These exercises are intended to stimulate discussion, and some might be set as term projects. Alternatively, preliminary attempts can be made now, and these attempts can be reviewed after the completion of the book. — ae Define in your own words: (a) intelligence, (b) artificial intelligence, (c) agent, (d) fationality, (e) logical reasoning. 1.2. Read Turing’s original paper on Al (Turing, 1950). In the paper, he discusses several objections to his proposed enterprise and his test for intelligence, Which objections still carry weight? Are his refutations valid? Can you think of new objections arising from develop- ments since he wrote the paper? In the paper, he predicts that, by the year 2000, a computer will have a 30% chance of passing a five-minute Turing Test with an unskilled interrogator. What chance do you think a computer would have today? In another 50 years? 1.3 Are reflex actions (such as flinching from a hot stove) rational? Are they intelligent? 1.4 Suppose we extend Evans’s ANALOGY program so that it can score 200 on a standard IQ test. Would we then have a program more intelligent than a human? Explain, 1.5. The neural structure of the sea slug Aplysia has been widely studied (first by Nobel wreate Eric Kandel) because it has only about 20,000 neurons, most of them large and manipulated. Assuming that the cycle time for an Aplysia neuron is roughly the same as for a human neuron, how does the computational power, in terms of memory updates per second, compare with the high-end computer described in Figure 1.3? 32 Chapter 1, Introduction 1.6 How could introspection—reporting on one’s inner thoughts—be be wrong about what I’m thinking? Discuss. ‘curate? Could I 1.7 To what extent are the following computer systems instances of artificial intelligence: © Supermarket bar code scanners. © Web search engines. © Voice-activated telephone menus. @ Internet routing algorithms that respond dynamically to the state of the network. 1.8 Many of the computational models of cognitive activities that have been proposed in- volve quite complex mathematical operations, such as convolving an image with a Gaussian or finding a minimum of the entropy function. Most humans (and certainly all animals) never Jeam this kind of mathematics at all, almost no one leams it before college, and almost no ‘one can compute the convolution of a function with a Gaussian in their head, What sense does it make to say that the “vision system” is doing this kind of mathematics, whereas the actual person has no idea how to do 1.9 Why would evolution tend to result in systems that act rationally? What goals are such systems designed to achieve? Is AL a science, or is it engineering? Or neither or both? Explain. “Surely computers cannot be intelligent—they can do only what their programmers tell them.” Is the latter statement true, and does it imply the former? 1.12. “Surely animals cannot be intelligent—they can do only what their genes tell them.” Is the latter statement true, and does it imply the former? 1.13 “Surely animals, humans, and computers cannot be intelligent—they can do only what their constituent atoms are told to do by the laws of physics.” Is the latter statement true, and does it imply the former? Ath Examine the Al literature to discover whether the following tasks ean curently be solved by computers: a. Playing a decent game of table tennis (Ping-Pong). Driving in the center of Cairo, Egypt. Driving in Victorville, California. |. Buying a week's worth of groceries at the market. Buying a week’s worth of groceries on the Web. Playi . Discovering and proving new mathema Writing an intentionally funny story. Giving competent legal advice in a specialized area of law. Translating spoken English into spoken Swedish in real time. ig a decent game of bridge at a competitive level 1 theorems. . Performing a complex surgical operation. 23SROY WHEE CF e. CE PIU) CUVEE GPP OCS TUNE IPERS RCSEUE. Papers QUUUE ew search algorithms—which, remarkably, continue to be discovered—appear in journals such as Artificial Intelligence and Journal of the ACM. nec The topic of parallel search algorithms was not covered in the chapter, partly because it requires a lengthy discussion of parallel computer architectures. Parallel search became a popular topic in the 1990s in both AI and theoretical computer science (Mahanti and Daniels, 1993; Grama and Kumar, 1995; Crauser et al., 1998) and is making a comeback in the era of new multicore and cluster architectures (Ralphs et al., 2004; Korf and Schultze, 2005). Also of increasing importance are search algorithms for very large graphs that require disk storage (Korf, 2008). Explain why problem formulation must follow goal formulation. 3.2 Your goal is to navigate a robot out of a maze. The robot starts in the center of the maze eS 113 facing north. You can turn the robot to face north, east, south, or west. You can direct the robot to move forward a certain distance, although it will stop before hitting a wall. a. Formulate this problem. How large is the state space? b. In navigating a maze, the only place we need to turn is at the intersection of two or more corridors. Reformulate this problem using this observation. How large is the state space now? ¢. From each point in the maze, we can move in any of the four directions until we reach a turning point, and this is the only action we need to do. Reformulate the problem using these actions. Do we need to keep track of the robot's orientation now? d. In our initial description of the problem we already abstracted from the real world, restricting actions and removing details. List three such simplifications we made. 3.3 Suppose two friends live in different cities on a map, such as the Romania map shown in Figure 3.2. On every tum, we can simultaneously move each friend to a neighboring city on the map. The amount of time needed to move from city i to neighbor j is equal to the road distance d(i, j) between the cities, but on each turn the friend that arrives first must wait until the other one arrives (and calls the first on his/her cell phone) before the next turn can begin. We want the two friends to meet as quickly as possible. a. Write a detailed formulation for this search problem. (You will find it helpful to define some formal notation here.) b. Let D(i, j) be the straight: heuristic functions are admi: ine distance between cities i and j. Which of the following le? (i) D(i, j); (ii) 2» DG, j)s Gli) D(i, 5) /2. . Are there completely connected maps for which no solution exists? d. Are there maps in which all solutions require one friend to visit the same city twice? Show that the 8-puzzle states are divided into two disjoint sets, such that any state is able from any other state in the same set, while no state is reachable from any state in the other set. (Hint: See Berlekamp et al. (1982).) Devise a procedure to decide which set a given state is in, and explain why this is useful for generating random states. 3.5. Consider the n-queens problem using the “efficient” incremental formulation given on page 72. Explain why the state space has at least /n! states and estimate the largest n for which exhaustive exploration is feasible. (Hint: Derive a lower bound on the branching factor by considering the maximum number of squares that a queen can attack in any column.) 6 Give a complete problem formulation for each of the following. Choose a formulation that is precise enough to be implemented. a. Using only four colors, you have to color a planar map in such a way that no two adjacent regions have the same color. b. A 3-foot-tall monkey is in a room where some bananas are suspended from the 8-foot ceiling. He would like to get the bananas. The room contains two stackable, movable, climbable 3-foot-high crates.[ee LL | The algorithm should run in time linear in the size of the sentence. (Altematively, use a version of this function from the online code repository.) b. Give three examples of sentences that can be determined to be true or false in a partial model that does not specify a truth value for some of the symbols. ¢. Show that the truth value (if any) of a sentence in a partial model cannot be determined efficiently in general. 4d, Modify your PL-TRUE? algorithm so that it can sometimes judge truth from partial models, while retaining its recursive structure and linear run time. Give three examples of sentences whose truth in a partial model is nor detected by your algorithm. e, Investigate whether the modified algorithm makes TT-ENTAILS? more efficient. Which of the following are correct? False |= True. True | False (AAB) (A # B). A@ BEAVB. A BE-AVB. (AAB) > CK(A > C)v(B = C). (Cv (AAA-B)) = ((A > C)A(B > C)). (AV B) A(-C V=Dv E) F (AV B). V B) A (-C V =DV E) (AV B) A (-DV E) 4 => B) is satisfiable. AA V B) is satisfiable. Feap an oe 1 (A & B) & Chas the same number of models as (A _B) for any fixed set of proposition symbols that includes A, B, C. Prove each of the following assertions: vis valid if and only if True a. ‘alse | a. a b. ¢. o & Gif and only if the sentence (a = 3) is valid. a. e. For any a, a = (if and only if the sentence (a < 3) is valid. . c Gif and only if the sentence (a A 8) is unsatisfiable. J&™ Prove, or find a counterexample to, each of the following assertions: a. If & yor BE ¥ (or both) then (a A 8) Ey b. Ifa = (8.4) thena — Banda ¢. Ifa & (8 Vy) then a & Bora 4 (or both). YF Consider a vocabulary with only four propositions, A, B,C, and D. How many models are there for the following sentences? a. BYC. b. -AV-ABV-CV-D. «. (A > B)AAA-BACAD 7.8 We have defined four binary logical connectives. a, Are there any others that might be useful? b. How many binary connectives can there be? ¢. Why are some of them not very useful? 7.9 Using a method of your choice, verify each of the equivalences in Figure 7.11 (page 249). 7.10 Decide whether each of the following sentences is valid, unsatisfiable, ify your decisions neither. Ver- sing truth tables or the equivalence rules of Figure 7.11 (page 249). a. Smoke => Smoke Smoke = Fire . (Smoke = Fire) = (Smoke + —Fire) Smoke V Fire V Fire . (Smoke A Heat) > Fire) <= ((Smoke = Fire) V (Heat = Fire)) f£. (Smoke => Fire) => ((Smoke A Heat) => Fire) b, « a. eear Ae | > explicit sentences in the map language. :. Give three examples of facts about the physical structure of your country that cannot be represented in the map language. |. Give two examples of facts that are much easier to express in the map language than in first-order logic. Give two other examples of useful analogical representations, What are the advantages and disadvantages of each of these languages? Consider a knowledge base containing just two sentences: P(a) and P(b). Does this knowledge base entail V.r P(r)? Explain your answer in terms of models 8.3 Is the semtence 3.r,y r= y valid? Explain 8.4 Write down a logical sentence such that every world in which it is true contains exactly cone object. Consider a symbol vocabulary that contains ¢ constant symbols, py, predicate symbols ofeach arity k, and fy function symbols of each arity k, where 1 < k < A, Let the domain size be fixed at D. For any given model, each predicate or function symbol is mapped onto a relation or function, respectively, of the same stity. You may assume that the functions in the ‘model allow some input tuples to have no value for the function (i., the value is the invisible object). Derive a formula for the number of possible models for a domain with D elements, Don’t worry about eliminating redundant combinations. a. (Gra=z2) = (Vy 32y=2) Ya P(e) V5P(2) ©. Va Smart(2) v (2 =2). 8.7 Consider a version of the semantics for first-order logic in which models with empty domains are allowed, Give at least two examples of sentences that are valid according to the 316 Chapter 8 __ First-Order Logic standard set outcome makes ‘more intuitive sense for your examples. tics but not according to the new sem 8.8 Does the fact ~Spouse( George, Laura) follow from the facts Jim % George and Spouse (Jim, Laura)? If so, give a proof; if not, supply additional axioms as needed, What happens if we use Spouse as a unary function symbol instead of a binary predicate? 8.9 This exercise uses the function MapColor and predicates In(.r, y), Borders(r,y), and Country(e), whose arguments are geographical regions, along with constant symbols for various regions. In each of the following we give an English sentence and a number of can- didate logical expressions. For each of the logical expressions, state whether it (1) correctly expresses the English sentence; (2) is syntactically invalid and therefore meaningless; or (3) is syntactically valid but does not express the meaning of the English sentence, a, Paris and Marseilles are both in France. (i) In(Paris A Marseilles, France) A In( Marseilles, France). v In( Marseilles, (ii) In( Paris, France (iii). In( Paris, France b. There is a country th (i) 3e Country(c) A Border(c, Iraq) A Border(e, Pakistan). (i) 3¢ Country(c) = [Border(c, Iraq) A Border(c, Pakistan) (iii) [Be Country(c)] + [Border(e, frag) A Border c, Pakistan) (iv) 3¢ Border(Country(e), Iraq A Pakistan), ance ) borders both Iraq and Pakistan, ¢. All countries that border Ecuador are in South America. (i) Ve Country(c) A Border(e, Ecuador) = In(c, SouthAmerica) (ii) Ve Country(c) = [Borter(c, Beuador) = In(c, SouthAmerica)] (iii) Ve [Country(e) = Border(e, Beuador)] => In(c, SowthAmerica). (iv) Ve Country(c) \ Border(e, Beuador) \ In(e, SouthAmerica) 4. No region in South America borders any region i () “Bed In(c, SouthAmerica) A In(d, Gii) Ve,d [In(c, SouthAmerica) A In(d, Burope)] = ~Borders(c, d) (iii) Ve In(c, SouthAmerica) = 3d In(d, Europe) A >Borders(c, d) (iv) Ve Inc €, No two adjacent countries have the same map color, ope. rope) \ Borders (c,d)| ‘outhAmerica) => Vad Ind, Europe) => ~Borders(c,d) () We,y >Country() ¥ >Country(y) V >Borders(.r, y) V (MapColor(x) = MapColor(y)) (il) ¥e,y (Country) A Country(y) A Borders(.x,y) A (2 = y)) > (MapColor(x) = MapColor(y)) (iii) Ve,y Country(x) A Country(y) A Borders(x,y) A (MapColor(x) = MapColor(y)) (iv) Vr,y (Country(x) A Country(y) A Borders(.r,y)) = MapColor(x # y).ee ee ee ee ee ee eS ee Te ee ee A Source Book in Mathematical Logic (van Heijencort, 1967). Textbooks geared toward automated deduction include the classic Symbolic Logic and Mechanical Theorem Prov- ing (Chang and Lee, 1973), as well as more recent works by Duffy (1991), Wos er al. (1992), Bibel (1993), and Kaufmann ef al. (2000). The principal journal for theorem proving is the Journal of Automated Reasoning; the main conferences are the annual Conference on Auto- mated Deduction (CADE) and the International Joint Conference on Automated Reasoning (CAR). The Handbook of Automated Reasoning (Robinson and Voronkov, 2001) collects Papers in the field. MacKenzie’s Mechanizing Proof (2004) covers the history and technology of theorem proving for the popular audience. EXERCISES wun Prove that Universal Instantiation is sound and that Existential Instantiation produces an inferentially equivalent knowledge base. 9.2 From Likes(Jerry, IceCream) it seems reasonable to infer 3.x Likes (x, IceCream). eXSIENIM. Write down a general inference rule, Existential Introduction, that sanctions this inference. State carefully the conditions that must be satisfied by the variables and terms involved, Exel 361 Suppose a knowledge base contains just one sentence, 32 AsHighAs(x, Everest) Which of the following are legitimate results of applying Existential Instantiation? a, AsHighAs(Bverest, Everest). b. AsHighAs( Kilimanjaro, uerest). ¢. AsHighAs( Kilimanjaro, Everest) A AsHighAs( BenNevis, Everest) (after two applications). 9.4 For each pair of atomic sentences, give the most general unifier if it exists: P(A, B, B), P(,y,2). Qu, GA, B)), AG(x,2),u)- Older(Father(y),y), Older (Father (zx), John). Knows(Father(y), y), Knows(x, 2). ere = 9.3 Consider the subsumption lattices shown in Figure 9.2 (page 329). Construct the lattice for the sentence Employs(Mother(John), Father(Richard)) b. Construct the lattice for the sentence Employs(IBM, y) (“Everyone works for IBM”). Remember to include every kind of query that unifies with the sentence. ¢. Assume that STORE indexes each sentence under every node in its subsumption lattice. Explain how FETCH should work when some of these sentences contain variables; use as examples the sentences in (a) and (b) and the query Employs(x, Father(z)). fs Write down logical representations for the following sentences, suitable for use with (Generalized Modus Ponens: . Horses, cows, and pigs are mammals. An offspring of a horse is a horse. a b. ¢. Bluebeard is a horse. 4. Bluebeard is Charlie’s parent. ¢. Offspring and parent are inverse relations. £. Every mammal has a parent. 9.7 These questions concern concem issues with substitution and Skolemization. a. Given the premise Vr Sy P(-r,y), it is not valid to conclude that 3q P(g. q). Give an example of a predicate P where the first is true but the second is false. b. Suppose that an inference engine is incorrectly written with the occurs check omitted, so that it allows a literal like P(x, F(.r)) to be unified with P(q,q). (As mentioned, most standard implementations of Prolog actually do allow this.) Show that such an inference engine will allow the conclusion 3y (qq) to be inferred from the premise Ya dy P(z,y) 362 Chapter 9. Inference in First-Order Logic ¢. Suppose that a procedure that converts first-order logic to clausal form incorrectlyee rere See ee er ne shows that constraint-based approaches such as GRAPHPLAN and SATPLAN are best for NP- hard domains, while search-based approaches do better in domains where feasible solutions can be found without backtracking. GRAPHPLAN and SATPLAN have trouble in domains with many objects because that means they must create many actions. In some cases the problem can be delayed or avoided by generating the propositionalized actions dynamically, only as needed, rather than instantiating them all before the search begins. Readings in Planning (Allen et al., 1990) is a comprehensive anthology of early work in the field, Weld (1994, 1999) provides two excellent surveys of planning algorithms of the 1990s. It is interesting to see the change in the five years between the two surveys: the first concentrates on partial-order planning, and the second introduces GRAPHPLAN and SATPLAN. Automated Planning (Ghallab et al., 2004) is an excellent textbook on all aspects of planning. LaValle’s text Planning Algorithms (2006) covers both classical and stochastic planning, with extensive coverage of robot motion planning. Planning research has been central to AI since its inception, and papers on planning are a staple of mainstream AI journals and conferences. There are also specialized conferences such as the Intemational Conference on Al Planning Systems, the International Workshop on Planning and Scheduling for Space, and the European Conference on Planning. ST ee ee ee ee TT ee meee eT en ee XERCISES xercises he Describe the differences and similarities between problem solving and planning. 10.2 Given the action schemas and initial state from Figure 10.1, what are all the applicable concrete instances of Fly(p, from, to) in the state described by At(P,, JFK) A At(P2, SFO) A Plane(P,) A Plane(P2) A Airport(JEK) A Airport(SFO) ? 10.3. The monkey-and-bananas problem is faced by a monkey in a laboratory with some bananas hanging out of reach from the ceiling. A box is available that will enable the monkey to reach the bananas if he climbs on it. Initially, the monkey is at A, the bananas at B, and the box at C. The monkey and box have height Low, but if the monkey climbs onto the box he will have height High, the same as the bananas. The actions available to the monkey include Go from one place to another, Push an object from one place to another, ClimbUp onto or ClimbDown from an object, and Grasp or Ungrasp an object. The result of a Grasp is that the monkey holds the object if the monkey and object are in the same place at the same height. a. Write down the initial state description. 397 Switch 4 q Door 4 Room 4 Switch 3; tt : Door 3 Room 3 " Shakey ‘Switch 2 Corridor q Door 2 Room 2 — Switch 1] Box 3) Box 2 N Door 1 Room | Box | Figure 10.14 Shakey’s world. Shakey can move between landmarks within a room, pass through the door between rooms, can climb climbable objects and push pushable objects, and can flip light switche: b. Write the six action schemas. c. Suppose the monkey wants to fool the scientists, who are off to tea, by grabbing the bananas, but leaving the box in its original place. Write this as a general goal (i.e., not ming that the box is necessarilv at C) in the lansuave of situation calculus. Can thisVwvoMaLY ‘The mapping that determines which actions are allowed is the Turing machine’s program. Your goal is to control the machine into the accept state. Represent the Turing machine acceptance problem as a planning problem. If you can do this, it demonstrates that determining whether a planning problem has a solution is at least as hard as the Turing acceptance problem, which is PSPACE-hard. 10.6 Explain why dropping negative effects from every action schema in a planning prob- lem results in a relaxed problem. 10.7 Figure 10.4 (page 371) shows a blocks-world problem that is known as the Sussman anomaly. The problem was considered anomalous because the noninterleaved planners of the early 1970s could not solve it. Write a definition of the problem and solve it, either by hand or with a planning program. A noninterleaved planner is a planner that, when given two subgoals G and G2, produces either a plan for Gy concatenated with a plan for G2, or vice versa. Explain why a noninterleaved planner cannot solve this problem, __ er Prove that backward search with PDDL problems is complete. 10.9 Construct levels 0, 1, and 2 of the planning graph for the problem in Figure 10.1. 10.10 Prove the following assertions about planning graphs: a. A literal that does not appear in the final level of the graph cannot be achieved. 399 b. The level cost of a literal in a serial graph is no greater than the actual cost of an optimal plan for achieving it. 10.11 The set-level heuristic (see page 382) uses a planning graph to estimate the cost of ieving a conjunctive goal from the current state. What relaxed problem is the set-level heuristic the solution to? 10.12 Examine the definition of bidirectional search in Chapter 3. a. Would bidirectional state-space search be a good idea for planning? b. What about bidirectional search in the space of partial-order plans? ¢, Devise a version of partial-order planning in which an action can be added to a plan if its preconditions can be achieved by the effects of actions already in the plan. Explain how to deal with conflicts and ordering constraints. Is the algorithm essentially identical to forward state-space search? 10.13 We contrasted forward and backward state-space searchers with partial-order plan- . saying that the latter is a plan-space searcher. Explain how forward and backward stat ne space search can also be considered plan-space searchers, and say what the plan refinement operators are 10.14 Uptonow we have assumed that the plans we create always make sure that an action’s preconditions are satisfied. Let us now investigate what propositional successor-state axioms such as HaveArrow't! < (HaveArrow' A >Shoot") have to say about actions whose preconditions are not satisfied. a. Show that the axioms predict that nothing will happen when an action is executed in a state where its preconditions are not satisfied, b. Consider a plan p that contains the actions required to achieve a goal but also includes illegal actions. Is it the case that initial state \ successor-state axioms \ p goal ? ¢. With first-order successor-state axioms in situation calculus, is it possible to prove that a plan containing illegal actions will achieve the goal? ale axioms of 10.15 Consider how to translate a set of action schemas into the successor- situation calculus. a. Consider the schema for Fly(p, from. to). Write a logical definition for the predicate Poss(Fly(p, from, to), s), which is true if the preconditions for Fly(p, from, to) are satisfied in situation s. b. Next, assuming that Fly(p. from, to) is the only action schema available to the agent, fate axiom for At(p, r,s) that captures the same information write down a successor-s as the action schema.