如何用AI在鴻蒙生態裏精準連接設備、人和服務？

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"萬物互聯時代，可以想象人均聯網設備會越來越多。但設備的增多不等同於方便和高效，設備互聯不易，設備“智聯”更難，最難的大概是我們如何能在設備增多的情況下享受到便利，而不用被操控多個設備、查找和使用各類應用等繁瑣的過程所困擾。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在近日舉辦的"},{"type":"link","attrs":{"href":"https:\/\/www.infoq.cn\/news\/8r3lsBU3yU1mQOdoZti6","title":"xxx","type":null},"content":[{"type":"text","text":"華爲開發者大會2021（Together）"}]},{"type":"text","text":"上，華爲給出了他們對上述問題的思考。簡單來說，"},{"type":"text","marks":[{"type":"strong"}],"text":"華爲認爲，人和設備的交互方式需要被重新定義，其中的核心利器是AI。這樣的答案聽起來並不稀奇，甚至可以說是常見的標準答案。但是別忘了，華爲正在構建鴻蒙生態，而基於這個生態底座，我們從中看到了更加“紮實”的可行性。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"關於人機交互的想象"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"讓我們先來看看，被重新定義後的人和設備的交互方式可能會是什麼樣的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"今天，健康是所有人都關心的命題，其中不少人通過健身、運動來管理身體。以“跑步”這項運動爲例，華爲設想了一個頗具“未來科技感”的交互場景——在智能手錶、智能跑步機、手機等多設備的聯動下，用戶在不知不覺中就擁有了專業的私人健身教練，並且這不是真人教練，而是基於AI的虛擬教練。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"據華爲消費者業務AI與智慧全場景業務部總裁王成錄介紹，這個“私人教練”可以根據用戶的身體狀態，制定最恰當的鍛鍊方案，在鍛鍊過程中會實時提醒用戶其鍛鍊方式是否正確，會不會對身體某個部分造成傷害，鍛鍊結束之後該如何放鬆，根據鍛鍊狀態的結果恰當的放鬆方案等等。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"而這個“私人教練”要做到如此智能，就需要獲取大量信息。比如該用戶在運動前的體脂率、心率區間、飲食和睡眠狀況，運動中也需要實時監測呼吸、心率，瞭解運動的情況如跑步時的配速\/坡度和跑姿等等。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/26\/260f97e5257e1ad12b2261a1b3367be4.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"而要獲取更多更精確的信息和數據，就意味着需要更多的設備協同。越多設備協同，就相當於圍繞用戶的傳感器會越來越多，這些傳感器不斷記錄用戶的行爲，“私人教練”才能更準確地捕獲用戶的意圖。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"更重要是的，在這個人機交互的的過程裏，用戶不需要圍繞多個設備去獲取服務、查找服務，而是設備主動來理解用戶並推薦相應的服務。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"當然，運動健康只不過是其中一個智慧場景。“我相信在鴻蒙系統的加持下，我們會有非常多的機會去構建各種各樣場景化的超級終端。”王成錄表示。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"HarmonyOS如何更精準感知消費者意圖"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"實際上，如今每個消費者身邊的設備已經有非常多的傳感器，它們可以表徵地理位置、空間信息，記錄生理指標、行爲活動等，但大家使用設備的方式是"},{"type":"text","marks":[{"type":"strong"}],"text":"“單設備”和“單維度”"},{"type":"text","text":"，而單設備、單維度無法準確捕獲消費者意圖，如果能針對場景把多個維度的信息做整合，就能夠最大限度地捕獲到消費者的意圖。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因此，在傳感器已經存在的情況下，技術上所面臨的問題是如何獲取傳感信號，如何建立多模態語義，以及如何用多模態語義跟場景匹配進而準確捕獲消費者的意圖。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"這裏面主要涉及三個核心技術點，分別是多設備傳感器信號的同步、多模態數據的標註和多模態自適應。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/3b\/3bb14a517f309c4d998885cd672fb800.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"多設備協同的第一個挑戰是複雜的設備異構性。要把傳感信號利用好，非常重要的第一步是保障所有傳感器上報信號的時序對齊。而由於設備大小不同、傳感器不同、上報信號的特徵不同等問題，要把信號之間的時序對齊挑戰很大。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"鴻蒙生態的軟總線技術恰好能夠很好地解決這個問題，軟總線的目標是無限逼近硬總線的能力，其工作原理中很重要的一點是要有軟時鐘。軟時鐘的工作原理是當工作設備組成一個超級終端，就從最強的設備的晶振上去捕獲時鐘信號，一旦時鐘信號捕獲，後續所有業務處理的時鐘將以軟時鐘爲準。當前，軟時鐘的精度已經可以做到誤差小於1毫秒，王成錄表示，HarmonyOS 3.0的Beta版本目標是做到500微秒，這樣的精度已經能完全滿足傳感器信號同步的需求。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第二個核心技術點是多模態數據的標註。數據標註一直是AI樣本數據方向很大的挑戰，爲此，華爲提出了一個創新性的互助標註方法，在時鐘同步和時間對齊的基礎上，有標註模態的數據對無標註模態的數據進行自動標註。通過這樣的方式，隨着傳感設備越多，標註的設備越多，關聯的模態會越來越多，自動標註的效率也會大幅度提升。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第三個核心技術點則與小樣本實時訓練模型有關。相比偏傳統的大規模語義訓練，傳感信號的樣本數據很少，其次，傳感器不夠穩定，無法保障超級終端裏面任何一個設備永遠在線永遠穩定，一旦發生波動就會對模型帶來較大的影響。因此，端側不能用現有的傳統AI模型來做訓練，華爲專門爲此研發出"},{"type":"text","marks":[{"type":"strong"}],"text":"端側的小樣本實時訓練模型"},{"type":"text","text":"，以解決小樣本數據波動和不穩定帶來的衝擊性問題。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"該模型的核心算法是稀疏編碼，它是一種無監督學習算法，核心在於找到一個完備的基礎向量機，向量信號進來以後利用基礎向量機的線性組合來表徵輸入信號。這種算法可以非常有效地避免由數據不穩定、樣本空間不穩定帶來的模型波動問題。如果要在端側用小樣本做實時訓練，這項技術是必不可少的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另外，樣本規模在訓練過程中有可能會無限變大，因爲它是在端側做訓練，硬件資源自然會被越佔越多，這樣一來其他應用就無法運行，所以華爲也研發了一個自適應的部署技術。簡單來講，自適應部署可以限制訓練模型消耗的硬件資源，當模型變大之後自動進行裁剪調整，保證模型在限定的硬件資源內完成訓練和部署。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"萬物互聯的統一入口"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"有了上述技術組合，傳感器的聯邦數據就能實時、準確地捕獲消費者的意圖，而下一步就是要根據意圖去找對應的服務，把匹配的動作都交由機器解決，而不是像今天這樣由消費者主動去查找、下載和安裝各種應用軟件。這裏面要做的是不斷增強隱式向量化意圖，從海量服務和應用裏找到數十個候選集，再加上顯式結構化的意圖，對候選集做進一步篩選和邏輯組裝，就可以做到對消費者意圖的精準反饋。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/3b\/3b4d249a1163c3c88953b8fcfaf2ca5f.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"實際上，HarmonyOS 2的“小藝建議”背後便是用了上述這些技術。王成錄指出，當前“小藝建議”的使用率已經很高，也正是因爲小藝能夠深入理解用戶意圖，並根據用戶的時間、地點、場景把可能需要的服務都放到小藝建議裏。這樣一來，用戶主動搜索的頻率就越來越低。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“像‘小藝建議’這樣的服務若繼續堅持下去，有希望打造一個將來面向萬物互聯的人機交互的統一入口，通過小藝就可以屏蔽各種各樣的硬件差異。”也就是說，在這個過程裏，用戶不需要在意自己是在跟哪個設備或服務交互，而只需要與小藝這個入口做交互，與此同時由於小藝的智能化，設備本身也在判斷用戶的意圖。王成錄將小藝稱作“未來萬物互聯統一的交互入口”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"與HarmonyOS的設計理念一樣，小藝也採用了完整的解耦化的架構設計，支持彈性化部署。換句話說，搭載HarmonyOS和OpenHarmony的設備，不論大小都能安裝上適配版本的小藝。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“有了像小藝這樣的部署，有了鴻蒙生態的互聯和融合，在任何一個場景下，無論你在哪一刻用哪個設備觸達，小藝都能感知到。小藝也會根據此刻觸達的方式把信息反饋給你，省去了在不同設備之間不斷配來配去、查找服務做邏輯組裝的過程。”王成錄表示，華爲希望通過重重賦能，讓真正的萬物互聯、萬物智聯時代加速到來。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"開發者如何獲取鴻蒙AI能力"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"目前，華爲已經將包括小藝在內的所有能力通過軟件服務包的方式提供給合作伙伴。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"今年，HarmonyOS Connect軟件服務包 3.0 新增支持垂類應用的帶屏設備，並升級了基礎服務包、增強服務包，以及包含小藝、服務中心、暢連等應用在內的應用服務包，開發者直接調用即可快速開發出具備智慧 AI 能力的設備，預計到 2022 年將全面適配支持豐富應用的帶屏設備。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"有了軟件包之餘，也需要考慮是否便於開發者使用。面向硬件開發者，華爲提供了一站式設備集成開發環境，支持遠程開發、按需定製、一鍵編譯和燒錄、一鍵集成服務包和設備仿真器，降低了開發者的接入門檻和開發週期，提升了開發效率，無屏設備開發週期已由 2 個月減少至 2 周，帶屏設備預計 2023 年將縮減到 2 個月以內。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下載地址："},{"type":"link","attrs":{"href":"https:\/\/device.harmonyos.com\/cn\/develop\/ide","title":"","type":null},"content":[{"type":"text","text":"https:\/\/device.harmonyos.com\/cn\/develop\/ide"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/b5\/b5506b1a995fd0798e3a39cf1a618eb2.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"面向應用開發者，華爲也推出了一站式設備集成開發環境，其中提供了超過 50 個原子化服務和卡片模板，支持多端雙向預覽、低代碼開發、分佈式模擬仿真以及分佈式調測等。據介紹，目前原子化服務開發效率已由 1 個月縮減至 15 天，明年的目標是一週，預計到2023年，只需要5天就可以完成開發。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下載地址："},{"type":"link","attrs":{"href":"https:\/\/developer.harmonyos.com\/cn\/develop\/deveco-studio\/index.html","title":"","type":null},"content":[{"type":"text","text":"https:\/\/developer.harmonyos.com\/cn\/develop\/deveco-studio\/index.html"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/29\/2966a0197eb5c34beaab24d16b78c857.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"王成錄表示，未來將不斷增強能力、增強工具，給開發者留出更多的時間和便利去做業務的創新，“把複雜交給我們，把簡便留給開發者。”"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"寫在最後"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2020年9月10日，鴻蒙操作系統正式發佈，邁出了鴻蒙生態的最關鍵一步。今年6月2日，HarmonyOS 2手機版本的發佈將鴻蒙生態構建按下了快進鍵。如今，搭載 HarmonyOS 設備數量已經超過 1.5 億。這個成績對於鴻蒙生態來說可以算是個不錯的開局。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"真正的萬物互聯時代的到來離不開AI，AI將重新定義人和設備、人和服務的交互方式。現在，鴻蒙已經在AI開發能力和服務上做了一些工作，但光靠華爲一家公司還不夠。會上，王成錄也表示，歡迎更多合作伙伴和開發者加入鴻蒙生態，持續添磚加瓦，開啓萬物智聯的新時代，這也是鴻蒙生態從誕生起就肩負的使命。"}]}]}