1.4 AI發展史

Summarize the history of AI development

suffer many mishaps（命運多舛）！ have a great future（前途無量）？
三起兩落（ Three ups and two falls ），命運多舛

1950–1956, The Birth of AI

• 1950, Alan Turing proposes the Turing Test as a measure of machine intelligence.
  1950年，艾倫·圖靈提出了圖靈測試，將其作爲機器智能的度量。
  -1956, the field of AI research was founded at a conference on Dartmouth College. 1956年，在美國達特茅斯學院的會議上，人工智能研究領域正式誕生。
  

---

• Professor Atanasoff of Iowa State University and his graduate student Berry built the world’s first electronic computer, Atanasoff-Berry Computer (ABC,“阿塔納索夫－貝瑞計算機), from 1937 to 1941.
• It laid the material foundation for the study of AI.
• Not the American mathematician Mokley and Echo invented in 1946!
  

1956–1974, The Golden Years

• After the Dartmouth Conference, the United States formed a number of AI research organizations, such as
• Carnegie RAND Co-working Group led by Newell and Simon ,
• the MIT Research Group led by Minsky and McCarthy,
• the IBM Engineering Research Group led by Samuel.
  達特茅斯會議後，美國形成了多個人工智能研究組織，如：
   紐厄爾和西蒙的Carnegie RAND協作組，
   明斯基和麥卡錫的MIT研究組，
   塞繆爾的IBM工程研究組等。
• In 1958, Rosenblatt developed a perceptron , a system that uses neurons for recognition. Its learning function has aroused wide interest and promoted the research of connection mechanism. However, the limitation of perceptron is quickly discovered, which can not solve the complex problem of recognition, so the research of connection mechanism has entered a low tide.
  1958年，Rosenblatt研製了感知機，這是一種將神經元用於識別的系統，它的學習功能引起了廣泛的興趣，推動了連接機制的研究。但很快就發現了感知機的侷限性，不能解決複雜的識別問題，從而連接機制的研究進入低潮。（感知機：單個神經元，線性不可分問題，如異或問題，現實世界大部分問題線性不可分）
• In 1958, Herbert Simon and Allen Newell demonstrated the first AI program, Logic Theorist (LT). 1958年赫爾伯特.西蒙和艾倫·紐厄爾演示了第一個AI程序，名稱爲邏輯理論家。
• 1958, John McCarthy (MIT) invented LISP programming language, which was developed as an important tool to build expert system.
  1958年麥卡錫發明了LISP (List Processing列表處理縮寫)編程語言，成爲建造專家系統的重要工具。
• 1960s, M. Masterman and his colleagues at University of Cambridge design semantic nets for machine translation.
  1960年代，M·馬斯特曼與劍橋大學的同事們設計了語義網絡，用於機器翻譯。
• 1963, Leonard Uhrand Charles Vossler published “A Pattern Recognition Program That Generates, Evaluates, and Adjusts Its Own Operators”, which described one of the first machine learning programs.
  1963年倫納德·武赫和查爾斯·瓦斯勒發表了關於模式識別的論文，描述了第一個機器學習程序。
  -1965, E. Feigenbaum initiated DENDRAL, a software to deduce the molecular structure of organic compounds. It was the first expert systems.
  1965年，E·費根鮑姆開創了Dendral，一個推斷有機化合物分子結構的軟件。這是首套專家系統。
  
• In 1970 s, Holland proposed genetic algorithm, which marked the beginning of evolutionary computing.
  In 1970s, Holland 提出了遺傳算法，標誌着演化計算研究的開始。
• In 1972, A.Comerauer of the University of Marseille, France, proposed and implemented the logical programming language PROLOG.
  1972年法國馬賽大學的科麥瑞爾提出並實現了邏輯程序設計語言PROLOG。
  -1974, T. Shortliffe demonstrated MYCIN program, a very practical rule-based approach to medical diagnoses.
  1974年，肖特列夫演示了MYCIN程序，一個非常實用的基於規則的醫學診斷方法。
• Since 1956, the study of AI has achieved many remarkable achievements in machine learning, theorem proving, pattern recognition, problem solving, expert systems and artificial intelligence languages.
  1956年以後，人工智能的研究在機器學習、定理證明、模式識別、問題求解、專家系統及人工智能語言等方面都取得了許多引人矚目的成就 。
• In 1969, the International Joint Conference on Artificial
  Intelligence（IJCAI） was established. （國際人工智能聯合會
  議）
  -In 1970, the International Journal named 《Artificial Intelligence》
  started its publication.

1974–1980, The First AI Winter

• In the late 1960s, AI research encountered difficulties, such as machine translation.
  20世紀60年代末，人工智能研究遇到困難，如機器翻譯。
  -1966, the machine translation failed. 1966年，機器翻譯失敗了。
  -The report of the US Advisory Board in 1966 affirmed that there is no universal scientific text machine translation, and the short term outlook for success remains gloomy.
  1966年美國顧問委員會的報告裁定：還不存在通用的科學文本機器翻譯，有很近的實現前景。
  -The United Kingdom and the United States suspended funding for most machine translation projects.
  英國、美國中斷了大部分機器翻譯項目的資助。
  -1970, the connectionism was abandoned. 1970年，連接主義遭到遺棄。
  AI兩大學派
  ——連接主義：採用神經網絡結構不是知識
  ——符號主義：採用符號表示知識（廣義表、LISP)
  -1971–1975, DARPA（美國國防部高級研究計劃局）felt frustrated at the Speech Understanding Research program at Carnegie Mellon University.
  1971年至75年，美國DARPA對卡內基梅隆大學的語音理解研究項目感到沮喪。
  -1973, the large decrease in AI research in the United Kingdom, in response to the Lighthill report “Artificial Intelligence: A General Survey”.
  1973年，受萊特希爾的“人工智能：綜合調查”報告的影響，英國大幅度縮減AI的研究。
• 1973-1974，美國DARPA削減了一般性AI學術研究經費。t

Jokes about English-Russian translation

(1) The spirit is willing but the flesh is week. （心有餘而力不足）
The vodka is strong but meat is rotten. （伏特加酒雖然很濃，但肉是腐爛的）
The reason for this error ：Spirit：1）精神 2） liquor (烈性酒)
(2) Out of sight, out of mind (blind and insane)
“眼不見，心不煩”vs.“又瞎又瘋”
(3) Time flies like an arrow.
“光陰似箭” vs. “蒼蠅喜歡箭”

Conclusion：

Only if the machine can understand , it can translate correctly.
and understanding requires knowledge.
結論：必須理解才能翻譯，而理解需要知識.

The mistake lies in the literal translation and not understanding.
錯誤在於僅字面翻譯，並非理解了。

1980–1987, AI Boom

• In 1977, Feigenbaum proposed the concept of “knowledge
  engineering” at the 5th IJCAI, which promoted knowledge centered research.
  1977年，費根鮑姆在第五屆國際人工智能聯合會議上提出了“知識工程”概念，推動了知識爲中心的研究。
• Knowledge is power–-Bacon;
• In the Knowledge lies the power–-Feigenbaum
  -The world entered the era of knowledge engineering. Knowledge
  representation and reasoning have made a breakthrough.
  進入知識工程時代，知識表示與推理取得了突破。
• Since 1978, China has taken “intelligent simulation” as the main
  research topic of national science and technology development
  planning.
  我國自1978年開始把“智能模擬”作爲國家科學技術發展規劃的主要研究課題。
• In 1981, the Chinese Society of Artificial Intelligence (CSAI) was
  established. 1981年成立了中國人工智能學會。
• Nowadays, AI has become a key technology in many fields such as computer, aerospace, military equipment, and industry.
  現在，人工智能已經成爲計算機、航空航天、軍事裝備、工業等衆多領域的關鍵技術。
• 1980, First National Conference of the American Association for
  Artificial Intelligence (AAAI) held at Stanford.
  1980年，美國人工智能學會（AAAI）在斯坦福大學召開了第一屆全國大會。
• 1982, Japan started Fifth Generation Computer System (FGCS)
  project for knowledge processing.
  1982年，日本啓動了第五代計算機系統（FGCS）項目，用於知識處理。
  
• In mid-1980s, the machine learning came, when the decision tree model was invented and distributed as software. The model is visualized and is easy to explain.
  1985年，機器學習出現了，當時發明了決策樹模型並且以軟件形式推出。該模型具有可視化、易說明的特點。

機器學習算法分類：監督學習、無監督學習、強化學習
監督學習算法 (Supervised Algorithms）:在監督學習訓練過程中，可以由訓練數據集學到或建立一個模式（函數 / learning model），並依此模式推測新的實例。該算法要求特定的輸入/輸出，首先需要決定使用哪種數據作爲範例。例如，文字識別應用中一個手寫的字符，或一行手寫文字。主要算法包括神經網絡、支持向量機、最近鄰居法、樸素貝葉斯法、決策樹等。
無監督學習算法 (Unsupervised Algorithms):這類算法沒有特定的目標輸出，算法將數據集分爲不同的組。
強化學習算法 (Reinforcement Algorithms):沒有數據集，強化學習普適性強，主要基於決策進行訓練，算法根據輸出結果（決策）的成功或錯誤來訓練自己，通過大量經驗訓練優化後的算法將能夠給出較好的預測。類似有機體在環境給予的獎勵或懲罰的刺激下，逐步形成對刺激的預期，產生能獲得最大利益的習慣性行爲。在運籌學和控制論的語境下，強化學習被稱作“近似動態規劃”（approximate dynamic programming，ADP）。（騎自行車例子）
基本的機器學習算法：線性迴歸、支持向量機(SVM)、最近鄰居(KNN)、邏輯迴歸、決策樹、k平均、隨機森林、樸素貝葉斯、降維、梯度增強

• Also in mid-1980s, multilayer Artificial Neural Networks (ANN) was invented. With enough hidden layers, a ANN can express any function, thus overcoming the limitation of perceptron.
  1985年，還發明瞭多層人工神經元網絡（ANN）。具有足夠多的隱藏層，一個ANN可以表達任意的功能，因此突破了感知的侷限性。
  

1987–1997, The Second AI Winter

• 1987, the LISP machine market collapsed.
  1987年，LISP 機(專家系統)的市場崩潰。（符號主義的代表）
• 1988, new funding on AI were cancelled by the United States government‘s Strategic Computing Initiative.
  1988年，美國政府的戰略計算促進會取消了新的AI經費。
• 1993, expert systems slowly slipped to the valley bottom.
  1993年，專家系統緩慢滑向低谷。
• 1990s, the fifth-generation computer project disappeared quietly because it did not achieve its original goals. (1982提出）
  1990年代，日本第五代計算機項目未能達到其初始目標，悄然退場。

1997–Present, Breakthrough

• In 1997-5-12, Deep Blue beat a reigning world chess champion, Garry Kasparov, and became the first computer chess-playing system. This event set off AI boom.
  1997年，深藍戰勝了衛冕國際象棋冠軍加里·卡斯帕羅夫，成爲第一臺計算機國際象棋系統。這一事件又掀起了AI熱潮。
• It marks the successful application of AI in game.
  它標誌着AI在博弈中的成功應用。
  （AI戰勝人類的要素：DP算法、算力、大數據，遊戲行業滿足）

---

• In 1956, Samuel developed the checkers program, which can learn from the checker games, but also from the practice of checkers.
  1956年，塞繆爾研製出了跳棋程序，這個程序能從棋譜中學習，也能從下棋實踐中提供棋藝。

• In 1959 it defeated Samuel himself; in 1962 it defeated a state champion.
  1959年，它擊敗了塞繆爾本人；1962年它擊敗了一個州的冠軍。

• In August 1991, IBM‘s Deep Thought computer system and Australian chess champion Johansson held a human-computer confrontation, which ended at a draw of 1:1.
  1991年8月，IBM的“深思”計算機系統與澳大利亞象棋冠軍約翰森舉行了人機對抗賽，以1:1平局告終。

• In 1996, IBM invited the chess champion Kasparov to fight with Deep Blue developed by IBM and Kasparov won at 4:2.
  1996年，IBM邀請國際象棋棋王卡斯帕羅夫與IBM研製的“深藍”計算機系統進行了六局的人機大戰，最終，卡斯帕羅夫以4:2獲勝。

• In 1997, Deep Blue beat Kasparov at 3.5:2.5.
  1997年，深藍終於以3.5:2.5打敗了Kasparov （1勝2負3平）。

• In 1957, Simon predicted that the computer could beat the world champion in 10 years, but 40 years later Deep Blue beat Kasparov, 30 years later than the prediction.
  1957年，西蒙曾預測：10年內計算機可以擊敗人類世界冠軍，雖然未成功，但40年後“深藍”擊敗了國際象棋棋王卡斯帕羅夫，比預測晚了30年。

---

• In 2005, a Stanford’s Stanley, an autonomous robotic vehicle, won the DARPA Grand Challenge.
  2005年，斯坦福的自主機器人車輛Stanley，贏得了DARPA無人駕駛汽車挑戰賽。
• In 2006, the term “deep learning” gained attention after a publication by Geoffrey Hinton and Ruslan Salakhutdinov.
  2006年，在傑弗裏·辛頓和魯斯蘭·薩拉赫丁諾夫在科學雜誌上發表了有關“深度學習”的論文之後，該術語成了熱門。
• Watson, the computer was specifically developed by IBM to answer questions on the quiz show Jeopardy!
  沃森，是IBM專門開發的、在智力競賽Jeopardy!回答問題的計算機。
• In 2011, Watson competed on Jeopardy! against former winners Brad Rutter and Ken Jennings.
  2011年，沃森在Jeopardy!上戰勝了上屆冠軍布拉德·路特和肯恩·詹寧斯。
• Watson received the first prize of $1 million.
  沃森獲得了1百萬美元大獎。

Note：Jeopardy! is an American television game show created by Merv Griffin. The show features a quiz competition in which contestants are presented with general knowledge clues in
the form of answers, and must phrase their responses in the form of questions. The original daytime version debuted on NBC on March 30, 1964, and aired until January 3, 1975.
危險！是 Merv Griffin創作的美國電視遊戲節目。該節目以競賽爲特色，競賽者以答案的形式呈現一般知識線索，並且必須以問題的形式表達他們的回答。最初的白天版本於1964年3月30日在 NBC上首次亮相，播出時間直到1975年1月3日。

• In 2011, Google started Deep Learning project, Google Brain, as one of the Google X projects.
  2011，谷歌啓動了深度學習項目，谷歌大腦，作爲 Google X (research lab) 項目之一。
  Google brain is a cluster of 16,000 computers dedicated to mimicking some aspects of human brain activity.
  谷歌大腦是由1萬6千臺計算機連成的一個集羣，致力於模仿人類大腦活動的某些方面。
  It had successfully recognized a cat based on 10 million digital images.
  通過1千萬張數字圖片的學習，已成功地學會識別一隻貓。
• In 2012, Siri (Speech Interpretation & Recognition Interface ) was introduced by Apple as an integral part of iOS since iOS 5, running from iPhone 4S.
  2012年蘋果公司引進了Siri，從iPhone 4S上運行的iOS5開始，已作爲iOS的一個組成部分。
• 2012年，計算機視覺界頂級比賽ILSVRC中，多倫多大學Hinton團隊所提出的深度卷積神經網絡結構AlexNet（8層）一鳴驚人（圖像分類錯誤率降低11%），同時也拉開了深度卷積神經網絡在計算機視覺領域廣泛應用的序幕。
• Siri is an intelligent personal assistant and knowledge
  navigator. Siri 是一種智能個人助理和知識導航軟件。
  Use a natural language user interface to answer questions, make recommendations, and perform actions.
  使用自然語言用戶接口來回答問題、做出建議和執行動作。
  Available in: English, French, German, Japanese, Chinese, Korean, Italian, Spanish. 支持英語、法語、德語、日語、中文、韓文、意大利語、西班牙語。
• In 2012, Rick Rashid, Microsoft’s Chief Research Officer, demonstrated a real-time English-to-Chinese universal translator that keeps your voice and accent.
  2012年,瑞克·拉希德，微軟首席研究官，演示了一款實時的英文-中文通用翻譯系統，可以保持你的聲音和口音。
  Not only is the translation very accurate, but the software also preserves the user’s accent and intonation.
  該軟件不僅翻譯非常準確，而且能夠保持講者的口音和語調。
• Apr. 2014,Microsoft demonstrated “Cortana”, an intelligent personal assistant on Windows Phone.
  2014年4月，微軟演示了“Cortana”，一款運行在Windows Phone上的智能個人助理。
• Jun. 2014, Microsoft China released chatbot “XiaoIce(小冰)” which allowed WeChat users to have conversations with it.
  2014年6月，微軟中國推出了聊天機器人小冰，微信用戶可與她交談。
• On Sept. 8, 2015, Baidu launched a robot assistant — Duer at the
  2015 Baidu World Congress, which provides a secretarial search service for users.
  2015年9月8日，百度在2015百度世界大會上推出了一款機器人助理—度祕，可以爲用戶提供祕書化搜索服務。
• Jun. 2014, chatbot Eugene Goostman, at a contest marking the 60th anniversary of Turing‘s death, 33% of the event’s judges thought that Goostman was human, so that the event’s organizer considered it to have passed Turing’s test.
  2014年6月，聊天機器人尤金·古斯特曼，在紀念圖靈逝世60週年的一個比賽上，被該活動33%的評委認爲古斯特曼是人類，因此組織者認爲它已經通過了圖靈測試。
  Eugene Goostman is developed in Saint Petersburg in 2001 by a group of three programmers.
  尤金·古斯特曼是由三個程序員小組於2001年在聖·彼得堡開發的。
• Aug. 2014, IBM announced “TrueNorth”chip to work like human brain.
  2014年8月，IBM發表了類人腦工作的TrueNorth芯片。
  TrueNorth is a neuromorphic CMOS chip, consists of 4096 hardware cores, each one simulating 256 programmable silicon “neurons” for a total of just over a million neurons. TrueNorth是一款神經形態的CMOS芯片，由4096個硬件核組成，每個仿真256個可編程的硅神經元，總計剛好超過百萬個神經元。 (4096*256=1,048,576)
• On April 9th, 2016, “I am a singer” came to a close in the fourth season finals. Li Wen won the championship. According to reports, before the final result was announced, Ali Cloud xiao Ai predicted that Li Wen won the championship.
  2016年4月9日，《我是歌手》第四季總決賽落下帷幕，李玟奪得總冠軍。據報道，在決賽結果宣佈之前，阿里雲小Ai 就預測到了李玟奪冠。
  Xiao Ai is an AI program developed by Alibaba Cloud. It is based on the principles of neural network, social computing, and emotional perception. It is good at understanding the essence and real-time prediction, and can understand human emotions. It can continuously evolve itself through powerful computing and machine learning capabilities.
  小Ai 是阿里雲研發的人工智能程序，主要基於神經網絡、社會計算、情緒感知等原理工作，善於洞察本質和實時預測，並能理解人類情感，可以通過強大的計算和機器學習能力不斷自我進化。
• In 2010, AI programmer and neuroscientist Demis Hassabis et al. co-founded DeepMind, a cutting-edge artificial intelligence company, located in London.
  2010年，人工智能程序師兼神經科學家戴密斯·哈薩比斯(Demis Hassabis)等人聯合創立了DeepMind，是前沿的人工智能企業，位於英國倫敦。
  The company combines state-of-the-art technology in machine learning and system neuroscience to establish a powerful general-purpose learning algorithm.
  該公司結合機器學習和系統神經科學的最先進技術，建立了強大的通用學習算法。
  In January 2014, Google bought DeepMind. for $400m
  2014年1月，谷歌斥資4億美元收購DeepMind。
  Feb. 2015, Google DeepMind published Deep Q-Network, the human-level control through deep reinforcement learning.
  2015年2月，谷歌DeepMind公司在Nature雜誌上發表了Deep Q-Network，通過深度強化學習達到人類水平的操控。

---

• Dec. 2015, program AlphaGo developed by Google DeepMind beat Fan Hui, the European Go champion. Five battles, five victories.
  2015年12月，谷歌DeepMind公司的程序AlphaGo打敗了歐洲圍棋冠軍樊麾，成績5戰5勝。
• Jan. 27 2016, the announcement of the news was delayed until this day, to coincide with the publication of a paper in the journal Nature describing the algorithms used.
  這個消息直到2016年1月27日才宣佈，目的是與描述所用算法的論文在《自然》雜誌發表的時間同步。
• Deep-learning software defeats human professional for the first time.
  深度學習軟件第一次擊敗了人類職業棋手。

---

• Mar. 9-15 2016, AlphaGo played South Korean professional Go player Lee Sedol, ranked 9-dan, in Seoul, South Korea. AlphaGo won all but the fourth game.
  2016年3月8日至15日，AlphaGo在韓國首爾對壘韓國九段職業棋手李世石。AlphaGo以5戰4勝贏得了比賽。

1997.5 Deep Blue beat Kasparov, 3.5:2.5（chess）
2017.5.27 AlphaGo beat 柯潔, 3:0
2016.3 AlphaGo beat 李世石, 4 :1

AlphaGo Zero Vs AlphaGo

• AlphaGo learned about human’s 30 million chess games before it defeat humanity;阿爾法狗學習人類三千萬棋局，纔打敗人類;（監督學習）
• AlphaGo Zero was self-taught from scratch, relying on reinforcement learning without any human game and priori knowledge;
  AlphaGo Zero，譯爲阿爾法元,， 從零開始自學，沒有任何人類棋譜和先驗知識，完全依靠強化學習;
  It played around 4.9 million chess games with itself within three days, using 4 TPU （Tensor Processing Unit）;
  阿爾法元僅用4個TPU，用三天時間自己左右互博490萬棋局;
  In 2017, AlphaGo Zero beat AlphaGo with a score of 100:0.
  2017年，阿爾法元以100:0的成績完勝阿爾法狗。

---

https://www.nature.com/articles/nature24270

• Demis Hassabis (born 27 July 1976) is a British AI researcher, neuroscientist, video game designer, entrepreneur, and world-class games player.
  Demis Hassabis （1976年7月27日出生）是英國人工智能研究員，神經科學家，視頻遊戲設計師，企業家和世界級遊戲玩.
  Professor David Silver (dob c.1976) leads the reinforcement learning research group at DeepMind and was lead researcher on AlphaGo .
  David Silver 教授（dob c.1976）領導DeepMind的強化學習研究小組，並擔任AlphaGo的首席研究員。

人工智能簡明趣史

三起兩落（ Three ups and two falls ）