踏入2025年，中國人工智能（AI）領域取得標誌性突破。深度求索（DeepSeek）推出的R1模型，以557萬美元的超低訓練成本（僅為GPT-4的5%）媲美美國OpenAI o1模型的性能表現。

從「追趕」到「並跑」的轉折

R1模型這一成就的關鍵在於其獨創性的科技路徑：利用純強化學習技術擺脫對監督微調數據的依賴，藉助「群體相對策略優化」（Group Relative Policy Optimization）演算法實現推理能力的自主進化。與此同時，其動態精度調節技術有助華為旗下昇騰AI平台的算力成本降低70%，而性能損失則只低於5%。尤其觸目的是，R1的開源模式打破了算力壟斷，支持開發者將大模型能力蒸餾為1.5B參數的小型版本，因而在東南亞、中東等新興市場迅速滲透。

美國OpenAI對此迅速作出回應，上月緊急發布了o3-mini推理模型。該模型主打可調節推理強度，在高強度模式下對科學、科技、工程、數學（STEM）領域展現出無可匹敵的統治力——在AIME 2024數學測試中取得87.3%的準確率，勝過R1的79.8%。其物理類比能力也被開發者譽為「教科書級別」，例如在四維超立方體內反彈小球的程式設計中，o3-mini對幾何結構與運動軌跡的分析精度高達97%。然而，其輸入輸出成本仍是R1的8倍，凸顯出閉源模式的效率瓶頸。

生態構建與地緣滲透

中國市場的獨特優勢體現在場景驅動的生態整合上。透過開放應用程式介面與行業定製策略，國內企業已深度嵌入醫療、教育、金融等垂直領域。以教育界為例，作業幫的大型語言模型通過逐步分析，提升中學生數學解題的效率達37%。這種本土化適配能力預計將推動內地AI大模型市場在2026年突破700億元人民幣，使用者每周平均使用頻率達4.5次，遠超歐美市場的2.1次。

美國企業則試圖依靠多模態AI功能穩固霸主地位。GPT-4o通過文本、語音和視覺的融合能力，在微軟Office智能助手中自動生成會議紀要和簡報排版等功能，企業用戶續費率高達89%。另一方面，o3-mini在非英語市場的表現卻顯露不足，在中文理解一環的準確率只及百度文心一言的82%，拓展東南亞市場的計劃也就舉步維艱。更為嚴峻的是，OpenAI預計2023至2028年的累計虧損將高達440億美元；公司行政總裁阿爾特曼（Sam Altman）罕見地承認「閉源策略可能犯下歷史性錯誤」，並啟動o3-mini免費開放，以吸引開發者。

美國科技霸權重構

近年來，美國在AI大模型領域憑藉閉源商業模式和專有技術，一直維持國際領先地位；然而，中國 DeepSeek-R1冒起，馬上引發一場嶄新的科技變革。R1模型通過開放創新的方式，達至核心演算法和模型共用，有利於前沿技術迅速反覆運算，同時也大幅降低科技壁壘。這一舉措對美國長期以來依賴封閉體系構建的科技霸權造成直接衝擊。

DeepSeek-R1 的開源，使環球科研機構、企業和獨立開發者能夠在一個開放透明的平台上，不但共同改進和應用AI技術，而且逐漸重塑科技標準和行業規則。各方藉助這一開源平台進行研發合作，加速達至創新成果和構建自主智能知識產權，從而開創國際科技生態中多極互動的新局面。這種模式正在促使科技管治由單一壟斷轉向開放協同，並為國際社會建立科技公平競爭的規則樹立典範。

面對異軍突起，美國政府與科技巨頭不得不調整其基於閉源模式的科技霸權體制。為了維持競爭優勢，美國正在探索如何在保障商業機密的同時，更加積極地參與制定國際規則與拓展跨國合作，致力平衡開放創新與保障所有權之間的關係。DeepSeek-R1的成功實踐表明，開源能夠釋放創新活力，亦能加速全球科技生態的健康演進，進而對美國的科技主導地位產生實質性挑戰。

香港充當科技冷戰的安全地帶

顯而易見，AI的迅猛發展和大規模模型在各行各業的廣泛應用，催化中美之間的劇烈競爭，呈現科技冷戰的危機。在這一背景下，特區政府應聚焦其國際金融中心、獨立法治與開放市場的制度優勢，著力打造一個中立監管平台連接中美雙方，而成為緩衝調節的試點城市。事實上，為確保AI既能安全發展，又能保持公平競爭，香港金融管理局現正和科技企業、研究機構攜手推出生成式AI沙盒，構建一個涵蓋模型安全、數據隱私、倫理審查與風險預警的專屬平台，各類企業和監管機構可在受控、獨立的環境下對AI演算法、數據處理流程和應用場景進行全面測試，及時識別與化解潛在風險。此一項目可為中美兩國的科技對話提供實證支援。

此外，特區政府積極搭建國際對話平台，主動與歐美、東南亞及中國內地的監管機構建立常規化的合作機制，全力在中美科技冷戰中發揮避風港的作用。通過定期舉辦以AI倫理、數據管治與風險管理為主題的政策研討會、專題論壇及雙邊交流，各方圍繞AI應用規範、模型透明度及跨境數據流動等熱點議題，進一步展開溝通與協作。除了有助於平衡各國利益、推動科技與監管同步向前，這種開放協商模式也為國際統一AI監管標準提供寶貴的香港實踐藍圖，進一步鞏固特區在全球科技管治中的樞紐地位。

開源與閉源博弈下的展望

在全球AI領域風雲變幻、競爭白熱化的時代，中美大模型的爭鋒既展現了AI技術的無限創新潛力，亦暴露了開源與閉源兩種模式各自面臨的風險與機遇。中國依託開源協作、演算法優化和務實創新不斷突破，而美國則通過閉源策略、全能整合及高性能應用牢牢佔據部分關鍵市場。這種互為競爭且互補的格局，正驅動全球科技標準和監管制度迅速演變，挑戰傳統管治模式。
憑藉作為國際金融中心並具備健全法制與開放市場的強項，香港特區政府宜利用跨境監管沙盒、多邊對話和區域合作，構建全鏈條技術監管制度，為應對中美大模型競爭提供全新管治思路。唯有政府、業界與國際監管機構協同創新，在開源活力與閉源保障之間取得平衡，香港才更能用好其豐富經驗和前瞻探索，在環球多極競爭中搶佔先機，為未來數碼生態管治創立務實的範例。

章逸飛博士

港大經管學院經濟學高級講師

（本文同時於二零二五年二月十二日載於《信報》「龍虎山下」專欄）

特區政府近6個財政年度中，有5個出現赤字，累計超過5千億元【圖】。面對此一難題，財政司司長將於本月底發表新一份《財政預算案》，有何妙策可以應付？

2024–2025年度首8個月的赤字為1,432億元（已包括發行債券收入）；羅兵咸永道估算全年赤字為948億元，而財政儲備將下降至6,398億元，相當於約10個月政府開支，屬有記錄以來最低水平，最高峰曾達28個月水平。若公共財政繼續入不敷支，並無措施出台扭轉趨勢，估算未來4年赤字亦將居高不下。政府必須盡快全面檢討支出，同時在管理財政及其對公眾的影響之間小心取得平衡。

圖  香港特區政府的財政狀況

（億港元）

資料來源：香港特區政府庫務署

以公債挹注基建

政府近年推出銀色和綠色債券，並積極增幅以應對財赤。去年2月陳茂波發表預算案時，提出2024年度將發債1200億元，其中零售部分為700億元（500億元為銀色債券、200億元為綠色債券及基礎建設債券）。當時訊息非常清晰，500億元銀債與基建債計劃無關。

發售銀債旨在為本港長者提供回報穩定的投資產品，雖沒有二手市場，但投資者可以在債券到期前讓政府提前贖回。值得留意的是，2024年9月根據基礎建設債券框架發行的第九批銀色債券，首度將債券資金用於基建工程。政府認為可更好地管理大型基建所需現金流，令惠及經濟民生的項目早日落成。預計2028–29年度，債務對本地生產總值比率介乎9%至13%。

過去發行的銀色債券均由金融管理局（金管局）負責投資，如去年投資回報有5.5%，稍高於銀債息率，政府不能償還債券的機會極微，但本財政年度開始，卻直接將銀債的資金用於政府公務工程，投資回報變得複雜，風險比以前高。縱使目前政府違約的機會不大，然而隨着發行債券的金額倍數擴大，風險也會大增。無獨有偶，目前銀色債券的發行額度正與財赤增加而同步擴大。例如2020–21年度財政赤字2,325億元，銀債就從此前每年發行30億元躍升至240億元，以致表面上錄得盈餘294億元。

回溯政府在2016年首次發行銀債，目標之一是協助長者抵抗通脹，但近年本港通脹偏低，如2023年的通脹率只是1.7%。除了發行首年，銀債息率一直高於通脹率，尤其是2021年的通脹率僅為0.6%，銀債息率卻達3.5%；認購量又由最初5手大增至去年21手，難免有以公帑補貼長者投資回報之嫌。筆者認為，公帑理應集中支援最需要照顧的弱勢長者。銀債推出已近10年，究竟如何能達到當年推廣長者投資市場的本意，值得深思。

千億債券救近火

綠色債券是政府可持續債券計劃的重要組成部份，為應對氣候變化、轉型至低碳經濟體等理念的綠色項目融資，反映了香港與綠色債券市場的國際標準接軌。本年度預算案列出，綠色債券計劃與基礎建設債券計劃的合共借款上限為5千億元，全部撥入基本工程儲備基金。根據預算案的中期估算，政府2025–26年度可轉虧為盈，盈餘63.3億元，但若扣除發債收支，則至2027–28年度才會錄得盈餘約141億元。

銀債在2020年起發行規模由30億元增加至150億元，翌年更倍增至300億元；同一時間在特區政府帳目中來自債券發行的淨收益，由2019–20年度首次有63億元年收入，隨後兩年倍增至193億元及291億元，可見認購銀債金額全數撥入財政儲備。上年度銀債發行額達550億元新高，並同時發行綠色債券，以致債券淨收入高達716億元。顯而易見，沒有這筆龐大收入，就會錄得1,700億元財赤，而非只是1千億元。政府不應將各類債券所募集的資金視為收入，更不應在同年將債款花掉，公眾才得以了解財赤的嚴峻程度。

新一批銀債集資不再放在外滙基金投資，意味着政府潛在收入減少，財赤亦會比預期多，加上賣地收入減少，財政儲備恐怕更快耗盡，政府有必要及早設法開源節流。金管局稱，未來會繼續按基礎建設債券框架發行銀債，或就合適基建項目發行綠債，視乎屆時的公共項目而定。在財政儲備將要跌穿6千億元的情況下，發行基建債券融資不能單靠儲備作支持，發債成本應與基建項目回報掛勾，但基建投資屬長遠性質，北部都會區、交椅洲人工島發展談不上何時能提供收益，恐怕融資成本將遠高於發行銀債的息率。政府應詳加解說，以釋公眾疑慮。

經濟需另譜新章

結構性赤字其實有好有壞，例如美國列根時代減稅所引發的財政不平衡，在不同程度上有助於經濟結構提升和改革。相反，日本的結構性財赤更關乎經濟政策出現問題，特區政府應引以為鑑，設法推出減低支出、提升收入的政策，否則單靠發債度日，難免有債務危機。政府亦需考慮，持續性發債會否影響金融市場的穩健性及評級。香港一如新加坡和挪威，是世界上極少數擁有積累財政儲備的經濟體，然而透過開源節流來削減財赤，實在談何容易！經濟不景時開徵新稅，必然備受社會反對，甚至因被視為有違簡單稅制的優良傳統，結果得不償失。至於削減公共開支，則會遇到既得利益者的阻力。教育、醫療、社會福利開支，極其量只有望凍結增長。

另一燙手山芋是公務員薪酬。筆者不建議公務員全面減薪，因每每引致私人機構效法，亦會進一步受到打擊消費者和投資者信心，而影響疫後經濟復甦。當然，現時19萬名公務員的龐大架構中，可刪減部分職位以便節流。政府不妨帶頭推動人工智能，提升工作效率，此舉更能釋出正面訊息。發債以外，政府亦可考慮下調股票印花稅的稅率，吸引更多資金流入股市，並制定吸引高端消費遊客的政策，以收開源之效。

財政赤字令特區政府忽略或擱置長遠的策略，無疑是非常危險的傾向。政府的本能反應是削減開支、增加收入。歸根究柢，解決方案還需依賴經濟增長。本港經濟的核心問題，在於地產市道崩潰之後，缺乏新的火車頭和增長動力。這既屬地產泡沫的後遺症，也是外在環境（特別是中國因素）出現重大轉變使然。根據金管局的研究報告，香港的潛在本地生產總值增長率從1980年代初的8%，跌至1998至2000年的3.5% 至4%，反觀南韓、中國台灣及新加坡的增幅卻達4%至6%。

香港經濟因疫情低迷，疫後經濟復甦緩慢，並失去發展方向。國內城市冒起，更迅速動搖香港的自信。特區的生產性投資、效率及人力資源質素，整體難以令人樂觀，加上供求錯配，這座城市的潛在增長率正逐漸下降。這必然影響政府財政收入的穩健性（revenue buoyancy）。由於開支方面出現名義剛性（nominal rigidity），財政赤字便會惡化，單以削減開支、增加收入的傳統方法來平衡財赤，就難以奏效。除非能顯著地提高生產效率，否則財政赤字只會揮之不去，更可能引致惡性循環。如何令經濟發展有所創新，重拾增長動力，才是關鍵所在。

參考資料

Jiming Ha and Cynthia Leung. “Estimating Hong Kong’s Output Gap and Its Impact on Inflation.” HKMA Research Memorandum, November 2001.

謝國生博士   港大經管學院金融學首席講師、新界鄉議局當然執行委員

何敏淙  香港大學附屬學院講師

院線大戰客廳沙發：香港戲院如何扳回一局

從新冠肺炎疫情至今，香港電影院經歷了一波關門大潮。截至2024年4月30日，累計16間戲院結業，其中包括UA連鎖影院在2021年3月全線結業。【註1】這波關門大潮似乎還遠未結束，上月12日MCL Cinema Plus+荷里活廣場戲院宣布結業，而以播放小眾電影知名的高先電影院，也傳出將於租約期滿後結束營業。【註2】

電影院紛紛倒閉並非香港的獨有現象，在作為電影大國的美國，同樣難逃厄運。根據英國一家研究機構統計，目前美國的影院銀幕總數為36,400，比2019年減少12%。世界最大的影院公司AMC娛樂，從2019年開始就陸續關閉了169家影院，雖然疫情之後又增加了60家，但是規模遠比疫情前要小。至於全球第二大連鎖影院巨頭Cineworld，也關閉了接近15%的影院。

影院頭號勁敵

衝着戲院而來的最大對手，無疑是Netflix、Disney+等串流媒體。比起動輒花上百元購買1張門票，同樣是大概100元的串流服務月費，卻可換來豐富的電影、電視節目選擇，對於消費者來說自然更加划算。加上用戶可以舒舒服服地躺在自家沙發上，隨時隨意暫停播放或重播，不難理解為什麼愈來愈多人選擇串流服務來看電影。

無庸置疑，新冠肺炎來襲令影院市道雪上加霜。疫情期間本港戲院共停業267日，市民被迫呆在家中，通過串流媒體觀看電影電視。3年多下來，消費習慣逐漸改變。據《華爾街日報》報導，疫情前有60%的美國消費者認為，看電影應去戲院，疫後降至只有35%；同期選擇在家看電影的美國人，則由40%增至65%。【註3】

作為影院的合作方，電影製作公司亦轉投串流媒體的懷抱。荷里活各大電影製作公司為了既在院線大銀幕上賺足戲票，又想在串流媒體上分一杯羹，通常會在新電影發行的首90天，授予電影院獨家放映權。【註4】消費者如果迫不及待先睹為快，就只能花錢進戲院。等到這個影院獨家檔期過去了，串流媒體才可加以播放。

在串流媒體盛行的今天，獨家檔期對於影院的重要性也就不言而喻。然而疫情令所有影院停運，電影製作公司只好將影片直接在串流媒體上發行。儘管獨家檔期在疫後恢復，但大幅減半至只有45天。盈利窗口大大縮短之際，經營成本又告上升，經營影院生意可謂舉步維艱。

力挽狂瀾之道

影院能否贏過客廳沙發？雖然串流媒體優勢明顯，影院卻可提供無法取代的觀影體驗。如何放大這一強項，是影院能否打贏家中沙發的核心關鍵。比方說，影院的超大銀幕和震撼音效能讓觀影體驗全面升級。在影院總體經營困難的情況下，大銀幕IMAX加杜比音效卻逆勢增長。2023年IMAX的年收益上漲25%，達到2億1400萬美元，其全球（除中國內地）票房收入的市場份額也從2019年的1.7%升至2.1%。【註5】為了強化這一優勢，美國德克薩斯州具百年歷史的連鎖電影院B&B，甚至安裝了一塊寬7層樓、高4層樓的超大銀幕以及一塊270度的環形銀幕，讓觀眾完全沉浸在電影特效之中。

即便是座椅，也有不少影院想勝過家裡的沙發。為了提高觀影體驗，一些影院不惜重金設置可調整溫度和傾斜度的座椅，甚至有服務員端上餐飲，觀眾們一享眼福之餘，還能一飽口福。阿聯酋的連鎖影院Vox更提供和牛豪華大餐，讓看電影變成奢華享受。另有創新者，如在亞洲擁有連鎖影院的韓國公司CGV，不僅在電影放映中為觀眾帶來風雨以及氣味的4D體驗，還在放映前提供音樂、遊戲、餐飲一條龍服務，把觀影塑造成獨特文化體驗。

影院與客廳沙發能否握手言和

電影分類眾多，一個很簡單的標準就是片長。《經濟學人》分析了從荷里活的黃金年代1930年至今10萬多部在全球放映的電影，發現平均片長的增幅為32%，從1930年代平均1小時21分鐘增至2022年的1小時47分鐘。這種趨勢在熱賣的電影中尤其顯著；2022年IMDB網站的評分排名中，十大電影平均片長為2個半小時，足足比1930年代的十大電影長了50%。【註6】

對於觀眾來說，這樣的長片很難集中精力盯着家中的小熒幕一口氣看完，而斷斷續續地看又讓觀影體驗大打折扣。反之，影院的大銀幕配合好音效，若再提供附加服務，就足以吸引消費者步出家門，走進影院，好好享受電影時光。電影製作公司的這種加時策略，在2019年漫威電影《復仇者聯盟：終局之戰》中取得豐厚回報。這部長達3小時的超級英雄電影成為2019年票房冠軍。近期北美電影票房排行榜就有不少這類長片，其中包括長達3小時的2024年奧斯卡最佳影片《奧本海默》。

這意味著電影院和串流媒體可以針對不同影片細分市場：影院專門放映長片或大製作，串流媒體則主攻其他類型的作品；各有側重，和氣收場。

影院應為跳出舒適圈另謀對策

上電影院和在家看戲還有一個很大的不同，就是在影院內和朋友或陌生人坐在一起，這就使得影院有可能成為社交娛樂的場所。影院不妨跳出電影的局限，另闢蹊徑。比如AMC影院正在考慮直播美國欖球聯賽；美國連鎖電影院B&B不僅貼心地為小朋友提供遊樂場地，還為大人提供保齡球、攀岩、匹克球、酒吧以及舉辦私人宴會的場所。其他影院為了鼓勵單身男女入場，甚至在放映前的等待時間組織配對約會活動，讓害怕一個人看電影的年輕人多了一個走進影院的理由。

電影院PK客廳沙發的大戰尚未結束。戲院如果不想舉手投降，只能變陣迎戰，由「觀影場所」打造成「升級體驗的休閒社交場所」，方能有望在這場沒有硝煙的戰役中收復失地。

註1：https://www.hk01.com/%E7%A4%BE%E6%9C%83%E6%96%B0%E8%81%9E/1015167/%E6%88%B2%E9%99%A2%E7%B5%90%E6%A5%AD-%E7%96%AB%E6%83%85%E8%87%B3%E4%BB%8A16%E9%96%93%E5%80%92%E9%96%89-ua%E5%85%A8%E7%B7%9A%E8%90%BD%E5%B9%95-%E5%8C%97%E4%B8%8A%E6%BD%AE%E6%AF%94%E9%98%B2%E7%96%AB%E6%8E%AA%E6%96%BD%E6%9B%B4%E5%85%87

註2：https://www.hk01.com/%E9%9B%BB%E5%BD%B1/1084551/%E9%AB%98%E5%85%88%E9%9B%BB%E5%BD%B1%E9%99%A2%E7%AA%81%E5%82%B3%E6%BB%BF%E7%B4%84%E5%BE%8C%E7%B5%90%E6%A5%AD-%E6%9C%89%E8%82%A1%E6%9D%B1%E6%9B%BE%E8%A1%A8%E7%A4%BA-%E7%94%9F%E6%84%8F%E5%94%94%E4%BF%82%E4%B8%8D%E5%A5%BD-%E8%80%8C%E4%BF%82%E5%B7%AE

註3：“How going to the movies is changing, in charts,” The Wall Street Journal, August 6, 2024.

註4：“Movie-theater industry pain intensifies even as pandemic eases,” The Wall Street Journal, April 4, 2023.

註5：“Cinemas may be dying. But IMAX and the high end are thriving,” The Economist, February 28, 2024.

註6：“Why films have become so ridiculously long,” The Economist, October 14, 2023.

范亭亭
港大經管學院市場學首席講師

（本文同時於二零二五年一月二十二日載於《信報》「龍虎山下」專欄）

特朗普2.0啟幕

還有五天，便是特朗普2.0時代的正式開始。8年前，特朗普以半冷門姿態在大選中贏了希拉莉，在缺乏政壇經驗下登上總統寶座。就職典禮後，他還在着意於出席人數的多少。然而，這次他卻有備而來，當選後迅即組成內閣，並高調地提出一些將要推出的政策，如高關稅和擴張領土等。

他同時又接見多位外國領導人及大企業總裁，海湖莊園一時冠蓋雲集，重要人物絡繹不絕，一番未就任已指點江山的氣象，無視現任總統拜登。特朗普的得力助手馬斯克亦不甘後人，公然稱德國總理肖爾茨為蠢才，並公開支持極右的另類選擇黨。他又和一些政客盤算如何在英國下次大選前促使現任首相工黨的施紀賢下台，似乎在幫助特朗普成功競選後，愛上了造王者（kingmaker）的角色。這些出位的言行，預告了未來數年的國際風雲變幻。

特朗普2.0的政策目的仍然是「讓美國再次偉大」（Make America Great Again, MAGA），做法是「美國優先」（America First），政策範圍則包羅甚廣。美國甚為保守的智庫傳統基金會（Heritage Foundation）在2023年出版了「2025計劃」（Project 2025）報告，詳細列出不同範疇的政策執行方向。這份長達900頁的報告，有100多個保守主義機構和一些前特朗普政府的官員參與，雖然特朗普口口聲聲說報告和他無關，但可說代表了保守派主流思想。2016年特朗普競選總統時傳統基金會也有類似報告，並認為特朗普首兩年任期內已跟隨該報告的64%建議。

在經濟領域上，特朗普2.0比較確實的政策範圍包括減稅、減少貿易逆差和減少進入美國的移民。特朗普於2017年成功在國會通過「減稅及職位法案」（Tax Cuts and Jobs Act），但其中多項條款只是臨時的，將於本年底失效。若沒有進一步的立法行動，稅率便會在年底回升至之前的水平。

特朗普一方面希望有關的低稅率可以持續，另方面亦想進一步減稅。上述的「2025計劃」的建議是將現有稅率推倒重來，新的個人所得稅分15%和30%兩種，企業所得稅則減至18%。有批評指出，這會增加低收入人士和減少高收入人士的稅務負擔，增加了稅後收入分配的不均。由於共和黨掌控了參眾兩院的多數，成功減稅的機會比較大，會為美國經濟帶來正面的刺激作用，同時也會提高美國企業的競爭力。

大削政府開支勢添民怨

另一方面，特朗普2.0亦計劃削減聯邦政府的財政開支，提高政府部門的效率。馬斯克曾經稍為隨意地提到，會減少政府開支2萬億元（美元下同），但這難以實現。在去年9月結束的財政年度，美國政府開支為6.75萬億元，2萬億元即佔總開支的30%。再看看開支中必須支付的，包括社會保障、醫療和國債利息，已達4.13億萬億元。餘下的再減2萬億便只有0.62萬億，連該年度的國防開支也不足，更無經費用於教育、基礎建設、退伍軍人福利等多個社會民生項目。當然，若要節省2萬億，肯定會影響政府運作和引發民怨。

特朗普2.0的減稅規模和幅度有多大仍未清楚，但無論如何，都會進一步增加美國國債。即使包括多次提及的對外國產品增加徵收關稅所得的財政收入，亦難以抵消削減所得稅的財赤效果。基於美元的「囂張的特權」（exorbitant privileges），美國政府已習慣了對財赤和國債的持續上升視若無睹，只要去美元化還未認真地打擊到美元在國際金融體系中的核心地位，美國兩黨仍會繼續透支未來。

特朗普及其顧問團隊常認為進口是給外國賺自己的錢，美國的貿易逆差是外國找美國的便宜。要減少貿易逆差，一是徵收關稅，另一是從匯率入手，迫使外國貨幣升值，兩者都會增加美國消費者購買外國產品的美元價格，減少他們的進口。但關稅可以選擇性地加於某些特定商品，因而也是制裁外國或保護本地產業的手段。

特朗普早前曾揚言要對中國產品加徵60%關稅和全球其他國家產品10%甚至20%的關稅，其後又以25%關稅威脅加拿大和墨西哥，要它們限制芬太尼和非法移民進入美國，最近又說以100%關稅對付那些要去美元化的金磚國家。要控制巴拿馬運河和格陵蘭島，也用關稅威脅巴拿馬和丹麥。

對眾多外地徵收大幅關稅，無異於減少外貿，趨向孤立主義，同時外國亦會還以顏色。全球分工已落實到零部件的層面，特朗普的多方面關稅容易誤中副車，亦會如迴力鏢般害人害己。到頭來美國的貿易或會趨向平衡，但那只是貿易收縮下的平衡，貿易帶來的增值皆付諸東流。首當其衝的美國企業或會進行多方面的游說，或以各種理由爭取豁免貿易限制，甚至因為競爭力不足而退出市場。估計美國會在特朗普2.0開始時對一些國家選擇性地加徵關稅，但稅率不會如傳聞般高。

在目前的環境下，美國難以迫使外國貨幣升值。一來面對美國的高關稅，正是需要讓貨幣貶值來抵消關稅，將貨幣升值就如送羊入虎口。二來目前全球眾多經濟體都處於增長疲弱情況，貨幣升值是倒行逆施。這情況和1985年的廣場協議時不同，當時的日本經濟如日中天，每每威脅着美國的領先地位，有條件讓日圓升值。當然，日本央行隨後的誤判和無效的政策，導致經濟呆滯多年，但那是後話。

加徵關稅將推高美國物價

反過來說，美國是否可以單方面讓美元貶值來促進出口？首先，直接影響美元匯價的是聯儲局的貨幣政策。在現有制度下，聯儲局毋須聽命於白宮。此外，美國的貨幣政策目標是2%的通脹率和最高就業的雙重使命，沒有匯率目標的考慮。換句話說，聯儲局實施貨幣政策，一貫以來都不是要調控美元匯率。若特朗普硬要更改貨幣制度的安排，自然會給美元及美元資產市場添上濃厚的不確定性和不透明性，美元亦會自然貶值，但此舉破壞性很大。

過去幾個月美國的通脹率似乎有很大黏性，雖然通脹率不很高，但遲遲未接近2%的目標，聯儲局減息的步伐也慢下來，似乎短期內改變不大，特別是特朗普2.0若真的加徵關稅，也會推高美國物價。

以特朗普唯我獨尊的性格，大概無論美國經濟如何如他所願，亦不會是他心目中的MAGA，因為世界上有一個足以與美國抗衡的中國，這也是他在2018年開始啟動貿易戰打擊中國的原因。然而美國並沒有贏得貿易戰，甚至在不同準則下都敗下陣來。隨後拜登政府對中國的多方面打壓都不成功。

如今特朗普再度上台，為了增加勝算，便巧取豪奪地增加美國的資源。主要是土地，和隨之而來的物資和戰略位置，於是有了特朗普2.0的啟幕，國際政治一下子回到二戰前的亂局。也許戰後的80年是歷史的偏差，規則制定者最終也是規則破壞者。

一個被法庭定罪的重犯（convicted felon），曾經在4年前挑動政變、兩次被國會彈劾、謊話連篇的政客，可以第二度登上全球最高權力的位置，應該是超乎政治學倫理和想像的異數，但特朗普2.0並非個人現象，在他後面有7700多萬投票給他的美國選民，以及眾多攀附的政客、資本家和科技巨頭。歷史不會終結。

陸炎輝博士
港大經管學院榮譽副教授

（本文同時於二零二五年一月十五日載於《信報》「龍虎山下」專欄）

行爲經濟學之效：從致命工業意外說起

致命工業意外在香港時有發生，2024年上半年共11宗，較前年同期有所增加。勞工處公布，這些意外中有5宗關乎建造業，2宗與製造業有關，餘下4宗則涉及運輸、倉庫、郵政及速遞服務。建造業在職業安全健康方面一向存在較多問題，該5宗相關致命意外關乎有害物質、有人受困、被墮下的物件擊中等。

有立法會議員關注建造業界的安全意識薄弱，爲趕工而導致意外。勞工處指出，業界貪快、貪方便、「要錢唔要命」的弊病頗爲常見；如果做足安全措施，大部分意外本可避免。

經濟學假設人都是理性的，每個人的行爲應對自己最有利。要防止一件事情發生，就應該加大其成本，讓决策者加以避免。工業意外牽涉到雇員和雇主雙方；就雇員而言，如果安全措施沒做足，因工受傷甚至喪命，難道成本還不够大、還不應加倍小心嗎？

至于雇主方面，除了承擔賠償責任，還要被迫停工、接受監控、暫禁投標等。立法會2023年通過修訂《職業安全及職業健康法例（雜項修訂）條例草案》，最高罰款額由50萬元大增至1000萬元。加大成本難道還不足以讓公司警覺幷作出改善嗎？實行更嚴厲的法規、加强教育、監督雇員，對雇主而言，减少致命意外應該是收益大于成本，爲什麽情况未見明顯改善？

疏忽在所難免

近年興起的行爲經濟學認爲，人在决策、行動時，常常不够理性，幷沒有做出對自己最有利的選擇。認知科學、心理學爲這一說法提供了很多證據。美國心理學家Christopher Chabris和Daniel Simons曾經進行一項廣爲人知的實驗，其中安排6名學生互相傳遞籃球，然後將相關視頻播放給觀察者，要求他們點算其中3名穿白色上衣的學生總共傳球多少次。球來球往之際，有個身穿大猩猩服裝的人走到球員中間，捶胸頓足一番後便離開。這件匪夷所思的事情，竟有一半觀察者完全沒有注意到。短短數分鐘傳球過程中，舞臺背後的布景顔色瞬間改變，更是幾乎沒有觀察者留意得到。這一實驗後來被稱爲「有史以來最著名的心理學演示之一」，兩位研究者還因此獲頒搞笑諾貝爾獎。

有人說這是人爲製造的場景，你讓人去數傳球的次數，他當然不會注意有沒有大猩猩走過。問題是，我們在日常生活或决策過程中，雖然幷無受命要專注什麽，却自然而然只會專注某些事而忽略另外一些事。上述兩位學者亦曾在街頭進行另一實驗，其中參加者假裝在陌生的城市迷了路，向路過的人問路。在對話中安排了另外兩人抬著門板，强行從問路者和回答者中間穿插而過，而問路者與其中一個抬門板者巧妙互換。等門板過去，回答者繼續回答。雖然「調包」兩人的衣著、相貌差別很大，有一半觀察者却完全察覺不到問路者已經換了人。

如何避免意外

兩位學者的一系列研究都揭示，專注于某一件事的人，很容易忽略其他事情。2010年，他們將相關主題寫成《看不見的大猩猩》（The Invisible Gorilla: How Our Intuitions Deceive Us）一書，大獲好評，影響廣泛。書中大量例子表明，我們以爲可以全面、準確地觀察到周圍所發生的事情，但其實只看到這個世界的極小部分，而錯過了很多。我們的注意力、感知、記憶、推理，都有重大缺陷和錯失，常常導致代價高昂甚至危及生命的錯誤。

每個人都會犯類似的錯誤。香港每年都有多宗高層住客在晾衫時，意外跌出窗外而喪命的慘劇。在工業意外中死亡的雇員，也很可能基于同一原因：不是不懂後果嚴重，而是大大低估了發生事故的可能性。

要改善此情况，當然不能單靠加大懲罰力度，而要從教育入手。勞工處稱，有些個案每每發生在缺乏安全意識的少數族裔身上，因此要特別針對這個群組推展有關宣傳。

雇主方面，則需要考慮到底是公司對問題認知不足，還是只因尚未找到合適、有效的方法。如果是前者，當然應該進一步加重處罰，有議員就建議設立舉報機制。若是後者，則政府和行業協會應該加强教育、督促公司切實執行有關法規，委派專人監督，組織有效的經驗交流。

此外，不妨借助高科技，例如勞工處正在研究利用無人機協助搜證和執法。除了執法和懲罰，高科技還有更正面的作用。例如智能相機鏡頭現已用來及時發出警報，以免司機因打瞌睡而出意外，或長者上洗手間時摔倒。應用到工業場景，可研究能否在搭棚工人的頭盔裝上智能鏡頭，以監察工友在工作時是否已扣上安全繩。至于在固定場所（如工廠），安裝智能相機確保工業安全，相信較易實行。

理性還是非理性

注意力一時疏忽，看似是人類的認知缺失，但換個角度看，倒可以視爲人類認知的巨大成就。幾百萬年前，在非洲大草原上的人類祖先，需要在瞥見移動物體的瞬間，斷定是否獅子、豹子之類的吃人猛獸，是否需要拔腿就跑，而不是全面、準確、客觀地認識世界；在極短的時間內，根據一鱗半爪的訊息，就要迅速决定如何行動。換句話說，認知是解讀而非接收，追求的是效率而非全面、客觀。爲了作出决定，只需抓取一點點關鍵資料，而忽略其他。

人到底是理性還是非理性，所做决策是否對自己最有利？就證據而言，幷不是非黑即白，既有仔細思考、權衡利弊的例子，也有鹵莽大意而丟失性命的個案。

就學術研究而言，理性抑或非理性都只是假設。采取什麽樣的假設，主要不是看假設是否絕對符合實際，而是看它能否最有效地幫助我們分析和理解問題。

至于公共政策，則不能直接假設人是理性的。行爲經濟學對于非理性行爲的關注，可以爲公共政策帶來新的思路。以器官移植爲例，世界各國都面對捐贈器官供應短缺問題。雖然德國與奧地利在語言、宗教、歷史、經濟發展、教育水平各方面都很近似，但德國的器官捐贈登記率只有15%，奧地利却高達90%以上。專家發現，造成這一重大差別的原因，在于德國的系統要求捐贈者填表，聲明同意捐贈，亦即自願捐贈（opt-in）；沒采取任何行動的人，等于不願意捐贈。奧地利的系統則倒轉過來，沒采取任何行動的人視爲「預設默許」（opt-out）捐贈，而不願意捐贈器官者，則需要填表聲明。

科學家的研究表明，在德國、美國這些推行自願捐贈機制的國家，國民認爲器官捐贈道德高尚但成本高昂，好比要决定把遺産的5%甚至50%捐給慈善機構，或相當于絕食抗議這樣代價高昂的行爲。而在奧地利這種實施預設默許安排的國家，國民認爲器官捐贈是件小事，無關道德，有點像讓某人在排隊時插隊，又或缺席孩子的畢業典禮和棒球比賽。

在香港，由此産生的政策建議也就清晰明確：如果希望提升器官捐贈的比率，一個簡單有效的辦法，就是將現行自願捐贈的機制，改爲預設默許捐贈。

周文教授
港大經管學院管理及商業策略副教授

（本文同時於二零二五年一月八日載於《信報》「龍虎山下」專欄）

數智時代的企業適老化實踐

數字化生活已經成為現代社會不可或缺的一部分，然而對於中國2.97億60歲以上的長者來說【註1】，新技術能否讓他們受益卻仍然存疑。儘管各界都在努力推動「智慧養老」，但現實是許多長者在跨越數字鴻溝時，遇到不少挑戰。

隨着全球老齡化趨勢加劇，如何確保龐大的銀髮族享受科技時代的便利，成為了亟待解決的社會問題。

世代數字鴻溝有待縮窄

長者使用數字技術常常面臨諸多障礙。一方面，很多數字產品設計缺乏對長者需求的關注，操作複雜、介面不友好，使得他們即使有心嘗試，也往往因為使用體驗不佳而望而卻步。另一方面，網路詐騙日益猖獗，一些不法分子專向防範意識較弱的銀髮族下手，導致他們在數字生活中遭受財物損失，甚至心理創傷。

資訊超載、隱私洩露等問題也讓不少長者對新技術產生恐懼感，更進一步限制其融入數字生活的步伐。為了使每位長者都能安全、自信地參與數字社會，許多企業都開始探索如何通過數字化技術，構建一個更加包容、人性化的環境。

包容長者如何惠及企業

從環境、社會及管治（ESG）的角度來看，企業在助老領域的投入不僅能帶來顯著的社會效益，還為企業創造多方面的價值。為此，筆者調研了頂尖數字化企業的助老實踐，總結出以下4種效果。

一、提升品牌形象和社會認同。通過積極參與助老項目，企業展示其對社會責任的承諾，增強公眾對其品牌的好感和信任。二、建立社區關係。深入社區提供服務的企業，能夠更好地理解當地需求，建立深厚的社區聯繫。三、增強員工參與感和忠誠度。透過助老項目，員工得以參與公益活動，從而增強其歸屬感和忠誠度。四、推動內部創新和人才發展。在開發適合長者的技術產品和服務過程中，企業能夠激發內部的創新思維和技術進步。

案例1：AI向善語料庫

騰訊於2024年8月啟動了一項面向全社會的共創行動——AI（人工智能）向善語料庫【註2】，旨在為那些在商業環節中失聲的群體和話題，構建一個更加人性化的AI語料庫。通過前期調研，騰訊研究院認識到，由於社會普遍認為長者在數字方面的消費力較低，因此相應需求長期不受重視。目前大部分AI產品亦並非專為長者而設，未能針對其實際需要。要使長者能夠真正受益於AI的發展，大模型需要能夠理解他們的特殊需要，提供資訊實用、滿載關懷的回答。

為此，該項目首先通過聯繫一眾專業人士，包括社工、志願者、社區工作者、心理輔導員、醫生等，加上熱心的高等院校師生，共同收集並整理了數千條長者日常生活中的真實問題及相關解答。這些問題涵蓋了情緒管理、網路購物、心理危機處理，乃至臨終關懷等多個方面，力求全面覆蓋長者的生活需求。

為了確保語料的質量，騰訊邀請不同專家確定了「好語料」的5個標準：一、精確的需求識別；二、充分的同理心；三、切實可行且效果可見的操作建議；四、簡短口語化的表達方式；五、穩定的回答風格。團隊發現，AI回答的同理心尤其關鍵，不僅需要應對長者的負面情緒，提供充分的理解和支持，也應有助於激發和強化他們的積極情緒與自我價值感。騰訊研究院計劃逐步開放AI向善語料庫，以推動社會各界在這些語料的基礎上，創建出更有人情味的產品。

騰訊研究院發布的研究報告。

案例2：藍馬甲行動

為了便於長者更安全地享受數字生活，並增強他們的反詐防騙意識，螞蟻集團及其公益基金會於2020年9月發起了「藍馬甲行動」【註3】。通過傳統方式和數字技術，這一項目旨在協助長者融入數字時代，避免讓長者成為網路犯罪的目標。

線下活動方面，藍馬甲在社區中心舉辦公益講座，講解數字設備使用方法和防騙知識；在社區廣場設立諮詢台，提供現場指導和答疑；組織志願者進行家訪，給予一對一的幫助和指導。至於線上活動，「藍馬甲行動」與20多家夥伴機構攜手合作，生產多元的專業助老科普內容。例如，徐州幸福105電台每天早晚高峰時段都會播放《幸福藍馬甲公益電台》小單元，包含反詐小劇場和手機使用指南等內容。

現時「藍馬甲行動」已經具備4個系統化的支援平台：助老內容創新平台、助老生態支持平台、助老價值宣導平台和助老議題研究平台。基於前期活動的經驗，螞蟻公益基金會在2022年發布《適老化設計與服務參考指南》，2023年發布助老防騙書《一件藍馬甲——關注數字時代的銀髮群體》。「藍馬甲行動」為長者提供了直接的幫助和支持，也促進了社會對銀髮族議題的關注。

藍馬甲志願者在社區駐點協助長者使用手機。

圖片來源：長沙縣知仁社會工作發展中心

政策建議

綜上所述，助老不僅是企業履行社會責任的重要途徑，也是提升品牌形象、促進社會和諧、加強社區關係，以及推動內部管治優化的有效方式。從ESG的角度看，這些活動為企業帶來可持續的競爭優勢和發展機遇，創造雙贏局面。

筆者呼籲更多企業充分發揮自身的技術、資源和服務優勢，積極參與並推動助老事業。通過深入了解長者用戶的需求和心理特徵，設計既易用又富有情感交流的產品；通過教育普及和社會服務，提高他們的數字素養和防騙意識，從而構建一個更加包容、溫暖的數字社會，讓每一位長者都能安全、自信地享受科技帶來的便利。

註1：https://www.mca.gov.cn/n152/n166/c1662004999980001780/content.html

註2：https://mp.weixin.qq.com/s/ggdoPoCtTmimzqu7IW8P9Q

註3：https://mp.weixin.qq.com/s/oqvkME_w1nUBmb7f89ir1w

張閏嘉　
北京大學光華管理學院博士生

顏示硼　
港大經管學院管理及商業策略助理教授

（本文同時於二零二五年一月一日載於《信報》「龍虎山下」專欄）

Professor Heiwai Tang and Mr Cyrus Cheung

18 December 2024

Through the ebb and flow of its economy in the aftermath of the Second World War, Hong Kong has sealed its status as an international financial and trade centre on the world’s economic stage. However, in light of the global economic downturn, fierce regional competition, and worsening geopolitical situation in recent years, coupled with the fact that Hong Kong–as a highly externally-oriented free economy–cannot afford to be complacent simply because it has historically managed to turn crises into opportunities. Times have changed. Now beleaguered by internal problems such as an ageing population as well as external challenges, the city may no longer be as “hardy” as it once was.

To address the challenges in the new era, Hongkongers should not stick to the old rut and must find ways to enhance competitiveness so that long-standing and thorny problems can be resolved.

Formidable problems facing the economy

The first challenge facing the Hong Kong economy over the past few years is the SAR Government’s persistent fiscal deficits. After racking up a record-high surplus of $149 billion for 2017–18, the Government registered a fiscal deficit of $100.2 billion for 2023–24. Initially projected to be downsized to $48.1 billion for the current year 2024–25, the deficit is now estimated to reach $100 billion. Barring the reduction in income from Government-issued bonds, the actual deficit would be even larger. Factors underlying the deficits include fast-increasing government expenditures as well as decline in government revenues. Government expenditure soared significantly from $470.9 billion for 2017–18 to $721.3 billion for 2023–24, of which non-recurrent, social welfare, and healthcare expenditures grew the fastest. The last two items are unlikely to be cut. Meanwhile, government revenue dropped from $619.8 billion to $549.4 billion.

The Figure shows that in 2017–18, 26.6% of the Government’s main sources of revenue came from land premium while 15.4% came from stamp duties. In 2023–24, the share of land premium plummeted to 3.6% and stamp duties dramatically slumped to 8.9%. Despite seeing its share rise from 3.5% to 13.6%, investment income is, after all, not a stable source of revenue. In addition, the Inland Revenue Department annual report 2023–24 reveals that in the year of assessment 2022–23, only around 1.83 million people were required to pay salaries tax, which means that the tax base is still narrow. Given the overall economic downturn, it is unlikely that the Government will be able to sharply reduce the persistently-high fiscal deficits in the short run.

Figure    Sources of the Hong Kong SAR Government revenue for 2017–18 and 2023–24

The second challenge facing the Hong Kong economy is the failure to fully leverage the economic benefits from the integrated development of the Guangdong-Hong Kong-Macao Greater Bay Area (GBA). Despite the high degree of integration of consumption activities in the GBA, Hong Kong’s professional services sector has yet to be fully integrated into the development of the GBA. While it has become a trend among Hongkongers to go north for spending, there is much less incentive for Mainlanders to spend in Hong Kong. At the same time, more and more local people prefer to shop on Mainland e-commerce platforms, inevitably impacting the sales of physical stores in Hong Kong. According to the Census and Statistics Department’s data, the Value Index of Retail Sales dropped from 144.8 points in 2018 to merely 121.3 points in 2023 while the monthly average for the first 10 months of 2024 went further down to 111.8 points.

Furthermore, the Volume Index of Retails Sales dropped from 148.9 points in 2018 to 113.9 points in 2023, and even averaged 103.3 points in the first 10 months of 2024. The decreases for both indexes are equally significant. Due to constraints in rent and labour costs, the local retail industry can hardly compete with its Mainland counterpart in terms of cost-effectiveness. It seems that the professional services industry, in which Hong Kong excels, has not been able to capitalize on the market opportunities in the GBA. This can be put down not only to the sluggish macroeconomy in recent years but also to delayed cross-boundary professional qualification certification, administrative red tape, and other factors.

The third challenge facing the Hong Kong economy is the gradual shrinking of the middle class and international talent drain. Hong Kong’s unitary economic structure is one of the primary reasons for the weakening middle class. Since middle-income jobs have always been concentrated in the financial, real estate, and professional services industries, problems will start to surface as soon as these sectors face headwinds. In addition, while there has been a mass migration of middle-class families overseas in the last few years, the highly-educated new immigrants to Hong Kong are less internationalized. The LinkedIn profile data analysed in an essay in the “Hong Kong Economic Policy Green Paper 2024” published by the HKU Business School demonstrates that the proportion of Asians among leavers is 58% and is as high as 79% among joiners, while the number of connections of leavers is 1.7 times that of joiners.

Four reform strategies to chart a new course

Cracking the above problems is no easy task. Let me outline below four policy directions to spark more valuable ideas from all sectors.

1. Broadening the tax base to tap new revenue sources

Hong Kong can take a leaf from Singapore’s book and consider introducing consumption tax progressively to relieve financial pressure on the Government. The goods and services tax (GST) implemented in Singapore at a rate of 3% in 1994 gradually rose to 9% in 2024. In 2023, the GST contributed to 15.7% of Singapore’s fiscal revenue. Based on private consumption expenditure in Hong Kong, and after deducting existing overlapping tax items, a 2% GST can bring the Hong Kong SAR Government an incremental income of $27 billion, roughly equivalent to 5% of the financial revenue in 2023. Referencing Singapore’s experience, retail sales fell following upward adjustments of GST in 2007 and 2023, but not after GST hikes in 1994, 2003, and 2004. An economist at RHB Bank points out in a recent study that the GST increase in January 2024 did not have a strong impact on Singaporeans’ spending habits. This suggests that a GST rise does not necessarily dampen retail sales. The key lies in a balanced measure that entails controlled tax increases, effective expectations management, and complementary welfare policies to maintain steady consumer sentiment.

As a matter of fact, the introduction of GST in Singapore did arouse controversy. For example, there were views that daily necessities should be exempted. The Singaporean government did not accept this suggestion because of the potential increase in compliance and audit costs. Instead, the authorities chose to alleviate pressure on low-income families by issuing GST vouchers, subsidizing public education and healthcare services, etc. In any case, if the GST rate is set too low, it will not be adequate to alleviate the Government’s financial problems. Conversely, if the rate is set too high, it will breed dissatisfaction among businesses and the general public and could build up excessive inflationary pressure. How to strike a balance in between is a great challenge for policy formulation. Moreover, the Government can consider selling idle assets, including unused premises and surplus equity, to ease financial pressure.

2. Fostering development via public-private partnerships

The SAR Government can consider strengthening public-private partnerships to promote infrastructure development. Apart from reducing the Government’s initial investment and operating costs, this approach helps to bring in technologies and management experience from leading enterprises. Many international cities have achieved remarkable results through this development mode. Examples include the Chicago Skyway and the Port of Long Beach Middle Harbour Redevelopment Project in the US, the Marina Bay Sands Integrated Resort in Singapore, the Beijing subway Line-4 project, and the Eastern Harbour Crossing in Hong Kong. There are, of course, different forms of public-private partnerships, with “Build, Operate, and Transfer”; “Build, Own, and Operate”; “Transfer, Operate, and Transfer” among the most common modes. Hong Kong can choose the suitable mode depending on its specific needs.

3. Leveraging unique advantages to integrate into the GBA

Hong Kong needs to optimize its integration with other cities in the GBA to give full play to the benefits of the regional economy. According to the Second Agreement Concerning Amendment to CEPA Agreement on Trade in Services recently signed by the SAR Government with Mainland authorities, the Mainland market will be further opened up to Hong Kong enterprises offering professional services. On this basis, the SAR Government can continue to maintain close liaison and cooperation with other GBA cities to ensure the successful implementation of policies. For instance, assistance can be provided for Hong Kong’s estate surveying companies to complete the filing of records to bid for consultancy services projects in joint ventures within the GBA.

Given the distinct advantage of Hong Kong’s higher education in the GBA, the SAR Government can maintain close cooperation with sister cities and continue to support higher-education institutions in building branch campuses in the GBA and achieving success in their subsequent development. Opening up the “four flows”―human flow, goods flow, capital flow, and information flow―is of vital importance in this regard. Only by doing so will it be possible for the branch campuses in the GBA to obtain invaluable resources from both the Mainland and abroad, and for Hong Kong’s higher education institutions to preserve their competitive edge in internationalization.

To maximize the removal of operating restrictions on Hong Kong’s professional service sectors, such as finance, law, and accounting, in the GBA, the SAR Government needs to continue to lift systemic barriers in the area. For example, in the First Phase Report on Survey of the Current Situation of Hong Kong Legal Practitioners under the Development of the Guangdong-Hong Kong-Macao Greater Bay Area, the Law Society of Hong Kong and the School of Law of Sun Yat-sen University point out that under Mainland laws, the associations formed by Hong Kong law firms with Mainland law firms shall not be in the form of partnership or legal entity. Therefore, cooperation between Hong Kong and Mainland law firms is mainly based on non-partnership associations. This gives rise to various problems, including differences in handling conflicts of interest, discrepancies in business acceptance and processing standards, and a lack of clarity on the legal responsibilities of non-partnership associations.

4. Proactively competing for talent and enticing foreign investments

Apart from talent and capital from the Mainland, Hong Kong must also focus on attracting talent and funds from abroad to maintain its relative advantages as an international metropolis. In view of the fact that the career development of ethnic Chinese technology experts in Europe and the US is thwarted by current geopolitical tensions, the SAR Government should seize this opportunity to encourage them to advance their careers in Hong Kong. Meanwhile, the authorities can also consider setting specific performance indicators for local universities, e.g. target percentages for international students, to reinforce local higher education institutions’ strengths in internationalization.

Needless to say, Hong Kong must continue to leverage the unique advantage of “one country, two systems” to draw in more direct foreign investments to the Mainland. At the same time, apart from enticing Mainland investments through the Government’s Office for Attracting Strategic Enterprises, it is also necessary to enhance the presence of leading foreign enterprises to maintain Hong Kong’s distinctive advantage as a bridge to the world. The growth of emerging sectors, e.g. artificial intelligence, biotechnology, financial technology, advanced manufacturing, and new energy sources, will determine if Hong Kong can produce more high-quality jobs in future, thereby expanding its middle class and furthering its economic prosperity.

As we mentioned in this column two years ago, priority should be given to creating a liveable environment when it comes to attracting talent. Otherwise, they will not stay after arriving. On the one hand, they need to find high-quality jobs in Hong Kong, connect with a thriving professional community, and enjoy a comfortable and vibrant living environment. On the other hand, high-quality human capital is a key consideration for companies with an eye to establishing their presence in Hong Kong. Hence, efforts to attract companies and capital, compete for talent, or even formulate cultural policies should be complementary rather than isolated from one another. Retaining talent and businesses is a systemic project that requires comprehensive policy coordination across the SAR Government to achieve success.

Dr Yifei Zhang

11 December 2024

Nowadays, with the advancement of artificial intelligence (AI) technology in leaps and bounds, AI applications have permeated various aspects of human life—not only from smart assistants to autonomous driving technologies but also from industrial production to medical diagnosis. According to the International Data Corporation, the global AI market value is expected to rise from US$132.4 billion in 2022 to US$512.4 in 2027.

While the convenience of AI innovation is applauded by all sectors of society, does it also raise the community’s awareness that the technological revolution is subtly exerting a tremendous impact on the global environment? As a matter of fact, the problem of carbon emissions arising from the AI development process has reached such a state that it can no longer be ignored.

The invisible killer: the carbon footprint of AI training

To understand the impact of AI on the environment, it is necessary to unveil the true face of AI training models. The training process for modern AI models, particularly large language models, requires massive amounts of data and calculation resources. The latest research by the University of Massachusetts Amherst indicates that carbon emissions from training a large AI model can reach 626,000 pounds, equivalent to the total emissions from five vehicles throughout their entire life cycle, from production to disposal.

Specifically, approximately 552 tonnes of carbon dioxide are emitted during the training process of GPT-3 while the CO2 emitted from training the even larger model, GPT-4, is estimated to exceed 1,000 tonnes. Of particular concern is that these figures continue to go up. Under the sectoral consensus that “large models are the order of the day”, giant technology companies have been vying to develop even larger models, resulting in exponential surge in energy consumption. The AI sector’s carbon emissions are forecast to account for 3.5% of the world’s total carbon emissions by 2030.

Data centres: an energy-guzzling beast in the AI era

The energy consumption of large AI models has now reached an alarming level. Data of the Stanford AI Laboratory shows that one single training session of GPT-3 typically uses 1,287 megawatt-hours of electricity, equivalent to all the power consumption of 3,000 Tesla electric cars each travelling 200,000 miles, emitting a total of 552 tonnes of carbon dioxide.

In daily use, every response generated by ChatGPT requires 2.96 watt-hours of electricity, almost 10 times that (0.3 watt-hour) for a standard Google search. Each Google search powered by AI even utilizes 8.9 watt-hours. The water resource consumption level is also alarming. During its training, GPT-3 consumes close to 700 tonnes of water. For every 20 to 50 questions, 500 millilitres of water are required. For cooling of its data centres alone, Meta used over 2.6 million cubic metres of water in 2022.

Root causes of escalating energy consumption

The colossal energy consumption of large AI models can mainly be attributed to two core factors. First, the rapid iterations of AI technology have significantly stimulated the demand for chips, directly pushing up electricity consumption. The training and inference processes of modern AI models deploy enormous computational resources, which primarily rely on high-performance hardware, including graphics processing units and application-specific integrated circuits. This hardware is highly energy-intensive when running complex computations. As AI models keep expanding in size, their computational capabilities have seen exponential growth, resulting in an ever-increasing demand for high-performance chips and, consequently, mounting energy consumption.

Furthermore, substantial computational power is essential for supporting the AI model training process. The around-the-clock data centres generate excessive heat, necessitating cooling treatments. Energy consumption is an especially severe issue for data centres, which serve as the core infrastructure for AI computation. Servers and storage devices running at high loads release a vast amount of heat. If the heat is not dissipated in time, both the performance and lifespan of the devices will be seriously compromised. Hence, data centres are equipped with super-efficient cooling systems to ensure that the devices operate at optimal temperatures.

In the operating cost structure of a data centre, electricity tariffs account for 60% of the total cost, of which over 40% is spent on cooling systems. At an air-cooling data centre in particular, more than 60% of electricity is used for cooling while less than 40% is used for computation. As a result of this energy utilization imbalance, the energy consumption of data centres around the world is now almost 10 times more than it was a decade ago. Traditional air-cooling systems are less costly but also less efficient, making them incapable of meeting the requirements for high-efficiency cooling. In comparison, except for a large-scale investment at the initial stage, liquid-cooling systems are more efficient, thus sharply reducing energy consumption at data centres.

In addition, the site selection and design of data centres have a significant impact on energy consumption. Many data centres are located in areas with lower electricity costs but in hot climates, placing a heavier burden on the cooling systems. To enhance energy utilization efficiency, priority should be given to locations with cooler temperatures and a stable energy supply. Besides, a modular design should be adopted so that resource allocation can be flexibly adjusted according to needs.

Finally, the training and inference processes of AI models also involve huge amounts of data transmission and storage, which inevitably boost energy consumption. As more and more data is created, data centres need extra storage devices and greater bandwidth to cope, which further expands energy consumption. These facts demonstrate that companies should make use of data compression and transmission optimization technologies to cut down energy consumption by minimizing unnecessary procedures.

Corporate solutions and policy suggestions

In the face of the environmental challenges from AI technology, companies and policy-makers need to take a series of carbon-reduction measures. First, businesses should maximize the use of green energy and energy-saving technologies. Investments should be made in renewable energy sources such as solar energy and wind energy to minimize reliance on traditional fossil fuels. Second, enterprises should optimize AI model training algorithm to streamline computation, cutting down energy consumption at source. Third, they should enhance data centre management and upgrade technologies; use high-efficiency solutions such as liquid-cooling to promote energy utilization efficiency; and minimize waste of idle resources through smart dispatch and load balancing. Fourth, through virtualization technology, companies can integrate computational resources to lower energy consumption.

In terms of policy-making, the government should first set strict energy efficiency standards and promote green development of AI technology; and, through tax concessions and funding, encourage enterprises to adopt energy-saving technologies and renewable energy sources. Second, regulation of data centres should be strengthened and energy efficiency evaluation standards should be established to promote overall energy efficiency. Third, governments and industry should join hands to spread environmental awareness among the public and businesses. The negative impact of AI technology on the environment can be minimized through such mechanisms as carbon trading and carbon offsets. Both education and publicity are indispensable, as only when the relationship between AI advancement and environmental protection is widely known can a social consensus be reached and concerted efforts be made to address the problems.

Glimmers of hope amid crisis

The trend of AI technology may well be overwhelming, but we must ensure that the environment will not be harmed as a result. Through technological innovation, corporate self-regulation, government guidance, and social oversight, environmental impact can be minimized while the convenience of AI can be enjoyed by all. The International Renewable Energy Agency predicts that, with proactive measures, the annual growth in carbon emissions by the AI industry can be controlled within 5% by 2030.

As witnesses and participants of this era, each and every one of us should be concerned about the environmental issues brought about by advancements in AI and take concrete actions to support its green development. Only through this approach can we ensure that the AI technology benefits mankind instead of becoming another burden on the Earth. In our quest for technological breakthroughs, environmental protection should be the bottom line that must be upheld, not just a token gesture. Let all sectors of the community make concerted efforts to drive AI towards a greener and more sustainable future.

Through policy guidance, technological innovation, and public engagement, the path will be paved for Hong Kong to achieve the AI industry’s carbon-neutral goals by 2035 and contribute to the sustainable development of the world.

Dr Tingting Fan

4 December 2024

When we listen to music through our earphones these days, do we realize that over 30% of it is already produced by AI? Last year, the AI-generated song “Heart on My Sleeve” got 20 million hits on Spotify. Early this year, the burgeoning AI-generated music scene drove Universal Music, the world’s leading music company, to remove all its records from TikTok, the largest platform for short-form mobile videos. After the two companies reached an agreement in mid-2024, TikTok agrees to label all AI music footage accordingly. The impact of AI technology on the music industry has been fast and furious.

Cause for celebration or concern

With the progress in AI technology, major technology companies have extended their reach over the music industry. For example, the text-to-music model named MusicGen was launched by Meta to users in 2023. The Stable Audio 2.0 model, introduced by Stability AI this year, even allows users to upload existing music to generate new tracks in a completely different style. The acoustic quality is comparable to that of a vinyl record.

It is a cause for celebration because AI enables ordinary people, who are not music professionals, to not only “create” music with ease but also earn money from these “creative” works. Boomy, an American start-up, supports users in uploading their AI-generated music to Spotify and other streaming platforms for a commission.

That being said, it is a cause for concern because if music can be “created” by an AI model, would professional musicians find themselves out of a job? Given the five consecutive months of protest from Hollywood actors and screenwriters in 2023, the looming fear is clear as day.

As a matter of fact, the great concern is not unjustified. In 2017, 87% of music tracks played on Spotify were from singers signed with record labels. By 2022, this percentage fell to 75%. As of 2023, over 100 million pieces of music were generated by AI, taking up around 30% of our music-listening time. The revenue generated by the AI music market is projected by industry members to reach US$7 billion by 2026, while AI music is expected to have a 50% share of the music sector by 2030.

Quantity or quality

At present, the advantages of AI-generated music lie in speed and quantity. Boomy claims that in just a few years, there are already 18 million pieces of AI-generated music, whereas only 100 million pieces of old and new music spanning all time periods have found their way to Spotify so far. Nevertheless, does the quality of AI-composed music rival that of the creative works by professional musicians? So far, the works created by AI have been based on past music. With the rapid growth in quantity, the quality of AI-generated music will eventually regress to the mean. When the excitement over this new technology wears off for the public, will people tire of AI-produced music and turn to works by music artists? Alternatively, one wonders if a “scientific division of labour” is possible, whereby AI-generated music serves as low-cost background music while the concert stage is reserved for music artists’ works to shine.

When it comes to people’s requirements for music, quantity and quality are never mutually exclusive. Striking a balance between the two is something that both AI companies and music artists should explore.

As a matter of fact, both music artists and record companies, which rely on music copyrights to survive, are on the receiving end of AI-generated music’s vexing challenge. The music works owned by record companies provide the raw materials for AI models to “create” new music after learning their characteristics. But should AI companies be obliged to pay royalties for these raw materials? And should AI-created music be under copyright protection?

Recent years have seen increasing challenges arising from these issues. In 2023, for instance, Universal Music accused Anthropic, an AI company with investments from Google and Amazon of illegally using works owned by Universal Music to train Anthropic’s AI models. In its defence, the company claims that using existing music to train AI models does not constitute copyright infringement.

Challenge or opportunity

Since advancements in AI technology have far outpaced developments in intellectual-property laws, these problems without quick solutions have become a grey area, presenting both challenges and opportunities for record companies and AI technology companies. In retrospect, this is not the first time music publishers have encountered copyright challenges. The late 20th century saw the migration of music from CDs to electronic MP3 files, which, coupled with the rise of sharing platforms like Napster, led to rampant pirated music that pushed many record companies out of business. It took an entire decade for record companies to develop a business model more profitable than the traditional approach of selling CDs and to eventually reach agreements with music streaming platforms regarding music copyright.

Taking lessons from history and embracing the unstoppable trend of AI music, record companies no longer regard it as an uncontrollable beast. Instead, they are striving to devise a new business model that can enable music copyrights to bring greater profit in the AI era. Robert Kyncl, CEO of Warner Music Group, once says that simply rejecting AI and fighting against it is out of the question. In promoting legal definition and protection of music copyright, record companies actively use AI to facilitate music creation by professional artists in cheaper and faster ways on the one hand. For example, AI is harnessed to produce multilingual versions of podcasts for the enjoyment of audiences around the world. Machine learning has even been used to extract a muddled demo song left behind by the Beatles’ lead singer, John Lennon, in 1973. The recent release not only gave new life to the song “Now and Then” but also reignited enthusiasm for their Beatles’ classic ballads. On the other hand, AI models are also trained to precisely detect copyright-infringing music for litigation purposes, if necessary. Additionally, record companies may even roll out a two-pronged carrot-and-stick strategy to protest against copyright infringement by AI companies while leveraging their advantage as copyright owners to become market pioneers through closer cooperation with AI companies.

Two centuries ago, the Fate Symphony strikes a chord with us, helping us to empathize with Beethoven’s struggle against destiny after losing his hearing. Two centuries later, if Beethoven came back to life, could AI restore his hearing to inspire him to compose even more masterpieces for posterity? Two hundred years ago, through musical notes, Beethoven issued the rallying cry: “Listen! Fate is knocking at the door!” Today, two hundred years later, hopefully the door opened by AI will lead to a new era of human creativity!

美國大選結果折射出的避責文化

互相指摘或卸責，也許只是個人沒擔當的怯懦行為，但放大到社會層面，就足以產生混淆公眾視聽的惡果。政治人物往往透過彼此指摘來轉移視線，力求貶低對手而抬高自己。在競爭白熱化的選舉中，不惜一切推卸責任已成政客的慣技，或對選舉結果以至未來管治和政策帶來難以想像的衝擊。

民主黨敗選背後

本月美國總統選舉結果塵埃落定，共和黨特朗普以壓倒性姿態勝出。賀錦麗慘敗後，其所屬民主黨內隨即出現大舉卸責現象，矛頭直指拜登，歸咎他未能及時退選，陷賀錦麗於尷尬境地。不少黨內成員亦認為拜登年老退化，不受選民歡迎，雖然他及後宣布退選，賀錦麗仍因受選民支持度不足，未能於明年入主白宮。

與此同時，民主黨在國會改選中失去參議院和眾議院的控制權，較4年前表現更糟。根據《紐約時報》的分析，自拜登在2020年出任總統以來，美國3100多個縣的選民大都轉向右傾。民主黨向來標榜的支持墮胎權和民主立場，無法像經濟和移民等迫切議題引起選民共鳴。

儘管美國失業率現正維持在歷史低位，股市暢旺，但物價高、房租貴也是事實。拜登任內，物價上漲超過20%。康奈爾大學的經濟學家巴蘇（Kaushik Basu）指出，各種經濟指標之中，通脹對政治影響最大。一般人無需數據，也對通脹有切身感受。再者，《金融時報》的分析顯示，在今年舉行選舉的10個國家中，執政黨的表現都不如上屆選舉，相信也與高通脹有關。

根據民調，三分之二的美國選民對經濟給予劣評，收入較低的一群傾向於支持特朗普。2020年，他以15個百分點的差距失去收入介乎5萬至10萬的選民，但在這次選舉中卻逆轉獲勝。民主黨人似乎忽略了馬斯洛的需求層次理論（Maslow’s Hierarchy of Needs）：基本需要（如財務穩健和身心健康）必須先行，然後再滿足其他方面。在競選活動中，民主黨聚焦於民主等議題而忽略經濟。曾經是該黨核心的工人階級選民不再予以支持，因愈來愈多人按自身的經濟利益來投票。黨內對敗選結果莫衷一是，更出現互相指摘。如此反應，是否就能把選票贏回來？答案不言而喻。政治指摘伎倆層出不窮，皆因政黨或領導人藉此進行政治操弄，以便大權在握。

企業卸責文化

在商業環境中，互相指摘確也頗為普遍。譬如一家公司面臨存亡危機，責任的分配將直接影響其股價和投資者的信心。假使管理層只管找替罪羊，哪怕是象徵式的代罪羔羊，公司亦難逃衰敗的厄運。從管理學的研究可見，將公司失敗歸咎於外部因素的管理層，其整體表現往往不及承認自身責任並自我反省的公司。在瀕臨破產的企業中，可以看到不少經理將業績欠佳委過於其他部門。相反，管理層若有責任感，則有可能轉虧為盈，讓業務重上軌道。前事不忘，後事之師，只有汲取教訓，才能避免重蹈覆轍。

此外，互相指摘也足以助長風險規避文化，員工因害怕受責備而不敢主動行事，或礙於不願分享想法而窒礙創意。眾所周知，成功的企業有賴暢順的運作；管理層必須致力培養團隊合作精神，以解決公司內部的分歧。

「無過失」調查的啟示

反觀一些行業早已認識到指摘的弊端，例如航空業所以在降低意外事故一環取得成效，很大程度上受惠於「無過失」調查的程序。在美國，負責調查有關事故的國家運輸安全委員會明確表示，調查目的並非追究責任，而是找出問題並提出建議，以防同類事件重演。航空業不進行追責的事後調查，為現代航空安全奠下重要基石。

這種調查方式有助於建立開放的安全文化，鼓勵業界報憂，最終目的是確保減少意外事故。英國的航空監管機構在誠實錯誤和其他錯誤之間劃界線，也是個好的起點。航空公司致力於營造一種文化，使機師不會因為與其經驗和培訓相符的決定或疏忽而受到懲罰。這種做法並非完全免責，只是將責任範圍收窄而已。

醫療保健領域也面臨類似情況。一旦發生醫療事故，世界各地對病人的補償制度各有不同。例如英國依賴找出過失的訴訟程序，而紐西蘭則是全球最早實施醫療事故處理制度的國家。紐西蘭率先以「無過失補償」的程序來處理醫療事故，並於1974年成立意外補償局負責，接受因工作、交通或醫療事故導致的傷害賠償申請。在這一制度下，無論醫療措施或副作用造成的傷害是否可以避免，病人均可向補償局提出申請。只要問題與醫療診斷或決策相關，申請便可獲批准。該制度推行後，除非醫療人員的行為嚴重違法，否則紐西蘭患者幾乎無法向醫療機構提出訴訟。

在航空和醫療領域，從錯誤中學習的動機特強，因為從業員在工作中生命隨時受到威脅，安全無疑至關重要。因此，軟件工程師和開發人員經常進行「無過失的事後分析」，以調查網站失靈或伺服器故障等問題。一般人不易理解這種不追責的思維，心理學家James Reason在1990年代為此提出一個框架，以釋除大眾對無能和犯錯者逃避責罰的疑慮。

問責而非卸責

要逃避指摘其實並非易事。一、當事人為了避責往往要大費心力，但指摘別人反而是毫不費力的快速反應，而且容易令人入信。至於記錄錯誤並確保流程得以改進，則難免涉及結構性的變化。例如無過失事後分析長期以來已屬谷歌企業文化的一部分，該公司為此提供模板、反饋和討論小組。二、企業管理層既然大權在握，指摘屬下僱員也就輕而易舉。

加州大學聖地牙哥分校和新加坡南洋理工大學的學者最近合作發表一篇研究論文，指出當權者往往認為其他人會將失敗歸咎於他們。在一項實驗中，參與者被隨機分配為主管或工人，然後檢視有關錯誤的紀錄。參與者都收到道歉信，聲明網絡連接不穩定，以致任務無法正常完成。結果扮演主管者每多認定抄寫員應為失誤負責，主張剋扣其報酬。由此可見掌權與施罰之間的因果關係。

指摘別人似乎也具傳染性。2009年，心理學學者David Sherman 和 John Klein發表合著論文【註】，其中一個實驗要求參與者閱讀有關政治失敗的新聞，然後寫下政客的過失。讀到關於政客將失敗歸咎於特殊利益的報道時，參與者更可能將自己的失敗責任推卸給別人。至於讀到政客承擔責任的參與者，則更可能肯為自身的不足負責。同理，管理高層若輕易指摘別人，公司員工也會有樣學樣。如此一來，不難衍生出一種推卸責任的指摘文化。

不同文化對於失責和指摘的容忍度不盡相同。例如集體主義可能導致共同指摘，而在個人主義的文化中，個人指摘則較常見。相互指摘的經濟學強調人類行為與經濟結果之間的相互作用。了解這些動態關係當有助於機構創造出更具建設性的環境，減少諉過於人，以鼓勵問責和合作。

註：Sherman, D. K. and John M. Klein, “Failure to Blame: The Effect of Collective Blame on Self-Attribution.” Psychological Science, 2009.

謝國生博士
港大經管學院金融學首席講師、新界鄉議局當然執行委員

何敏淙先生
香港大學附屬學院講師

（本文同時於二零二四年十一月二十七日載於《信報》「龍虎山下」專欄）