Course Contents TDDC17 • 16 Föreläsningar • 6 Labs • (1) Introduction to AI [Doherty] • Intelligent Agents • (2,3) Search [Doherty] • Search Seminar 1 • (4,5,6) Knowledge Representation • Bayesian Networks Introduction to Artificial Intelligence [Doherty] • Planning Some State of the Art Successes • (7,8) Uncertain Knowledge and • Machine Learning/RL Historical Precursors Reasoning [Doherty] • Machine Learning/DL Intelligent Agent Paradigm • (9,10) Planning [Kvarnström] • Reading • (11,12,13) Machine Learning [Liu] • Russell/Norvig Book (4th Ed) • (14,15) Perception and Robotics [Wzorek], [Rudol] • Additional Articles (2) • (16) Course Summary/ Discussion • Exam [Doherty] Patrick Doherty • Standard Written Exam Dept of Computer and Information Science • Completion of Labs Artificial Intelligence and Integrated Computer Systems Division http://www.ida.liu.se/~TDDC17/index.en.shtml 1 2 Course Book What is Intelligence? 4th Edition recently came out. Book Store has some copies, have ordered more It is only a word that people use to name those unknown processes with which our brains solve problems we call Much more up-to-date than 3rd edition hard. [Marvin Minsky, MIT] New Chapters Official course book But if you learn the skill yourself or understand the mechanism behind a skill, you are suddenly less impressed! 3rd Edition Our working definitions of what intelligence is must necessarily Free copy on the web change through the years. We deal with a moving target which makes it difficult to succinctly explain just what it is we target in AI. 3 4
What is Artificial Intelligence? What is Artificial Intelligence? (Agent methodology) A Definition: An agent’s behavior can be described formally as “the scientific understanding of the Stimuli an agent function which maps any percept mechanisms underlying thought and Percepts sequence to an action intelligent behavior and their embodiment in machines.” (AAAI) An agent program implements The Grand Goal: an agent function “a freely moving machine with the intellectual Commands capabilities of a human being.” (Hans Moravec, CMU) Agents interact with the environment through sensors and actuators 5 6 Some Perspectives on AI Empirical Sciences Mathematics/Engineering Fidelity to human performance Ideal concept of Intelligence Human-Centered Rationality-Centered Systems that think like humans Systems that think rationally Some State-of-the-art Thought Processes ”The exciting new effort to make computers ”The study of mental faculties through the use of think. . .machines with minds, in the full and literal computational models.” (Charniak and McDermott, Achievements in Artificial sense.” (Haugeland, 1985) 1985) Reasoning ”[The automation of] activities that we associate with human thinking, activities such as decision- ”The study of computations that make it possible to Intelligence Research making, problem solving, learning...”(Bellman, perceive, reason, and act.” (Winston, 1992) 1978) Systems that act like humans Systems that act rationally r ”The art of creating machines that perform ”Computational Intelligence is the study of the o functions that require intelligence when performed design of intelligent agents.” (Poole et al., 1998) i v by people.” (Kurzweil, 1990) a h ”The study of how to make computers do things at ”AI . . . Is concerned with intelligent behavior in e which, at the moment, people are better.” (Rich and artifacts.” (Nilsson, 1998) B Knight, 1991) 7 8
IBM’s WATSON Historically: AI and Robotics Artificial Intelligence Traditional Robotics “Brains without Bodies” “Bodies without Brains” Cultural & Technological ABB Watson - IBM Gap! Stanford AI Lab “Shakey” Big Dog Google Go-Deep Minds Now strong attempts at Integration 200 million pages of info/10 racks of 10 Power 750 servers 9 10 Google/Deep Minds Alpha Go Kiva Systems - Smart Warehouse Logistics Integration Monte-Carlo Tree search Deep Learning Extensive training using both human, computer play First computer Go program to beat a human professional Go player without handicaps on a full-sized 19x19 board Purchased by Amazon in 2012: Now called Amazon Robotics 11 12
Robotics- Boston Dynamics: ATLAS & HANDLE Deep Minds - Emergence of Locomotion Integration Integration Simulation 1:26 13 14 Socrates Aristotle (384-322 BC) Plato Aristotle What is a good argument? Historical Precursors to the Grand Idea of AI Mortal Major Premise Man All men are mortal Some Highlights Socrates is a man Minor Premise Socrates _______________ Socrates is mortal Deductive Conclusion Deduction 15 16
Automatons (1600 -) Leibniz (1646-1716) Natural Laws are capable of producing complex behavior Perhaps these laws govern human behavior? Calculus Ratiocinator • A universal artificial mathematical language • All human knowledge could be represented in this language • Calculational rules would reveal all logical relationships among these propositions Let us • Machines would be capable of carrying out such calculations Calculate! Addition Subtraction Multiplication Square root extraction 1772 Binary Arithmetic Precursors to Robotics 17 18 Boole (1815 - 1864) Frege (1848 -1925) Begriffsschrift “Concept Script” Turned “Logic” into Algebra The 1st fully developed system of logic Classes and terms (thoughts) could be manipulated encompassing all of the deductive using algebraic rules resulting in valid inferences reasoning in ordinary mathematics. Logical deduction could Boolean Logic be developed as a branch of mathematics • 1st example of formal artificial language with formal syntax Subsumed Aristotle’s syllogisms In essence Leibniz’ • logical inference as purely mechanical calculus rationator (lite) operations (rules of inference) Intention was to show that all of mathematics could be based on logic! (Logicism) 19 20
Russell’s Paradox Russell (1872 - 1970) defined recursively by 0 = {} (the empty set) Frege’s arithmetic made use of sets of and n + 1 = n ∪ { n } Principia Mathematica (Russell & Whitehead) sets in the definition of number 0 = {}, 1 = {0} = {{}}, 2 = {0,1} = {{},{{}}}, 3 = {0,1,2} = {{},{{}},{{},{{}}}} An attempt to derive all mathematical truths Russell showed that use of sets of sets from a well-defined set of axioms and can lead to contradiction inference rules in symbolic logic. Dealt with the set-theoretical paradoxes in Frege’s work Ergo...the entire development of Frege through a theory of types was inconsistent! • Extraordinary set: It is member of itself • Ordinary set: It is not a member of itself It must be one, Logicism Take the set E of ordinary sets but it is neither. A contradiction! Is E ordinary or extraordinary? 21 22 Hilbert (1862 - 1943) Hilbert’s Program 1st Problem: Decide the truth of Cantor’s Continum Hypothesis Consistency Completeness Logic from the Decidability, etc 2nd Problem: Establish the consistency of the outside axioms for the arithmetic of real numbers Only use Finitist Methods Metamathematics Proof Theory 24 problems Is 1st-order logic complete? for the Is PA complete? 20th century Logic from the inside Formal axiomatic Business as usual 23rd Problem : Does there exist an algorithm that can theories determine the truth or falsity of any logical proposition in Peano Arithmetic a system of logic that is powerful enough to represent the natural numbers? (Entscheidungsproblem) 23 24
Metamathematics Gödel (1906 - 1978) Syntax Semantics Showed the completeness of 1st-order logic in his PhD Thesis Δ ⊢ ω Δ ⊧ ω Inference Entailment Model Theory Proof Theory Develop metamathematics inside a formal logical system by encoding propositions as numbers Soundness Not too strong The logic of PM Hilbert’s 2nd Problem (and consequently PA) Completeness is incomplete Strong enough As a consequence, the consistency of the There are true mathematics of the real sentences not Consistency numbers can not be provable within the Correct proven within any logical system system as strong as PA Can not infer both ω and its negation 25 26 Gödel’s Argument Turing (1912-1954) Assume: Anything provable in PM is True Turing wanted to disprove the 23rd problem U is a proposition that states that 23rd Problem: Does there exist an algorithm that can determine the “U is not provable in PM”. truth or falsity of any logical proposition in a system of logic that is Self-referential: powerful enough to represent the natural numbers? (Entscheidungsproblem) 1. U is true: Suppose U were false. Then what it says would be false. So U would have to be provable, To do this, he had to come up with a formal and therefore True (assumption). This contradicts the supposition characterization of the generic process underlying the computation of an algorithm that U is false. 2. U is not provable in PM: Since U is true, what it says He then showed that there were functions that were not must be true. effectively computable including the 3. The negation of U is not provable in PM: Because U Entscheidungsproblem! is true, its negation (that U is provable) must be false, and therefore U is not provable in PM. As a byproduct he found a mathematical U is a true (from the outside [1]) proposition, model of an all-purpose computing machine! but an undecidable (from the inside [2,3]) proposition. 27 28
Recommend
More recommend
