what you’re going to say before you even open your mouth. The device can answer many queries within seconds, send private messages, and internally record streams of information to access at a later time—all without any observable external actions. AlterEgo doesn’t really read minds, although it may sound that way. Instead, the device effortlessly facilitates private human machine communication by interpreting electrical impulses in the jaw that are triggered when words or phrases are internally vocalized. Although university-based researchers are still in the process of collecting data and training the system, AlterEgo might also eventually serve as a platform for communication between users in high-noise environments, such as the flight deck of an aircraft or a factory floor, as well as a mode of communication for those with speech impediments. And while AlterEgo could radically speed up the process of writing, planning, and communicating, for now, humans would still be the ones stuck actually reading all those emails.

34%

Percentage of people who believed they had interacted with A.I., according to a study by Pegasystems. (Percentage who actually had: 84%.)

Hiring Smarter

THE HIRING PROCESS IS fraught with challenges. Humans may be subtly or unconsciously swayed by a last name, a college, even the font size of a résumé. Now some companies are seeing if A.I. can help.

Applicants at Vodafone, Nielsen, and Unilever, for example, play a smartphone game designed by A.I. startup Pymetrics that measures cognitive and emotional traits with an algorithm designed to avoid racial, gender, or other bias. Unilever then asks top candidates selected by the software to record a video on HireVue, answering questions about how they would handle various situations encountered on the job. Another algorithm sifts the best candidates by reviewing not just what the individuals said but also how quickly they responded and what emotional cues they revealed in their facial expressions. Those candidates who pass the early tests are rewarded with regular job interviews with a live person.

Unilever says that since it instituted the system it’s getting a higher rate of acceptances when it offers a job, and has increased applicant numbers across several diversity measures, including race, ethnicity, and socioeconomic status—and that it’s drawing from a more diverse pool at three times as many colleges and universities.

20%

Percentage of people who want their voice assistants to help them be “funnier or more attractive,” according to a study by Computerlove in the U.K.

Building The Ultimate Manager

JUDGMENT OF HUMAN behavior was once reserved for, well, humans. But increasingly, algorithms are the ones evaluating and drawing conclusions on our actions and even intentions. That’s especially true in the workplace, where HR departments are turning to A.I. for more scalable (and hopefully, more reliable) insights into possible attrition risks, attributes of high performers, and what makes teams tick. Boston-based Humanyze is experimenting with smart ID badges that track how employees interact with each other throughout the day, enabling employers to look for patterns to figure out how work actually gets done.

Textio, a Seattle startup, uses A.I. to help companies craft the right recruiting ads (the “augmented writing platform” is particularly effective at surfacing language that will attract more diverse candidates). Big companies are getting in on H-less HR too: Intel is looking at using artificial intelligence to power a new internal tool that would match employees to other opportunities within the company, all in the name of retention.

These new capabilities could help companies attract and retain the talent they need (and cut down on on-boarding and recruiting costs by automating these processes). One possible downside? They also risk alienating the very people they claim to serve—employees might not like the increasingly intrusive workplace of tomorrow.

How AI Is Shaking Up Banking and Wall Street

Meet Your New Robot Mortgage Lender

ONE THEORY HAS ARISEN in the decade since the subprime mortgage crisis: Machines may be better than humans at giving out home loans. A new Fannie Mae survey of mortgage lenders found that 40% of mortgage banks have deployed A.I.—using it to automate the document-heavy application process, detect fraud, and predict a borrower’s likelihood of default. San Francisco–based Blend, for one, provides its online mortgage-application software to 114 lenders, including lending giant Wells Fargo, shaving at least a week off the approval process. Could it have prevented the mortgage meltdown? Maybe not entirely, but it might have lessened the severity as machines flagged warning signs sooner. “Bad decisions around data can be found instantaneously and can be fixed,” says Blend CEO and cofounder Nima Ghamsari. While banks are not yet relying on A.I. for approval decisions, lending executives are already observing a secondary benefit of the robotic process: making home loans accessible to a broader swath of America. Consumers in what Blend defines as its lowest income bracket—a demographic that historically has shied away from applying in person—are three times as likely as other groups to fill out the company’s mobile application. Says Mary Mack, Wells Fargo’s consumer banking head: “It takes the fear out.”

A New Edge For Pro Investors…

IN THE WORLD OF FINANCE, there’s been such an explosion of data collected over the past decade that even those twenty-something analysts working around the clock don’t stand a chance of being able to process it all. But machines might. Bloomberg, FactSet Research Systems, and Thomson Reuters have all developed an array of data science tools and techniques—including machine learning, deep learning, and natural language processing (NLP)—to quickly unearth valuable insights for thousands of financial professionals.

Bloomberg was a pioneer of sentiment analysis (an example of NLP), which it began developing around a decade ago, in which machine-learning techniques are used to flag a news story or tweet as being relevant to a stock and assign a sentiment score. A.I. is also spreading to wealth management. Investment groups have more than quadrupled their number of “alternative data” analysts over the past five years, as asset managers scramble to unlock the potential of trading signals contained in website scrapes, language analysis, credit card purchases, and satellite data. Firms reported to be using A.I. for investment research include BlackRock, Fidelity, Invesco, Schroders, and T. Rowe Price. BlackRock, the world’s largest asset manager, has been a forerunner in adopting A.I. and is setting up a BlackRock Lab for Artificial Intelligence.

72%

Percentage of people who are afraid of robots taking over their tasks, according to Pew Research.

…And For The Amateurs Too

“ROBO-ADVISER” SERVICES, offered by startups like Betterment and traditional discount brokerages like Charles Schwab, are already using A.I. to serve the investing masses. Their low-fee investment tools rely on algorithms to determine how your assets should be split among stocks, bonds, and other assets, based on your needs and your stomach for risk. Their A.I. can automatically rebalance your portfolio; it can also nudge a (nonrobotic) adviser to call you when the algorithms predict you need help with tax strategy or estate planning.

The next frontier: A.I. smart enough to help savers make good decisions about long-term, buy-and-hold investments. Bank of America Merrill Lynch and Morgan Stanley are among the bigger players in an emerging discipline known (awkwardly) as quantamental analysis. They aim to combine the quantitative processing for which basic A.I. is best suited (basically, the capacity to spot patterns in gargantuan loads of data) with additional algorithms trained in the sophisticated analysis associated with super smart humans—assessing, say, the growth potential of an industry or the strategic acumen of a company’s management. Machine learning could eventually enable a quantamental system to learn from its mistakes. The ultimate goal: Warren Buffett–like stock-picking wisdom at low prices—and perhaps a name catchier than “quantamental.”

How AI Is Changing How We Build Things

Designing More Efficiently

SURE, COMPUTER ALGORITHMS ARE TAKING over tech and science and medicine … but the creatives are still safe, right? Not exactly. A new program from software developer Autodesk called Dreamcatcher (rendering above) can use A.I. techniques to assist human designers as they go about their creative tasks. Already in use by companies including Airbus, Under Armour, and Stanley Black & Decker, the software is an example of the burgeoning field of generative design. A designer inputs requirements, limitations, and other qualities into the program—even the total cost of materials. The software then produces hundreds or even thousands of options. As the human designer winnows the choices, the software susses out preferences and helps iterate even better options. Airplane manufacturer Airbus used the software to redesign an interior partition in the A320 and came up with a design that was 66 pounds, or 45%, lighter than the previous setup.

Melding Humans And Robots

ROBOTS HAVE BEEN ON THE ASSEMBLY LINE doing all kinds of manufacturing for decades. Lately, a new feature is being added to the automated work machines: humans. Dubbed “cobots,” short for collaborative robots, the new setups range from robotic helpers that can hand the correct part to a human worker to an almost Ironman like robotic exoskeleton suit that a person wears to gain added strength and A.I. software guidance. BMW has a cobot nicknamed Miss Charlotte that is helping assemble doors at its Spartanburg, S.C., plant. Mercedes-Benz is turning to cobot technology to help personalize each car that the luxury-automaker assembles in some of its most expensive categories. Replacing larger automated systems, humans with more nimble cobot helpers can be quicker at choosing from among the huge variety of parts needed to customize S-Class sedans, for example. MIT professor Julie Shaw is working on software algorithms developed with machine learning that will teach cobots how and when to communicate by reading signals from the humans around them. Some researchers have even looked at connecting cobots to human brainwave readouts. Mind-reading assistive robots? Now that’s collaboration.

48%

Percentage of people who found chatbots pretending to be human “creepy,” according to Mindshare.

Powering Clean Energy

IF WIND ENERGY IS TO BE decisively cheaper than fossil-fuel power, the process of transforming wind into electricity must get more efficient. Machine-learning technology developed at Siemens is helping. Researchers realized that huge wind turbines could use data on weather and component vibration to fine-tune themselves continually, for example, by adjusting the angles of rotor blades. But “you cannot analytically calculate this,” says researcher Volkmar Sterzing.

That’s the right kind of problem for A.I. and machine learning. Sensors were already generating the needed parameters, but “previously, these were used only for remote maintenance and service diagnostics,” says Sterzing. “Now they are also helping wind turbines generate more electricity.” The technology can even adjust turbines to the unpredictable airflows coming through the turbines in front of them.

Deploying this A.I. broadly is now an opportunity for Siemens Gamesa Renewable Energy, an independent company formed last year by combining Siemens’s wind operations with the wind power business of Spain’s Gamesa.

Keeping An Eye On The Mortals

HUMANS ARE NOT GREAT at knowing their own limits—they eat too much, sleep too little, and overestimate what can be achieved in a period of time. That may seem a matter of little consequence when it comes to, say, Thanksgiving dinner, but in certain professions—like long-haul trucking and heavy-equipment operation—such fallibility can be dangerous and catastrophically costly.

That’s why companies are increasingly using A.I., guardian angel–like, to safeguard employees in high-risk jobs. Systems, trained on hundreds of hours of employee sensor data, monitor conditions—like an operator’s heart rate, body temperature, and indicators of fatigue level or nervousness—in real time and signal when that individual needs to rest or take a break, explains Mike Flannagan, an SVP at business software firm SAP. (SAP has a Connected Worker Safety product that does this.)

As for the rest of us? We can expect to see this type of technology soon in our own garages, where automakers are dreaming up ways for our cars to keep an eye on us. While the tech is currently limited to a coffee cup icon that flashes on the dash in a few models, Nils Lenke, head of innovation management for automotive at Nuance Communications, an A.I. firm that works with most of the major carmakers, says fatigue-detecting voice and facial recognition technology will soon be standard in new vehicles.

3 Ways AI Is Making You Safer

Making Weapons That Pick Their Targets

ONCE THE STUFF OF APOCALYPTIC SCI-FI tales, killer robots capable of choosing and taking out our nation’s enemies are now within reach—if companies and the Pentagon decide to go that far. Defense officials have so far stopped short of developing Lethal Autonomous Weapons Systems (the government’s official term), which could theoretically strike without a human order as easily as Facebook can tag friends in your photos without your say-so.

But the A.I.-driven technology that could form the basis for such attacks is well underway. Project Maven, the Pentagon’s most high-profile A.I. initiative, aims to use machine-learning algorithms to identify terrorist targets from drone footage, assisting military efforts to combat ISIS (more than 20 tech and defense contractors are reportedly involved, though they have not all been publicly named). Although supporting war efforts is nothing new for the defense industry, the Pentagon has increasingly looked to Silicon Valley for expertise in A.I. and facial recognition. That growing relationship has recently sparked controversy, with Google announcing this summer that it would withdraw from Project Maven after several employees quit in protest. Going forward, companies’ only barrier to winning lucrative new A.I. defense contracts may be their own unwillingness.

2022

Year in which A.I. will be better than humans at folding laundry, according to researchers at Oxford and Yale.

Averting Threats

THE FAILURE TO prevent attacks in cyberspace and IRL (in real life) is an expensive line item—the average cost of an individual data breach was nearly $4 million in 2017. But the surge in attacks of late has an upside: It means there’s also more data to mine. Machine-learning techniques have been used to detect patterns and filter emails for decades, but newer systems from vendors like Barracuda Networks can use A.I. to actually learn the unique communication patterns of particular companies and their execs in an effort to pinpoint potential phishing scams and other hacking attempts. In the world of physical security, A.I. is even being used in security cameras to “see” and try to stop threats. New cameras from startup Athena Security can identify when a gun is pulled and even automatically alert the police. In short: The more data we have, the more we can use A.I. to fight crime.

Embezzler Beware!

HOW DO YOU catch a financial criminal? Instead of bulking up compliance staff to sift through thousands of transactions in search of suspicious activity, banks across the globe like HSBC and Danske Bank are increasingly turning to A.I. to flag financial scams, money laundering, and fraud. (This push has gained even more momentum recently as several banks were hit with huge fines for failing to detect illegal funds flowing through their accounts.) HSBC partnered with A.I. startup Ayasdi to automate some of its compliance. In its 12-week pilot with HSBC, Ayasdi’s A.I. technology achieved a 20% reduction in false positives (transactions that looked suspicious but were legit), while retaining the same number of suspicious-activity reports as human review.

7 Ways AI Is Changing How You Shop, Eat, and Live

Driving So You Don’t Have To

“THE MACHINE KNOWS WHERE IT’S GOING!” CRIED Michael Scott, protagonist of NBC’s The Office, before launching a Ford Taurus rental car into a lake near Scranton, Pa. Getting an autonomous vehicle to drive safely under idealized road conditions has technically been possible for a while now, but for the real world, the cars are going to have to learn to drive a little bit more like us. That’s where Comma.ai, a startup founded by notorious iPhone hacker George Hotz, comes in. Rather than teaching its computer systems what a tree or a stop sign looks like, Comma.ai’s Openpilot technology analyzes the patterns of everyday drivers to train its self-driving models. The company is pulling in millions of miles of driving data from a dashcam app called Chffr and a plug-in module called Panda, then aggregating that data to create an autonomous system that mimics human drivers. The company—whose technology currently works with select Honda, Toyota, and Hyundai vehicles—is styling itself as the Android to Tesla’s Autopilot iPhone, an open-source system that is pegging its success to the notion that users will make it better. Let’s just hope Michael Scott isn’t one of them.

16%

Percentage of women who said they would feel comfortable riding in a driverless car, vs. 38% for men, according to a Reuters/IPSOS poll.

Your New Travel Companion

EYJAFJALLAJ?KULL, it turns out, has stayed with us long after the ash faded away. The Icelandic volcano that erupted in 2010 affected millions of fliers and, in doing so, ushered in a new era in travel communications. With information flow capabilities strained, airlines discovered social media as an effective, real-time way to reach passengers. “Once that happened,” says Rob Harles, Accenture Interactive’s head of social media and emerging channels, that mode of communication “was an unstoppable force.” Since then, however, the number of travelers has ballooned, with 1.25 billion arrivals in 2016, an increase of 30%. Human-powered social media interaction on that scale is “impossible,” Harles says.

Enter the customer service chatbot that’s able to answer travelers’ basic questions: Is my flight delayed? When is my hotel checkout? Booking.com, for instance, has a bot that the company claims can solve 60% of customer queries automatically. The next stage of this technology is for bots to understand the nature of your trip—business or pleasure?—and to make recommendations based on your preferences throughout your journey, ranging from suggesting a flight upgrade to reserving a table at the best vegan café in, say, Pittsburgh. So what’s currently known as a chatbot may soon resemble a full-blown automated concierge.

Upgrading The Call Center

“CAN I HELP YOU?” By 2020, IBM estimates that 85% of customer service interactions will be handled without a human agent. Machine learning and natural language processing make it possible for chatbots, enhanced phone support, and self-service interfaces to perform most of the functions of human representatives.

As for the 2.7 million Americans who are employed as customer service representatives? Some may be redeployed to tasks that bots can’t do (like dealing with truly irate customers). Companies relying on this technology say it can help eliminate human error, drastically increase speed in data retrieval, and remove bias from customer service interactions.

And don’t think this ends with bots. Swiss investment bank UBS recently teamed up with New Zealand A.I. expert FaceMe to digitally clone chief economist Daniel Kalt to interact with clients just as he would in the flesh. The bank said the avatar, built using IBM Watson A.I. technology and trained by the real Kalt, is part of its exploration to provide a “mix of human and digital touch.”

點球成金2.0

2017年，美國國家冰球聯(lián)盟的選秀團隊看到19歲的防守隊員肖恩·德茲（上圖），并沒怎么留意。僅僅一年之后，德茲在第二輪便被選入多倫多楓葉隊。之所以差別如此巨大，是因為總部位于蒙特利爾的創(chuàng)業(yè)公司Sportlogiq開發(fā)的人工智能軟件，數(shù)據(jù)顯示出德茲強大的組織能力。這款軟件叫點球成金2.0。Sportlogiq只是利用人工智能技術(shù)幫球隊尋找新星的公司之一。澳大利亞數(shù)據(jù)分析公司Brooklyn Dynamics的聯(lián)合創(chuàng)始人卡姆·波特表示：“關(guān)鍵在于在人才未成型時及時鑒別，尋找可塑之才。”該公司曾與幾家美國職業(yè)棒球大聯(lián)盟球隊合作，還為2017年環(huán)法自行車賽開發(fā)了機器學(xué)習(xí)人工智能系統(tǒng)，收集實時數(shù)據(jù)并預(yù)測比賽結(jié)果。

Brooklyn Dynamics正開發(fā)一款應(yīng)用程序，繁忙的選秀團隊和教練可使用機器學(xué)習(xí)技術(shù)，分析潛在球員和當(dāng)前的球員，創(chuàng)建全球各地大學(xué)和專業(yè)團隊均可訪問的集中式數(shù)據(jù)庫。“這是一個獨特的工具，可成為招募人員的絕招。”波特說。“該組織的其他成員可以查看[統(tǒng)計數(shù)據(jù)]并加入前期討論，判斷哪些球員能為俱樂部帶來價值。”

改變購物方式

現(xiàn)在實體商店有了新吸引力，店面都是絕佳的人工智能數(shù)據(jù)收集實驗室。家居建材零售商家得寶就在分析數(shù)百萬筆交易數(shù)據(jù)，弄清楚顧客還需要什么東西，例如廚房整體翻新，提供詳細(xì)的家庭裝修指南以及超級精準(zhǔn)的交叉銷售。絲芙蘭利用ModiFace（最近被歐萊雅收購）的人工智能支持面部識別，幫助購物者選擇最合適的眼影。該軟件分析了數(shù)百萬歷史用戶，更好地預(yù)測適合當(dāng)前顧客的商品。 MIT-spinoff Celect利用機器學(xué)習(xí)預(yù)測購物者的行為方式，判斷商店哪些地方更適合哪些促銷活動，研究售賣哪些產(chǎn)品業(yè)績最佳。

想高效檢查完10條過道上價簽？沃爾瑪就已在50家商店測試機器人，機器人負(fù)責(zé)掃描貨架上的缺貨商品，將客戶放錯位置的產(chǎn)品放回原位，檢查錯誤價簽等。對于人類來說都是耗時繁瑣的工作。創(chuàng)投調(diào)研機構(gòu)CB Insights表示，雖然沃爾瑪對應(yīng)用技術(shù)守口如瓶，但Navii和Simbe之類制造人工智能機器人的公司非常引人注目，投資人也在密切觀察。

廣告能逗你笑嗎？

現(xiàn)在營銷人員想達成目標(biāo)越發(fā)困難，超模肯達爾·詹娜出演的百事可樂廣告就明顯效果不佳。但越來越多營銷人士依靠人工智能降低失誤的幾率。情感人工智能公司Affectiva公司表示，《財富》美國500強企業(yè)里有四分之一在創(chuàng)意開發(fā)流程中使用其技術(shù)，主要在人工智能技術(shù)支持的調(diào)查研究測試用戶對備選廣告的反應(yīng)。 Affectiva的系統(tǒng)已接受87個國家700萬張面孔（以及38億個面部框架）的圖像訓(xùn)練，解碼了個人的面部表情，技術(shù)可識別人們看到廣告一刻的20種面部表情以及8種情緒，包括“厭惡”。

2011年以來，媒體研究巨頭Kantar Millward Brown已應(yīng)用Affectiva的產(chǎn)品（鑒別了3萬個廣告），發(fā)現(xiàn)耐克廣受贊美的四分衛(wèi)科林·卡珀尼克廣告評分達到了微笑。“由此能確定，卡珀尼克關(guān)于犧牲和夢想的信息引發(fā)了積極的反應(yīng)。”該公司董事總經(jīng)理格雷厄姆·佩奇表示。他們還發(fā)現(xiàn)，觀眾對世界杯廣告中的女運動員反應(yīng)回應(yīng)，比較出乎意料。

佩奇指出，除了幫客戶提升廣告效果，Kantar也從中累積了對所有客戶有意的經(jīng)驗。該公司表示，觀眾描述為“進步”的廣告，即主角更為現(xiàn)代（而非傳統(tǒng)）時，效果提升了25%。

自己種出食物

表面上看農(nóng)業(yè)很簡單：在土里播種、澆水、收獲，然后重復(fù)。但實際上種植糧食基于一系列復(fù)雜的因素。“我們在農(nóng)業(yè)中處理的大量數(shù)據(jù)非常復(fù)雜。”室內(nèi)垂直農(nóng)業(yè)企業(yè)Plenty的聯(lián)合創(chuàng)始人兼首席科學(xué)官奈特·斯托瑞說。環(huán)境因素（舉幾個例子：氣流、二氧化碳、光照和濕度），植物遺傳以及施肥和澆水之類人類行為都是相互作用的變量。現(xiàn)在Plenty和許多創(chuàng)業(yè)公司都在用人工智能技術(shù)協(xié)助管理農(nóng)業(yè)中各種復(fù)雜決策。例如，Plenty及其競爭對手Bowery和Gotham Greens都在搭建系統(tǒng)收集和分析圖像數(shù)據(jù)，通過機器學(xué)習(xí)確認(rèn)植物是否缺氮、缺鐵或遭遇蟲害問題等，然后及早應(yīng)對。“軟件可以發(fā)現(xiàn)問題所在，而且能實現(xiàn)大規(guī)模自動檢測，人力很難做到。”斯托瑞說。

人工智能改變醫(yī)療的三種方式

讓醫(yī)療重歸人性

當(dāng)前美國醫(yī)療行業(yè)前景相當(dāng)不明朗：每年超過1200萬個嚴(yán)重診斷錯誤，醫(yī)療領(lǐng)域3.6萬億美元有三分之一浪費，預(yù)期壽命將連續(xù)三年減少（此前從未出現(xiàn)），醫(yī)生倦怠、抑郁和自殺的水平均達到頂峰。與此同時，每個人產(chǎn)生的醫(yī)療數(shù)據(jù)超過以往任何時候，舉幾個例子，可穿戴傳感器生理學(xué)、掃描解剖學(xué)、DNA測序，腸道微生物組生物學(xué)均是來源。進入深度學(xué)習(xí)人工智能領(lǐng)域，神經(jīng)網(wǎng)絡(luò)會影響各類臨床醫(yī)生，實現(xiàn)準(zhǔn)確辨識掃描片、載片、皮膚病變和眼底等，還能在衛(wèi)生系統(tǒng)應(yīng)用，促進遠(yuǎn)程監(jiān)控推廣，最終不需要實體醫(yī)院；在消費者層面，可提供虛擬醫(yī)療顧問更好地管理甚至防治疾病。這仍是人工智能整合人醫(yī)療實踐的早期階段，宣傳熱鬧卻實證寥寥。但這是我們應(yīng)對各種嚴(yán)峻挑戰(zhàn)的好機會，可以利用豐富的數(shù)據(jù)減少錯誤和浪費，并節(jié)省時間，顯著改善臨床醫(yī)生與患者的關(guān)系。

Moneyball 2.0

WHEN NHL SCOUTS looked at Sean Durzi (above) in 2017, they decided to pass on the 19-year-old defenseman. Just one year later, Durzi went second round to the Toronto Maple Leafs. The difference was powerful new A.I. software by Montreal-based startup Sportlogiq that parsed terabytes of data to uncover his powerful playmaking ability—call it Moneyball 2.0. Sportlogiq is just one of several companies using A.I. to help teams spot the next star. “It’s all about identifying talent in hidden pockets and finding that diamond in the rough,” says Cam Potter, cofounder of Brooklyn Dynamics, an Australian data-analytics company that has worked with several Major League Baseball teams and even developed a machine-learning A.I. system for the 2017 Tour de France that collected real-time data points and spit out race predictions.

Brooklyn Dynamics is developing an app that will soon allow time-crunched scouts and coaches to run machine-learning analytics on both prospects and current players, creating a centralized database that can be accessed by college and pro teams around the world. “It’s a unique tool to add to the recruiter’s repertoire,” Potter says. “Other members of the organization can then look at [the stats] and join in on the draft discussion to eventually decide who’s going to bring value to the club.”

Change How You Shop

BRICK-AND-MORTAR stores have a new calling: They’re perfect A.I. data collection labs. Home Depot is using data from millions of transactions to figure out what else you—the DIY-er grappling with, say, a big kitchen renovation—might need and provide detailed home project guides as well as hyper-targeted cross-selling. Sephora has used A.I.-powered facial recognition by ModiFace (recently bought by L’Oréal) to help shoppers select that exact right makeup shade: The software analyzes millions of other past users to better predict what will look good on you. And MIT-spinoff Celect uses machine learning to forecast how shoppers behave, determine what kinds of promotions work better in what part of the store, and figure out where products should be placed for optimal results.

As for that price check in aisle 10? Walmart, for one, has tested robots at 50 stores that scan shelves for out-of-stock items, products placed back in the wrong spot by customers, and incorrect prices —all time-consuming and cumbersome jobs for humans. Though retailers are tight-lipped about in-store tech, companies like Navii and Simbe that make A.I.-powered robots are attracting attention and investors, according to CB Insights.

Will This Ad Make You Smile?

MARKETERS DON’T ALWAYS hit the mark—we’re looking at you and that Kendall Jenner spot, Pepsi—but increasingly, they’re leaning on artificial intelligence to make those misfires less likely. Emotion A.I. firm Affectiva says a quarter of Fortune 500 companies use its tech in their creative development processes, testing the reaction to potential ads in A.I.-enhanced survey research. Affectiva’s system, which has been trained on images of 7 million faces (and 3.8 billion facial frames) from 87 countries, decodes the facial expressions of individuals—the tech identifies 20 specific ones as well as eight emotions, including “disgust”—moment by moment as they watch ads.

Media research giant Kantar Millward Brown, which has deployed Affectiva’s product since 2011 (30,000 ads’ worth) found Nike’s lauded Colin Kaepernick ad scored smiles at key points. “We were really able to pinpoint the fact that it was Kaepernick’s message about sacrifice and dreams that triggered the positive response,” says Graham Page, the firm’s managing director. They also found that viewers responded positively, and with surprise, to women players featured in World Cup ads.

Page noted that beyond helping clients sharpen their campaigns, Kantar has gained broad insights that benefit all clients. Ads that viewers describe as “progressive,” with protagonists featured in modern roles (rather than traditional ones) are 25% more effective, according to the firm.

Growing Your Next Meal

ON THE SURFACE, FARMING SEEMS LIKE a simple endeavor: Pop seeds in the ground, water, harvest, repeat. But in reality, how food is grown is built on a series of intricate equations. “A lot of the data we deal with in agriculture is very complex,” says Nate Storey, the cofounder and chief science officer of Plenty, an indoor vertical-farming enterprise. Environmental factors (airflow, carbon dioxide, light, and humidity, to name a few), the genetics of the plant, and the things we do to it, like fertilizing and watering, are all interacting variables. Now Plenty and a number of other startups are using A.I. to help manage the complex decisions that go into farming. For example, Plenty and its indoor-ag rivals Bowery and Gotham Greens are all building systems that collect and analyze data sets of images that can help identify whether a plant has an issue, like nitrogen or iron deficiency or a pest problem, through machine learning and then preemptively treat it. “The software can learn what the problems are and do it in an automated fashion at a large scale that we couldn’t individually do,” Storey says.

3 Ways AI Is Changing Healthcare

Making Health Care Human Again

THE CURRENT U.S. HEALTH CARE PICTURE is pretty bleak: more than 12 million serious diagnostic errors each year, a third of the $3.6 trillion spent attributed to waste, reduction in life expectancy for what will be three years in a row (which is unpre?cedented), and peak levels of physician burnout, depression, and suicide. That’s all happening at a time when there is more medical data per individual than ever, imagined with wearable sensor physiology, scan anatomy DNA sequencing, gut microbiome biology, just to name a few layers. Enter deep-learning A.I., with neural networks that will impact every type of clinician, from helping to accurately read scans, slides, skin lesions, eyegrounds, and more, to health systems, promoting the use of remote monitoring that ultimately obviates the need for regular hospital rooms, and at the consumer level, by providing a virtual medical coach to better manage or even prevent diseases. It’s still early in the integration of A.I. into medical practice, with far more hype than validation. But it’s our best shot to deal with all of the formidable challenges: to use the wealth of data to reduce errors and waste, and the gift of time to markedly improve the clinician-patient relationship.

超越醫(yī)生

就在過去幾年，一系列技術(shù)越發(fā)可信，當(dāng)然仍在不斷進步，即通過人工智能技術(shù)讀取放射掃描（如Imagen），識別腫瘤并跟蹤癌癥的擴散（Arterys），用視網(wǎng)膜成像檢測眼睛狀況（谷歌的DeepMind），通過“不流血的血液測試”（梅奧合資企業(yè)和AliveCor）標(biāo)記血鉀水平危險的異常，并以其他方式協(xié)助診斷甚至預(yù)測疾病等棘手問題。從歷史上看，診斷錯誤率在5%到20%之間，某些病癥的錯誤率更高，與此同時醫(yī)療系統(tǒng)因醫(yī)生短缺和倦怠承受壓力，人工智能或許能幫忙緩解。

Outsmarting Your Doctor

IN JUST THE PAST few years, there have emerged credible if still-in-the-works A.I.-powered technologies that can read radiology scans (like Imagen), identify tumors and track the spread of cancer (Arterys), detect eye conditions using retinal imaging (Google’s DeepMind), flag dangerously abnormal potassium levels via a “bloodless blood test” (Mayo Clinic Ventures and AliveCor), and otherwise assist with the tricky business of diagnosing, or even predicting, disease. Historically, diagnostic error rates have been put at 5% to 20%, though the rate is higher for some conditions, while the health care system is strained by doctor shortage and burnout—some things A.I. may be able to treat.

重塑藥物研發(fā)

醫(yī)藥行業(yè)從不乏命運起伏的案例。某種藥可能在早期研究中看起來安全，卻在大規(guī)模臨床試驗中出現(xiàn)問題，代價極其高昂。德勤數(shù)據(jù)顯示，2017年美國大型生物制藥公司的投資回報率降至令人沮喪的3.2%。這也是BERG和Roivant Sciences等美國公司，以及英國的Exscientia等都希望借助人工智能更好地調(diào)配資源。BERG與阿斯利康和賽諾菲巴斯德等大制藥公司合作，利用算法提供的臨床數(shù)據(jù)，為藥物和分子找出可能奏效的生物靶點，從而治療帕金森病等疾病。賽諾菲在分析大量數(shù)據(jù)，希望了解為什么流感疫苗對某些人有效而對其他人無效（考慮到去年嚴(yán)重的流感疫情，這是很重要的公共衛(wèi)生問題）。利用人工智能協(xié)助制藥工具仍處于早期階段。但前景很明確，將制藥研發(fā)工作集中在有希望的目標(biāo)上，避免浪費大量的時間和金錢，希望有一天能令藥物開發(fā)過程更加簡化，不管是藥企還是患者都可從中受益。

7%

非營利組織Ideall的一項研究顯示，認(rèn)為“機器人可以代替自己工作”的人力資源行業(yè)員工占總?cè)藬?shù)7%。

逆轉(zhuǎn)疾病

美國的醫(yī)療系統(tǒng)一直被批評只注重分診，卻不主動尋找更便宜也更積極的治療方法， 企業(yè)也為生產(chǎn)力損失和醫(yī)療成本暴漲付出巨大代價。Virta Health的首席執(zhí)行官薩米·因肯能另辟蹊徑，他想用人工智能防止有糖尿病風(fēng)險的患者發(fā)病，甚至在早期試驗中通過純數(shù)字平臺治療2型糖尿病。Virta為顧客安排健康顧問，努力改變顧客的生活方式，顧問會提供飲食和其他因素方面的個性化建議。數(shù)字平臺還提供數(shù)字連接工具測量血糖、酮、血壓和體重等指標(biāo)。臨床醫(yī)生了解患者預(yù)期的血糖和體重改善情況之后，工作流程中可按患者情況安排診療次序。Virta也有競爭對手，IBM的沃森健康部門和醫(yī)療技術(shù)巨頭美敦力正合作開發(fā)一款名為Sugar.IQ的應(yīng)用程序，提供類似的工具。（財富中文網(wǎng)）

譯者：Min, Feb

Reinventing Drug R&D

THE MEDICINE BUSINESS IS FILLED WITH TWISTS OF FATE. A drug may appear safe for humans in early studies with small groups of patients only to crash and burn in spectacularly expensive fashion in a large-scale clinical trial. In fact, return on investment for the largest biopharmaceutical companies in the U.S. fell to a dismal 3.2% in 2017, according to Deloitte. Which is why American companies like BERG and Roivant Sciences and U.K.-based Exscientia want to harness the power of A.I. to better deploy resources. BERG has partnered with major drugmakers like AstraZeneca and Sanofi Pasteur to use clinical data fed through an algorithm to identify promising biological targets for drugs and molecules that may be able to treat diseases like Parkinson’s. Sanofi is also analyzing huge amounts of data to gain a deeper understanding of why certain flu vaccines are effective for some people but not for others (a critical public health question considering last year’s devastating flu season). A.I. as a central medicine-making tool is still in its early stages. But the promise is clear: Being able to funnel pharma R&D efforts to the most promising targets can avoid a huge waste of time and money and, hopefully one day, lead to a more streamlined drug development process that benefits companies and patients alike.

7%

Percentage of HR employees who think “a robot could do their job,” according to a study by Ideall.

Reversing Disease

AMERICA’S HEALTH care system has been criticized for favoring triage over cheaper, proactive approaches—and businesses pay the price in lost productivity and skyrocketing health care costs. Virta Health CEO Sami Inkinen is taking a different tack, using A.I. to prevent patients at risk for diabetes from developing the full-blown disease and, in early trials, even reversing Type 2 diabetes through its purely digital platform. Virta aims to shift customers’ lifestyles by connecting them with coaches who give them personalized recommendations on diet and other factors. It also provides digitally connected tools to measure blood sugar, ketones, blood pressure, and weight. Using a patient’s anticipated blood sugar and weight improvement, clinicians can prioritize patients in their hourly workflow. Virta is not alone: IBM’s Watson Health unit and medtech giant Medtronic are collaborating on an app called Sugar.IQ that offers similar tools.