Lesson X Artificial Intelligence 1 Prof. S. Lucci CSc 10500 Predicate. Unification In propositional logic, it is easy to determine that two literals cannot both be true at the same time Simply look for L and L In the predicate logic, this matching process is more complicated since the arguments of the predicates must be considered as well Example: man(John) and —man(John) isa contradictic whereas man(John) and ~man(Spot) is not! Why not? In order to determine contradictions, we need a matching procedure that compares two literals and discovers whether there exists a set of substitutions that makes them identical.Lesson X Artificial Intelligence. Prof. S. Lucci CSc 10500 The unification algorithm is a recursive procedure To attempt to unify two literals, we check if their initial predicate symbols are the same, If so, 0. k. Else they cannot be unified, regardless of their arguments example try_assassinate(Marcus, Caesar) hate(Marcus, Caesar) cannot be unified If the predicate symbols match, then we check the arguments one pair at a time. If the first matches, we continue with the second and so on. To test each argument pair, one calls the unification procedure recursively.Lesson X. Artificial Intelligence 3 Prof. S. Lucci CSc 10500 Matching Rules: * Different constants or predicates cannot match. only identical ones can * A variable can match another variable, any constant, or a predicate expression with the restriction that the predicate expression must not contain any instances of the variable being matched However We must find a single, consistent substitution for the entire literal, not Separate ones for each piece of it. To do this, we must take each substitution that we find and apply it to the remainder of the literal before we continue trying to unify them. Example: P(x, x) Ply, z) .«predicates match Next, compare x and y. Suppose we decide to substitute y for x written as ybx (note, we could have chosen xly instead ... the algorithm will simply choose one of these). Ply, x) So far, we have: ae note ylx z) If we continue, and attempt to match x and z, we produce the substitution zlx.Lesson X Artificial Intelligence Prof. 8, Lucci CSc 10500 But we cannot substitute both y and z for x. We have not produced a consistent substitution. Instead, after finding the first substitution ylx one is to make this substitution throughout the literals, yielding: Ply, y) Py, 2) Next, we would attempt to unify the arguments y and z; this succeeds with the substitution zly yielding: Ply, z) PCy, 2) ..they’re identical The entire substitution process has succeeded with a substitution that is the composition of two substitutions: (zly)(ylx) _...as with functions, start at the right end and go left << 4Lesson X Artificial Intelligence Prof. 8, Lucci CSc 10500 Usually, if there is one substitution, then there are many Exam hate(x, y) hate(Marcus, z) can be unified with any of the following: 1. (Marcuslx, zly) 2. (Marcuslx, ylz) Note | and 2 are equivalent 3. (Marcustx, Caesarly, Caesarlz) 4, (Marcusix, Poloniusly, Poloniustz) Note that substitutions 3 and 4 are more restrictive that is necessary We want the most general unifier (mgu) possible. Either substitution 1 or 2 would qualifyLesson X. Artificial Intelligence Prof. 8. Lucci CSc 10500 Unification — examples 1. P(x, y) and Pa, z) {alx, ylz} note a is a constant 2. P(x, x) and P(a, b) alx, blx is not legal! We cannot substitute two distinct values for x. Hence unification is not possible 3. ancestor(x, y) and ancestor(bill, father(bill)) {billlx, father(bill)ly} Note: father() is a function mapping, here bill to his father. 6Lesson X Artificial Intelligence 7 Prof. 8. Lucci CSc 10500 Resolution Principle ~ describes a way of finding contradictions in a database of clauses with minimum use of substitution Resolution Refutation proves a theorem by negating the statement to be proved and adding this negated goal to the set of axioms that are known (have been assumed) to be true. Resolution refutation proofs involve the following steps: 1) Put the premises or axioms into clause form 2) Add the negation of what is to be proved, in clause form, to the set of axioms. 3) Resolve these clauses together, producing new clauses that logically follow from them 4) Produce a contradiction by generating the empty clause 5) The substitutions used to produce the empty clause are those under which the opposite of the negated goal is trueLesson X Artificial Intelligence 8 Prof. 8, Lucci CSc 10500 Resolution is refutation complete; that is, the empty or null clause can always be generated whenever a contradiction in the set of clauses exists. Resolution refutation proofs require that the axioms and the negation of the goal be placed in a normal form called clause form. Clause form represents the logical database as a set of disjunctions of literals. Resolution produces a proof similar to one produced already with modus ponens. However, the inference rules are not equivalent— resolution is more general. Example*: We wish to prove “Fido will die” from the statements: “Fido is a dog” “All dogs are animals” “All animals will die” ‘© This example and the following procedure for converting to clause form are from Artificial Intelligence by George F. Luger, Addison- Wesley, 5" Ed. 2005Lesson X Attificial Intelligence Prof. 8. Lucci C Se 10500 Changing these three premises to predicates and applying modus ponens gives: 1) All dogs are animals: V(X)(dog( X) animal( X)) 2) Fido is a dog: dog{ fido) 3) Modus Ponens and {fidolx} gives: animal ( fido) 4) All animals will die: (CY Y)(animal( Y)— die Y)) 5) Modus Ponens and {fidoly) gives: diel fido) Equivalent reasoning by resolution converts these predicates to clause form: Predicate Form Clause Form 1) V(X)(dog)(X)= animal( X) © | dog X\vanimal( x) | I 2) dog fido) dog fido) 13) ¥(¥)(animal(¥)— diel ¥)) | animal Y\v diel Y) Negate the conclusion that Fido will die : “del fido)Lesson X Artificial Intelligence 10 Prof. 8, Lucci CSc 10500 Resolve clauses having opposite literals, producing new clauses by resolution. This process is often called clashing. TN dog fido) ~ dog Y)V diel Y) \ fido, Y die fido) ~die( fido)Lesson X Artificial Intelligence 11 Prof. 8. Lucci CSc 10500 Resolution — A Second Example Heads I win; tails you lose. Use resolution to show that I win. The following axioms describe the situation: . If the coin comes up heads, then I win. If it comes up tails, then you lose. . If it does not come up heads, then it comes up tails. Roepe . if you lose, then I win. Which may be represented as: . H— Wome) 10H: heads , W: win T > L(you) IT: tails, L: lose “HOT . L(you) > Wine) RwpesLesson X. Artificial Intelligence 12 Prof. S. Lucci CSc 10500 Next, our argument is converted to clause form 1. =H v W(me) 2. aT v L(you) 3. HvT 4, -L(you) v Wome) ‘Then, add the negation of the conclusion 5. =W(me) /falso in clause form Finally, we attempt to obtain a contradiction 24 =TvW(me) 6 13 Tv Wome) 7 67 W(me) 8 58 a Heontradiction! Hence W(me) iM win!Lesson X Attificial Intelligence 13 Prof. S. Lucci CSc 10500 Converting to Clause Form ‘* The following algorithm which consists of a sequence of transformations, may be used to reduce any set of predicate logic statements to clause form «The clause form that results will not be strictly equivalent to the original set of predicate calculus expressions, in that certain interpretations may be lost. * This occurs because skolemization restricts the possible substitutions for existentially quantified variables + However, these transformations may be used to reduce any set of predicate calculus expressions to a set of clauses that are inconsistent iff the original set of expressions is inconsistent. Notation: uppercase letters indicate variables: W,X,Y,Z - Lowercase letters in the middle of the alphabet denote constants or bound variables 1, m,n - early alphabetic lowercase letters a, b, ¢, d, e predicate namesLesson X Artificial Intelligence 14 Prof. S. Lucci CSc 10500 In this example X, Y, Z are variables and | a constant. (PX) ((alX) AL(X)) > (c(X, A( BY (AZ ¥, Z)) > A(XY))) v WX HX) Step 1: Eliminate implication. Recall that P>9="PVq We obtain: BL (WX)(alX)AB(X))¥(o( X, NAB YNZ) YZ) aX, Y)VOV XA X)) Step 2: Reduce the scope of negation a) >(ra)=a b) -(X)a(X) co) WY X)b(X)S(E X) 5X) a) “(aab)=navab e) ~(avb}s7annbLesson X Artificial Intelligence 15 Prof. S. Lucci CSc 10500 Using d) i. Above becomes Hi, (YX Y(a(X) v B(X)) v (CAK,1) A BY BZ eLY, Z)) v dX,Y))) v OWXHe(X)) Step 3: Standardize variable names. Rename all variables so that variables bound by different quantifiers have unique names. Example (V X)a{ X)v(W X)b( X)= V(X) al X)vW{Y)b(Y) Hence we obtain: fh, (YXM(malX) v D(X) v (0X, 1) A ( BYN BZN(eLY, Z)) v aX, ¥)) v (WHO W)) We have renamed this second occurrence of X Step 4: Move all quantifiers to the left without changing their order. iv. (VX AYN BZN VWH(a(X) v (XY) VOLK) A (ef ¥,Z)) ACK Y))) ¥ (OW)? This is referred to as prenex normal form.Lesson X Artificial Intelligence 16 Prof. S. Lucci CSc 10500 Step 5: Remove all existential quantifiers using skolemization; a process wherein a name is given to an assignment that must exist. Examples: (3x)(dod X)) may be replaced by dod fido) //fido is a skolem constant. « . (WX)(BY)(mother(X,¥)) replaced by (WV X)mother(X,m(X)) — //mQ isa skolem function. In general, the argument(s) of a skolem function will be all universally quantified variables preceding the existentially quantified variable being replaced. eo (WX YNS2Z)(V W)( foo X,Y, Z, W) is skolemized to: OW XIV YICW W( fool X,Y, F(X, ¥),) After skolemizing v. we obtain: Ve COX VWYralX) v (BOD) v (CCD) (eC, BCD) OX, ADDY) v CW) Y was replaced with f(x) and Z with g(x)Lesson X Artificial Intelligence 17 Prof. S. Lucci CSc 10500 Step 6: Drop all u1 ersal quantifiers. Vi, (a(X) v ABIX) v (e(X I) a (eC G1X), B(X)) v aX, IX) v e(W)) Step 7: Convert to Conjunctive Normal Form(CNF) ie, every expression is a conjunction of disjunction terms. Here we employ the associative and distributive properties: Vv (bv c)=(avb)ve aA(bAc}=(anb)nc Hassociative laws av(bAc)=(avb)v( anc) /distributive laws aa(bvc) Halready in clause form! (raf X) Vv AB(X) v e(X,Dv ef WY) 0 WE aX BODY WlfO), A) dK, KI) CW)Lesson X Artificial Intelligence 18 Prof. S. Lucci CSc 10500 Step 8: Call each conjunct a separate clause viii. a) 7a(X)v ab(X) v fv e( W) b) ral X)vb( Xv F(X), X) VAX, F(X) )Ve(W) Step 9: Standardize the variables apart again We know: (VX) a(X)) v bX) = (WX)aX) v (WYIB(Y) Finally, we obtain our original predicate logic expression in clause form: ix a) 7a(X)V=b(Xvel X,Nve(W) b) 7a U)v=b( U)v-0f F(U),g(U))va(U, F(U))ve(V) Note in x. b) X is rename by U and W by V.Lesson X Artificial Intelligence 19 Prof. S. Lucci CSc 10500 Resolution ~ an example Prove that the following argument is vali 1) All philosophers are Greek 2) All Greeks are happy 3) Either Zachos or Cox is a philosopher 4 5 Cox is not a philosopher Therefore, Zachos is happy. We employ the following predicates: P(X): X is a philosopher G(X): X is Greek H(X): X is happy We may then represent this argument as: 1) (WX)P(X}>G(X) 2) (WX)G(X)- H(X) 3) P( Zachos) P( Cox) 4) P(Cox) 5) OH (Zachos)Lesson X. Artificial Intelligence 20 Prof. S. Lucci CSc 10500 Next, we convert the premises to clause form. 1) PAUXV G(X) J! P>q="PVq,, all variables universally quantified 2) GYVH(Y) /remove implication as above standardize variable name 3) P( Zachos)v P( Cox} 4) ~P(Cox) Finally, we negate the conclusion, convert it to clause form and add it to our clause base. 5) 7H{( Zachos) We attempt to obtain a contradiction 34 P(Zachos) 6 16 G(Zachos) 7 ZachosIX 2,7 H(Zachos) 8 ZachoslY 58 0 Hence our argument is validLesson X Artificial Intelligence 21 Prof. $. Lucci CSc 10500 Suppose 1) is replaced by “Some philosophers are Greeks”: (3X)(P(X)A G(X) ‘Then we have: 0) P{ Tom) J/skolem constant Tom enables us 1) G{Tom) Jo remove existential quantifiers 2) 7G YV H(Y) 3) P( Zachos)v P( Cox) 4) >P( Cox) 5) 7H{( Zachos) We are unable to obtain a contradiction, Hence, this argument is invalid.Artificial Intelligence 22 Lesson X CSc 10500 Prof. S. Lucci Resolution - another example Premises: The custom officials searched everyone who entered this country who was not a VIP. Some drug pushers entered this country and they were only searched by drug pushers. No drug pusher was a VPP. Conclusion: Some of the custom officials were drug pushers Use resolution to prove that this argument is valid. Let E(X)represent “X entered this country” v(x) “X was a VIP” S(%Y) “X was searched by Y” cx) “X was a custom official” P(X) “X was a drug pusher” Premises:Lesson X Artificial Intelligence 23 Prof. 8. Lucci CSc 10500 1) (YXEIX) AVX) > (BYMS(XY) ACT 2) (AXYP(X) a E(X) A W¥NS(X,¥) > PCY))] 3) (VXI PIX) 99 UX)] Conclusion: 4) (AXILP(X)AC(X)] We transform our premises into clause form. D) (XIE) A>V(X)) > (BYASIXY) ACY] Remove implication (V XIE AVX) V(EN(SX, Y)AC(Y))] Reduce the scope of negation CVX) [E(X) v VOX) v (AYMS(XY) 0 COD)] Remove existential quantifiers via skolemization (CVX)[AE(X) v VIX) v (S(X, 8) 0 CURD)Lesson X Artificial Intelligence 24 Prof. S. Lucci CSc 10500 Drop all universal quantifiers AEX) v VX) v (S(X, (0). 6 CX) Convert to CNF (EX) v VOX) v SIX, FO) 0 (BOO) v VOX) v CO) Call each conjunct a separate clause TEX) VIX) v SUX, f()) AEX) v VOX) v CCAD) Standardize variable names apart 1. 7BIX) VIX) v SUX, 100) 2. ABIX) VIX) v CUI)Lesson X Artificial Intelligence 25 Prof. S. Lucci CSc 10500 Working on Premise 2 (AXILPLX)AE(X)A(Y YV(MX, Y)> PY) In the same manner will yield 3. Pla) /skolem constant 4, Ea) Haitto 5. S(aW)v P(W) And finally, transforming Pret (W X)[P(X)39V(X)] yields: 6. =P(X)VAVIX) Next, we negate the conclusion_(3X)[P(X) a C(X)) and convert it to clause form as well 7, -PU)v CU)Lesson X Artificial Intelligence 26 Prof. S. Lucci CSc 10500 Then we search for a contradiction 3,6 =V(a) 8) alX 24 Vea) v C(l@)) 9) alZ 89 c(f(a)) 10) 14 V(a) v Sta, fla) 11) alX 8,11 S(a, fa) 12) 512 P(f(a)) 13) Faw 713 >C(fla)) 14) f(aylU 10,14 9 15) Hempty clause Therefore, the original argument is valid.Lesson X Artificial Intelligence 27 Prof. S. Lucci CSc 10500 Resolution ~ Practice Problems 1. Consider the following argument: 1) Whoever can read is literate 2) Dolphins are not literate 3) Some dolphins are intelligent Conclusion: Some who are intelligent cannot read. A. Express the argument in the predicate logic B. Convert it to clausal form C. Prove that this argument is valid using resolution Il. A. Symbolize the following argument using the predicate names provided B. Use resolution to prove that this argument is valid 1. All mayor are Republican(M,R) 2. All Republicans are wealthy(W) 3, Either Green or Bloomberg is mayor(G,B) 4. Green is not mayor Conclusion: Bloomberg is wealthy. C. Analyze what happens to this argument if sentence 1. is replaced by 1, Some mayors are republican