Artificial Intelligence

1. Artificial Intelligence Problem solving and search (introduction)

2. “In which we look at how an agent can decide what to do by systematically considering the outcomes of various sequences of actions that it might take.” “In which we see how an agent can find a sequence of actions that achieves its goals, when no single action will do.”

3. Problem Solving • Problem-solving agent • A kind of goal-based agent that decides what to do by finding sequences of actions that lead to desirable states • But what is a “problem”? • What is a “solution”?

4. Romania Example • Imagine agent in Arad, Romania (on holiday) • Performance measure could be • Sightsee, eat good food, enjoy nightlife, etc. • Suppose have nonrefundable ticket to fly out of Bucharest tomorrow • Agent now adopts goal of getting to Bucharest on time • Courses of action that do not reach Bucharest on time can be rejected • Simplifies decision making • “Goals help organize behavior by limiting objective that agent is trying to achieve”

5. Problem-Solving Agents • [Step 1] Formulate goal • Based on current situation • Goal: a set of states in which the goal is satisfied • Actions cause transitions between world states • Need to find sequence of actions to get to goal • [Step 2] Formulate problem • Decide what actions and states to consider, given a goal • Decide how to quantify best solution • Need higher level of detail, else becomes intractable • Organize actions and states (graph structure, map) How find best sequence (path) of actions?

6. Problem-Solving Agents • [Step 3] Search • Process of looking for best action sequence (to reach goal) • Input: formulated problem • Output: solution (as action sequence) • [Step 4] Execution phase • Execute recommended actions • [Step 5] Find a new goal (repeat) Note: “offline” problem solving • Complete knowledge of problem and solution

7. Romania Example • On holiday in Romania, currently in Arad • Flight leaves tomorrow from Bucharest • [Step 1] Formulate goal • Be in Bucharest • [Step 2] Formulate problem • States: various cities • Operators: drive between cities • Best solution: shortest time • [Step 3] Search: find solution • Sequence of cities, e.g., Arad, Sibiu, Fagaras, Bucharest

8. Romania Neamt Oradea Zerind Iasi Arad Sibiu Fagaras Vaslui Rimnicu Vilcea Timisoara Pitesti Lugoj Hirsova Urziceni Mehadia Bucharest Debreta Craiova Eforie Giurgiu

9. Environment • Static • Formulating and solving problem in a fixed environment • Fully observable • All states knowable • Discrete • Cities are nodes, actions are links • Deterministic • No randomness assumed • Agent • Single

10. Well-Defined Problems and Solutions • A problem is defined by four items • Initial state • “In Arad”: In(Arad) • Actions/Operators available to agent • Successor function for state x: SUCCESSOR-FN(x) = <action, successor> pairs SUCCESSOR-FN( In(Arad) ) = {< Go(Sibiu), In(Sibiu) >, < Go(Zerind), In(Zerind)>, < Go(Timisoara), In(Timisoara)>} State Space

11. Well-Defined Problems and Solutions (con’t) • Goal test • Determines whether a given state is a goal state • Explicit: In(Bucharest) • Implicit (property): “Checkmate” • Path cost • A path is sequence of states connected by actions • Used to check if one solution path is better than another • Assigns numeric cost to each path • Should reflect performance measure • Example: Additive → sum of distances (time), number of operators executed, etc. • Example: Multiplicative → product of probabilities • A solution is a sequence of operators leading from the initial state to a goal state • Optimal solution has lowest path cost

12. Selecting a State Space • Real world is absurdly complex • State space must be abstracted for problem solving • Consider route planning GPS system in car • Abstract state • Set of real states, “In Arad” • Abstract action/operator • Complex combination of real actions • “Arad  Zerind” complex set of possible routes, detours, rest stops compressed into “change of location” • Abstract solution • Set of real paths that are solution in real world • “Arad  Sibiu  Fagara  Bucharest”

13. Example Problems • “Toy” problem • Intended to illustrate or exercise various problem-solving methods • Given a concise, exact description • Useful for comparing performance of algorithms • Real-world problem • Solutions people actually care about • Tend not to have universal description • Not everyone agree on problem description

14. Toy Problem: Vacuum World R L R L S S R R L R L R L L S S R S S L R L S S • States: • Initial state: • Actions: • Goal test: • Path cost: Integer (discrete) dirt and robot locations Any state Validity? (Left, Right, Vacuum) All squares clean Additive, each step costs 1 (???)

15. Toy Problem: The 8-puzzle 5 4 1 2 3 6 1 8 8 4 7 3 2 7 6 5 Start State Goal State Integer (discrete) locations of each tile and blank • States: • Initial state: • Actions: • Goal test: • Path cost: Any state Validity? (Left, Right, Up, Down) Matches goal configuration Additive, each step costs 1

16. In-Class Exercise • Have 2 empty water jugs • One is 3 gallon capacity, other is 7 gallon capacity • Can “add to” or “pour from” a jug • “Add water” by pouring from other jug, or by filling from a faucet • Can “empty jug” down drain • Must completely fill (no spill-over) and empty jug • e.g., pouring a full jug-7 into an empty jug-3 leaves 4 gallons left in jug-7 and 3 gallons in jug-3 • GOAL: have exactly one gallon of water in smaller jug

17. In-Class Exercise • States: ? • Initial state: ? • Actions: ? • Goal test: ? • Path cost: ?

18. In-Class Exercise • States: <gal-jug3, gal-jug7> • Initial state: <0, 0> • Actions: pourYintoX(jugX, jugY) Valid: jugY is not full, jugX is not empty Result: jugY = jugX+jugY or full. jugX = empty or remainder emptyJug(jugX) Valid: jugX is not empty Result: jugX = 0 fillFromFaucet(jugX) Valid: jugX is not full Result jugX = X • Goal test: <gal-jug3=1, gal-jug7=X> • Path cost: Additive, each step costs1?

19. In-Class Exercise • One Solution INIT: <gal-jug3=0, gal-jug7=0) fillFromFaucet(jug7) <gal-jug3=0, gal-jug7=7) pourYintoX(jug3,jug7) <gal-jug3=3, gal-jug7=4) emptyJug(jug3) <gal-jug3=0, gal-jug7=4) pourYintoX(jug3,jug7) <gal-jug3=3, gal-jug7=1) emptyJug(jug3) <gal-jug3=0, gal-jug7=1) pourYintoX(jug3,jug7) <gal-jug3=1, gal-jug7=0)

20. Real-World Problems • Route-finding/planning • GPS navigation, airline travel • Internet search • Robot navigation • Configuration layout • Etc.

21. Searching For Solutions • Have formulated problem, now need to solve it • Search through state space • Using search tree • Search node • Initial state • Test if already goal state!!! • Expand current state • Apply successor function to generate new states • Which state to examine (and expand) next? • Use search strategy

22. Searching for Solutions • Basic idea: • Offline, simulated exploration of state space by generating successors of already-explored states • Expanding states • Continue choosing, testing, and expanding until find a solution • Choice of which state to expand first (each time) determined by the search strategy

23. Romania Neamt Oradea Zerind Iasi Arad Sibiu Fagaras Vaslui Rimnicu Vilcea Timisoara Pitesti Lugoj Hirsova Urziceni Mehadia Bucharest Debreta Craiova Eforie Giurgiu

24. Arad General Search Example

25. Sibu Zerind Timisoara Arad General Search Example

26. Zerind Timisoara Arad Oradea Arad Fagaras Sibu General Search Example Rimnicu Vilcea Need to use a search strategy to choose which state to expand (Will discuss different strategies next lecture)

27. Zerind Timisoara Arad Oradea Arad Sibu Sibu Bucharest Fagaras General Search Example Rimnicu Vilcea

28. Zerind Timisoara Arad Oradea Arad Sibu Sibu Bucharest Fagaras General Search Example Rimnicu Vilcea Goal

29. Measuring Performance • Completeness • Does it find a solution when one exists? • Optimality • Is it the solution with lowest path cost? • Time complexity • How long does it take to find a solution? • Space complexity • How much memory is needed to perform the search?

30. Main Types of Search • Uninformed search • Given no information about problem (other than its definition) • Informed search • Given some idea of where to look for solutions

31. Summary • Before searching for solutions, must formulate goal and problem • A problem consists of • Initial state • Actions/Operators (successor function) • Goal test function • Path cost function • Toy vs. real-world problems • Many real world problems can fit into state space model • Search strategy needed to select states to examine • Have different performances