Introduction
What is AI ?
AI 從二戰後急速成長，而在 1956 年 被正式確立
一些定義將 AI 分為四大類型 :
Thinking Humanly
Thinking Rationally
Acting Humanly
Acting Rationally
Acting Humanly : The Turing Test
當一個 Computer 在進行 Turing test
並且詢問者在詢問完一些問題後
不知道他詢問的對象是 Computer 還是 Human
AI 就達成了 Acting Humanly
這個電腦需具備 :
Natural language processing
Knowledge representation
Automated reasoning
Machine learning
Computer vision
Robotics
The study of how to make computers do things at which, at the moment, people are better. (Rich and Knight, 1991)
Thinking Humanly : Cognitive Modeling
當電腦具備一些類似人類的思考模式時
Through introspection (反思)
Through psychological experiments
Through brain imaging
Cognitive Science 結合了 AI 和 Psychology，嘗試建立起精準又可以測試的人腦理論。
The art of creating machines that perform functions that require intelligence when performed by people (Kurzweil, 1990)
Thinking Rationally : Law of Thought
當 AI 能夠自己解決任何 logical solvable problem (Logicist tradition) 時
有兩大障礙 :
Logical notation 沒辦法輕鬆表達 informal knowledge
解決問題 in principle 和解決問題 in practice 有非常大的差別
簡單的問題都有可能有幾百種 facts 可以讓電腦資源癱瘓
The study of the computations that make it possible to perceive, reason, and act. (Winston, 1992)
Acting Rationally : Rational Agent
當沒有預設的最佳解時，AI 依然可以給出最佳解
有兩大優點 :
More general than "Laws of thought"
因為找到最佳解只是在數種答案中找到其中一種
More amenable (經得起檢驗)
AI ...is concerned with intelligent behavior in artifacts. (Nilsson, 1998)
Foundations of AI
Philosophy
智慧從哪兒來、智慧怎麼變成動作
Mathematics
什麼可以被計算、如何從不確定的資訊來推論
Economics
怎麼做可以得到最大效益
Neuroscience
大腦怎麼處理資料
Psychology
人類跟動物是怎麼思考跟動作的
Computer engineering
怎麼樣建立一個有效率的 computer
Control theory and cybernetics
如何控制人工智慧
Linguistics
語言跟思考的關係在哪裡
History of AI
The gestation of artificial intelligence (1943–1955)
McCulloch and Walter Pitts (1943) 推出一種 artificial neuron 可以計算一些 function & logic
Donald Hebb (1949) 升級並改造了兩個 neurons 的連結
Marvin Minsky and Dean Edmonds (1950) 建立第一個 neural network
Alan Turing 在 1947 年就有類似課程
The birth of artificial intelligence (1956)
John McCarthy, Marvin Minsky, Claude Shannon, Nathaniel Rochester ...
Early enthusiasm, great expectations (1952–1969)
Newell and Simon (1952-1969) 發明 General Problem Solver (GPS)
可能是第一個 "thinking humanly" approach.
Newell and Simon (1976) 定義 physical symbol system hypothesis
Nathaniel Rochester (1952-1969) 在 IBM 建立一些 AI 專案
Herbert Gelernter (1959) 建立 Geometry Theorem Prover 可以證明一些理論
Arthur Samuel (1952-) 寫出 checkers
McCarthy (1958) 定義 high-level language Lisp，在之後三十年主導 AI 領域
Advice Taker (hypothetical program)
Minsky with microworlds
McCulloch and Pitts with neural networks
A dose of reality (1966–1973)
AI researchers’ overconfidence
在 machine translation efforts 上心有餘而力不足
Scalability: 找到 principle 的 solution 不代表找到方法可以實作
fundamental limitations on the basic structures
Knowledge-based systems: The key to power? (1969–1979)
之前的方法都是 Weak methods
要突破 weak methods 必須要有更多專業知識進到 AI 領域
DENDRAL program (Buchanan et al., 1969) 可以解決 inferring molecular structure
Feigenbaum and others at Stanford developed MYCIN to diagnose blood infections
domain knowledge 的重視也在 natural language 這塊出現
AI becomes an industry (1980–present)
expert systems, vision systems, robots, and software and hardware specialized for these purposes
AI Winter
The return of neural networks (1986–present)
mid-1980s 一些科學家重新設計 Bryson and Ho (1969) 的 back-propagation learning algorithm
造就新的 connectionist models 出現，他們被視為是 symbolic models & logicist approach 的兢爭對手
AI adopts the scientific method (1987–present)
AI 理論大致上不會再被改變，並且開始出現一些 real-world applications
Speech recognition
Machine translation
Data mining
Probabilistic reasoning
The emergence of intelligent agents (1995–present)
Whole agent problem 重新出現
search engines, recommender systems, and website aggregators
AI 與更多領域結合在一起
The availability of very large data sets (2001–present)
以前的 computer science 重視 algorithm
而新的說法認為 data 更為重要一些
要表達各種 Knowledge 的方法比起 hand-coded，使用 learning methods on big data 可能更加有效
The State of the Art
Robotic vehicles
Speech recognition
Autonomous planning and scheduling
Game playing
Spam fighting
Logistics planning
Robotics
Machine translation
…
Last updated