Newsletter No. 542

02 # 5 4 2 | 0 4 . 0 9 . 2 0 1 9 p 金國慶教授 Prof. Irwin King T he perception of artificial intelligence (AI) is varied, as shown in popular movies. For R2-D2 in Star Wars , T-800 in The Terminator , and the eponymous hero in Wall-E , AI is personified by a robot which has cognitive ability and identity. When it comes to J.A.R.V.I.S. in Iron Man , which is the hero’s servant and council, AI takes the shape of a virtual system. With so many things and products given the tag ‘AI,’ how can we know what AI truly means? In 1950, Alan Turing proposed his now-famous ‘Turing Test’: A machine could be classified as AI if it could talk with a human without it being recognized as a non- human entity. The definition of AI, however, has continued to evolve with technological advancement. Prof. Irwin King of the Department of Computer Science and Engineering proposed a new definition: ‘AI has the capability to learn and improve, and can adapt to the users’ needs.’ For example, the more facial pictures are input into a deep-learning network, the higher its facial recognition accuracy will become. It can even distinguish the age, ethnicity, and emotion of a subject. As such, a fuzzy logic fan is ‘smart’ instead of ‘AI’, although it can adjust its speed as the temperature changes. In view of the increasingly high demand for AI professionals in local and even global employment markets, the Department of Computer Science and Engineering has launched an undergraduate programme in Artificial Intelligence: Systems and Technologies (AIST) this year. The Hong Kong government has formulated policies to promote the development of innovative technologies, including plans for the expansion of the Science Park in Tseung Kwan O Industrial Estate and the establishment of HK-Shenzhen Innovation and Technology Park in Lok Ma Chau Loop. It is expected that 50,000 jobs will be created. Professor King pointed out that the new programme addresses the needs of society on the strengths of the Faculty of Engineering. From the studies of statistics, machine learning and algorithm design, students could learn to construct an entire AI system step by step. Based on their interests and career goals, students may choose a specialized stream from four options: Biomedical Intelligence, Intelligent Multimedia Processing, Large-scale Artificial Intelligence Theory and Systems, and Intelligent Manufacturing and Robotics. Professor King said, ‘The programme offers four specialized streams, one theoretical and three application-oriented. Like building a house, it starts from the foundation. “Large-scale Artificial Intelligence Theory and Systems” is the foundation.’ 對 於人工智能,不同人有不同想像,從大眾電影中可見端倪。以 《星球大戰》的R2-D2和C-3PO、《未來戰士》的T-800,以及 《Wall-E》的同名主角為例,人工智能等同有思想和身分的實 體機械人,但在《鐵甲奇俠》中的「卓維」,卻是一套無形系統,既是主 角的管家,也是戰友。 人工智能盛行,很多產物都高舉AI幌子,叫人眼花撩亂,我們應如何分 辨?艾倫 • 圖靈於1950年提出「圖靈測試」,如果一台裝置能夠與人對 話而沒有被認出,便屬於人工智能。然而,科技發展一日千里,人工智能 的定義也與時並進,計算機科學與工程學系 金國慶 教授提出一項標準: 「人工智能是懂得學習的,可以進步,並能根據使用者的需要適應配 合。」以人臉辨識為例,我們把愈多人臉圖像輸入深度學習網路,它識 別人臉的準確度就愈高,甚至可以分辨出對象的年齡、種族和情緒等。 相反能感應環境、自動調節參數的快思邏輯(fuzzy logic)電風扇雖然 可以因應溫度高低調節風速,但只算智能,不屬於人工智能。 有見本地以至全球缺乏人工智能專才,中大計算機科學與工程學系今年 開辦「人工智能:系統與科技」學士課程。香港積極推動創新科技和建設 相關配套,例如擴充將軍澳工業邨的科學園和成立落馬洲河套區港深創 新及科技園等,預料創造五萬個職位。金教授指出,新課程既是回應社會 發展的需要,也是發揮中大工程學院的強項。學生可以由統計學、機器學 習,以至演算法的設計,逐步學習構建一套完整人工智能系統。 另一特點是,學生可以按興趣和志願選擇專修範圍,包括智能生物醫 學、智能多媒體處理、大規模人工智能理論與系統,以及智能製造與機 器人學。金國慶教授說:「課程有四個專修範圍,一個以理論為主,三個 以應用為本。就像建屋一樣,要有根基才能向上興建,『大規模人工智 能理論與系統專修』便是根基。」 「智能生物醫學」是應用專修範圍之一,現時基因組研究崛起,若要從 海量的基因資料中,探索個別基因與某些疾病的關連,演算法便大派 用場。若發現某一基因是引致某種癌症的高危因素,擁有該基因的人便 可及早調節生活習慣,降低發病機會。 除了人臉辨識外,「智能多媒體處理」亦涵蓋無人駕駛、醫療造影、語音 和音訊處理等。以「錯字和粵語檢測系統」為例,該系統由系統工程與 工程管理學系 黃錦輝 教授的團隊研發,他們把高中生的中文作文試卷、 課本和辭典等材料輸入系統,讓其「學習」。當中小學生把作文輸入,系 統便會按上文下理找出錯別字和區別簡體字。 「智能製造與機器人學」則是把人 工智能應用於機械裝置,例如搬 運機械人。機械與自動化工程學系 劉雲輝 教授及其團隊研發的「視覺 導航無人搬運車」,把廠房的平面圖 輸入智能系統,系統收集廠房環境 的相片並加以分析,便可規劃出搬 運路線,自主行走,不用人手操作。 縱然現時人工智能大多只有單一領 域的功能,不像「卓維」般身兼多 職,可以處理複雜任務,但它確確實 實在各方面協助人類改善生活。猶 如上世紀末互聯網和電郵誕生,人工 智能勢必滲入各行各業,改變其營運 模式,繼而改變世界。 AI in All Walks of Life 人工智能 課程 孕育百業專才 Biomedical Intelligence is one of the application- oriented specialized streams. Genomic research has gained currency in recent years. Algorithms could be applied to a large database of genetic information to explore the relationship between a gene and some diseases. If it is found that a particular gene is a risk factor for a cancer, a person carrying that gene may be advised to adjust her lifestyle to reduce the cancer risk. In addition to facial recognition, ‘Intelligent Multimedia Processing’ covers driverless cars, medical imaging, and speech and audio processing. Take ‘Automatic Colloquialism and Typo Detection System for Chinese Language’ as an example. The research team led by Prof. Wong Kam-fai of the Department of Systems Engineering and Engineering Management developed the system which can ‘learn’ from the Chinese compositions of senior secondary students and feed on the content of textbooks and dictionaries. When a secondary or primary school student uses the system on her compositions, the system can identify the typos or misused words. It can also differentiate between the simplified Chinese characters and the traditional characters. ‘Intelligent Manufacturing and Robotics’ applies AI in mechanics, one such product being the robots for transporting goods. The team led by Prof. Liu Yun-hui of the Department of Mechanical and Automation Engineering has developed a self-piloted forklift truck. Its AI system will analyse the floor plans and pictures of the warehouse and devise the route the autonomous truck takes to transport the goods. Although most AI applications have only functionality in one particular area, unlike the versatile J.A.R.V.I.S., they are making the lives of many easier. AI will proliferate like the Internet and the e-mails born at the end of the last century in all walks of life to change many ways of working and living. M. Mak