G51IAI Introduction to Artificial Intelligence

1. G51IAIIntroduction to Artificial Intelligence Andrew Parkes http://www.cs.nott.ac.uk/~ajp/

2. Why study AI anyway?

3. Because it is �cutting edge research� What is the most complicated and impressive thing that you know of? Your own brain Curious how it works?

4. Why do we need AI anyway? Will now see selected examples of currently active areas in which AI, and AI techniques, are important

5. Autonomous Vehicles Think of �robot wars� but with a computer doing the controls � not a human on the radio control Suppose you are going to take a winning robot from robot wars, and make it autonomous, and still win. What software components do you think you will need? Partial List: Vision processing Path planning don�t want to fall into holes Action planning: �if I move to the right of my opponent then I can use my hammer to attack the weak spot on their left flank�

6. Robosoccer Program robots to play football Autonomous not remote controlled by humans! There is work on this at Nottingham! http://grumpy.cs.nott.ac.uk/~robots/wiki/index.php/The_Robot_Soccer_Project http://grumpy.cs.nott.ac.uk/~robots/wiki/index.php/3rd_Year_Project_Ideas

7. DARPA grand challenge Take an all-terrain vehicle or SUV Give it a camera Give it a computer Connect computer to steering controls etc Have it drive by itself 130 miles through the desert Racing against others For $2 million in cash http://www.darpa.mil/grandchallenge05/gcorg/index.html

8. DARPA grand challenge

9. Europa Europa is one of the moons of Jupiter It has an ice surface, and it is believed possible that there is an ocean underneath (like our artic) Oceans might support life even in extreme conditions � e.g. thermal vents �Life on Europathe hunt for life on Mars will be a fossil hunt - the hunt for life on Europa will be for actual life.� http://www.resa.net/nasa/europa_life.htm �Nasa is preparing plans for a possible mission to Europa that would put a lander on the surface. It has even been suggested that a mission could be devised that would drill through the ice layer and release a probe into the liquid underment. This is the "hydrobot" concept.� The probe would need to be autonomous and quite intelligent to cope with the unknown conditions

10. Europa Hydrobot http://www.resa.net/nasa/images/gem/HYDROBOT.JPG

11. �Semantic Web� Current web pages are HTML Text written with a markup language of �tags� to tell the browser what to do Hypertext links �click here to go to a different page� Structural Markup Only E.g. <h2>This is a level 2 header</h2> <p>This is a paragraph with <b>this</b> in bold</p>

12. �Semantic Web� Want to move to �semantic markup� Text written with a markup language of �tags� to tell the reader the meaning of the entries In contrast, �structural markup� only says how to format them on the page Example: XML, e.g. <person> <last-name>Berners-Lee</last-name> <first-name>Tim</first-name> </person>

13. �Semantic Web� There is a lot of work on �Ontologies�: Ontology = Sets of rules for the meanings of the entries in a semantic markup language (extensions of XML) Want to use the ontologies in order to verify that the information is consistent with the ontology e.g. for housing might have tags for house type and numbers of floors and could check that a �bungalow� has only one floor deduce new information e.g. that the bungalow only has one floor even if we are not told so explicitly

14. �Semantic Web� For example, a current proposed standard is OWL �The OWL Web Ontology Language is designed for use by applications that need to process the content of information instead of just presenting information to humans.� http://www.w3.org/TR/owl-features/ If you are interested in web technologies then you might want to do web searches for terms such as �semantic web�, XML, RDF, OWL, etc Reasoning with and about ontologies can give very difficult search problems far harder than we will do in this course but often using extensions of the techniques we will talk about

15. Informal Notion of �AI completeness� Many problems once you delve deep enough to end up requiring a large subset of the AI techniques Sometime known, very informally, as �AI completeness� Means that if can you solve one of these deep hard problems well, then the techniques will probably help you solve others Do not need to worry too much about whether the area of study too restricted because the techniques might well be portable

16. Why is AI hard? Two usual ingredients (for standard AI) Representation need to represent our knowledge in computer readable form Reasoning need to be able to manipulate knowledge and derive new knowledge many possible ways to do this, but most give rubbish finding the successful way usually involves search Both of these are hard.

17. Why is search hard? The �Combinatorial Explosion�

18. The Travelling Salesman Problem (TSP) A salesperson has to visit a number of cities (S)He can start at any city and must finish at that same city The salesperson must visit each city only once For example, with 5 cities a possible tour is:

19. The Travelling Salesman Problem The cost of a solution is the total distance travelled Solving the TSP means finding the minimum cost solution: Given a set of cities and distances between them then find the optimal tour, that is, the shortest possible such tour

20. The Travelling Salesman Problem Suppose we have n cities For 1st city have n choices For 2nd city have n-1 choices � n * (n-1) * (n-2) *� * 1 = n! Hence, have n! sequences But reversing the sequence is considered the same tour Hence n!/2 possible tours

21. Combinatorial Explosion

22. Combinatorial Explosion

23. Scaling properties Big Oh notation: O(f(n)) n is a measure of the size of the problem function f(n) is an upper bound on the asymptotic (large n) behaviour (of the runtime) ignores constant factors e.g. ( 5 n2 + 4 n ) is O(n2) �Polynomial� means O(n), O(n2), etc. That is, O(nk) for some fixed k �Exponential� means O(2an) for some fixed a. E.g. the runtime might be O(2n/20)

24. Scaling properties Why is exponential so much worse than polynomial? Suppose that have two problems A and B with scaling properties for the runtimes of the best algorithms: A: O(n2) polynomial (quadratic) B: O(2n/20) exponential Suppose that currently, for A and B, can manage to solve problems of size n=1000 in 10 mins Now, suppose we need to solve n=1100 A: will take (1100/1000)2 * 10 = 12.1 mins B: will take (21100/20/21000/20) * 10 = 2100/20 * 10 = 320 mins Polynomial scaling: as problem gets bigger runtimes increase at a reasonable speed Exponential scaling: as problem gets bigger, runtimes get worse very rapidly For more examples see the course web pages

25. The Travelling Salesman Problem �Decision version� Given a set of cities and distances between them, and also given some upper bound D on the total distance to be travelled then answer Yes/No to whether or not there exists some tour of distance D or less

26. The Sorting Problem Sorting: Given a set of cities and their populations then find a tour such that cities are visited in order of increasing population There are still an enormous potential number of tours But it is easy to find an �ordered� tour There is a polynomial, O(n log(n) ), algorithm for sorting Maybe the large number of tours in the TSP does not preclude a polynomial time algorithm to solve it?

27. Extra Coursework Take the TSP and ... Give any polynomial time algorithm to solve the TSP ... or show why a polynomial algorithm cannot exist Marking scheme for correct and complete answers: you get a 100% mark on the course you get a Ph.D. you get a professorship anywhere you want you get to be on TV you get $1 million No penalty for failure to produce an answer

28. Extra Coursework Take the TSP, and give any polynomial time algorithm to solve it, or show why a polynomial algorithm cannot exist This is one of the major unsolved theoretical problems in Computer Science No-one has found a polynomial time algorithm Most believe none exists No-one can prove that a polynomial time algorithm does not exist The problem is known as the �P equal to NP?� problem

29. �TSP-like� problems Take the TSP, and give any polynomial time algorithm to solve it, or show why a polynomial algorithm cannot exist This is one of the major unsolved theoretical problems in Computer Science A lot of other problems are equivalent to the TSP Many such problems occur in AI However, in practice, one can improve the scaling of algorithms e.g from O(2n/10) to O(2n/20) such �minor� improvements can make an enormous practical difference e.g. n=200, 2n/10=1048576, 2n/20=1024 A lot of AI effort goes into improving the scaling of algorithms on such problems so that large problems can be solved in practice

30. Suggested Self-Study In preparation for the programming within the coursework: In the �Programming Language Of your Choice� (Java, etc), revise, or start to learn, how to store collections of objects queues, stacks, vectors, or whatever �linked lists� (and trees) Remark: these topics will be useful for other programming tasks in general, as well as for this course

31. Next Week: Trees & Graphs Introduction to trees and graphs Will see that trees can easily become exponentially big How this makes AI difficult and why good search techniques are essential Will start to describe blind search techniques