From: AAAI-86 Proceedings. Copyright ©1986, AAAI (www.aaai.org). All rights reserved. POINTWISE CIRCUMSCRIPTION: PRELIMINARY Vladimir Department REPORT Lifschitz of Computer Science St an ford IIniversity Stanford, CA 94305 Abstract Let us accept valued Circumscription subject We propose instead an is the to restrictions a modified of being “infinite tions; notion a single formulas. of circumscription condition, of “local” conditions of changing the value of a predicate We argucl that this it becomes from than the traditional time, leads with atl&tionaJ. flexibility of commonsense at circumscrip- simpler “global” to generalizations needed in applications cm expressed be dcsc*ribcd predical,e other the intuitions behind tliey suggest dilfercsnt i(.ittC" Illigllt If a ;uitl predicate satisfying a given derslantling OII These li”. with two ap- dill’erent,, “minimizing of its a predi- equivalent; are somewhat, views on what a set tlicn predicates it is natural of a prctlicate stronger of making makes at other as minimnlity is a stronger condition without to this order. A predicate if it cannot tl~c corltlition. Ieatls to the usual This be un- they of them be required cm rc~sc~nrch was partiall<y 406 / SCIENCE remain to remain definition new expresses, icate “at every nite conjunction” same; way; fixed, or we or some and the others of circumscription intuitively, point”. minimality expresses the impossibility from true conjunction” ‘Ynfinitc by a universal that this is in some traditiortal at one point. will be represented “pointwise” approach wa,ys conceptually “global” to each of changing quantilier). We argue scription as an “inf- conditions; to fulse in of a pred- It can be interpreted of “local” (Formally, this proposed the minimality the value of a predicate nccdccl the to vary. The leads As far *IS we can require, in an arbitrary approach gc.licr;~lizat,iOIis in applications alid, with to circum- simpler at than t,he iLdtlitiOllid to the theory the the* sa111e time, Urxibility of commonsense yea- soning. Proofs be published of the mathematical facts stated below will in the full paper. Basic Cost Let us start one predicate as parameters. icate constant with the simplest with all other Let n(P) P. for Circumscripl,ion case of circumscribing non-logical constants treated b e a sentcncc containing a pred- R.ccall that A(P) p < P stands s~~pport~ctl b,y DAIU’A of Point,wiso of P in rl( 1’) is, by delinilion, definii~ion of - uodrar (:or~t,~~;\(.t,NOO:l!)-,Y't-C-0'250. are concerned, to change them Here p is a predicate This points case, that allow circurnscriplion. . - ----- at a poht ( E II” as chang- from true to false. point as now we can also think can by the mini- predicate. is minimal violating of nliniltla1it.y relative but of the “smaller” a pred- are orderc’d to urltlcrstatid sense; in the simplest but and of elements IJnderstanding smaller at that as Boolean- as families of truth words, is a family true}. a predicate ing its value 2. The ])rcYlici~l~cis A smaller made a predicate lncnn, set inclusion, mality a k-nry II is the universe Inathem;~t,ically them in a model is to represent function of coItrsc, sub- formulas. syn~bol identifies possibility by R 13oo1c;u1-vnlucd are, of predicates One is I.0 represent of I/“‘, where ‘I‘his approach is logical 1986) by predicate in two ways. The proaches 1980, of a predicate by ;I srlbsct extension. cate (McCarthy iriterprct,ation the model. still this paper that, ;s, the minimization to restrictions The predicate {false, “stronger” to reasoning. or, in other of these conditions Circumscription nhirnimtiion, Each set allowed 1. Introduction ject values. ordered the values true to f&e “pointwise” and, at the same the condi- the impbssibility approach the theory so that, minimality expresses one point. tion is conceptually of predicates by predicate minimality conjun(*tion” each of these minimization expressed now the view of predicates functions, the (glohl) circumscription the second-order A Vpn( [l(P) A p < P). variable of the sntne arity formrlla (1) as P, and (Z is a tuple Circum( of object The Notice that by Cp(A( cannot A VxjPx denote by (1) this P)). is a first-order A model quantifier mentioned minimality the as a minimality that these Several Predicates case is important, cumscription because (McCarthy To illustrate 1986) both Any and b, is a model x. Cp(A)‘. of cir- predicates in the axioms. (I) and (2), take P identically with P) and Cp(A). false In addition, at exactly of the pointwise special applications between P is true are posi- This version, any two points, a but not of the one. Even when applications wc usually gcncral 1ll0rc to formalizing common- riecd forms of circuxiisc.ription t,liati tliosc defiiicd in the previous section. Let us start A(P, Z), where % is a tn- with a formula ple of predicate and/or function constants. circumscription of P iI1 A( P, 2) Tllc (global) with 2 dlowecf A v +x(P;x minimization different is which the predicates show and that circumscrib- circumscribing A, Circum(A; Pi; Z). of PI, . . . , P,, in A are positive, also holds; to hence, in this case, /\; Cr. (A; Z). whenever one is changed in arbitrary an counterpart Let, 11s turn considrr, 7’11c c.irciiInscript,il)ll a higher s1g11s CircuIti( by the sarnc pret,ed lexicogral,bic;l.lly; circumscription p, of structure formula is the circumscription; there definition. circumscription, 14; 1’1 > to the task formula that, PI ,“‘, (3), 1’2; Z), which of minimizing but for details, is equivalent and PI, Pz. the case of two prcdicntcs priority defined This This to prioritized now is equiv- asserts predicates of A. a special for simplicity, 11; P; 2) and the interpretation of parallel need to introduce no case when way, the resulting be a model implies the converse conjunction tr?Le to false, possibly special Circum( Tl lis valuck of one of the from 2 is changed cannot In the important each P; Z) Circum(A; all occurences alent with below). between parallel to is it from the of P are minimized is discussed relationship in Pl , * - * , Cl It is easy Pi? variables 3 p,x). of several members priorities, What ing is in simple knowledge slightly of joint pointwise 3. A Generalization of predicate with as parallel circumscriptior~ (to distinguish different The by (3), i=l form case the minimized model Circum(A; of A in which called occurences the difrerence A t.o be Pu E Pb. This the again n vx p;x 2 Pi,) ( A it asserts of P in A(P) for a tuple i=l conjunction” P) implies time n by the quantifier are equivalent. this circumscription PI,. . . , P,,. is given and p < P understood Pl,***,Pn, which of A(P) of global constants P; 2) of Circum(,l; p standing “infinite in standard have no negative meaning following Circum(A; all occurences (2) generalization P is a tuple of predicate at one point. condition: two formulas (2) of A(P) model of P at point of the value If we assume tive then satisfies another aud the formula (3), We denote of (2) is a model into It is easy to check that model formula. Vx represents above, can be viewed srusc 4. Minimizing In a further is A 2 # y))]. the value of P from true to false changing usually of P in A(P) A il(Xy(Py be transformed global We circumscription pointwise A(P) The variables). A( P); P). with as- 1’1, is p < P intcr- see (l,ifschitz 1985). to to vary is A( I’, Z) A vpz- I( A(p, z) A p < P), whcrc t is variables Circum( cannot il of t4plc similar to %. tl( P, Z); P; Z), bc made smaller intcrprebations prcdicatc This asserts the Cuticbioti dcnotcd included by of P extetlsion even at the price of chauging of the symbols The corrcspotltling formula, that (9 and/or the in 2. wlic~tic~vc~r/‘, , cori.iuuc4ion cumscription: 5. A Further Our variables Z, is cquivalcnt positive. (4) is, 110 value it1 /I. ‘I’his of prioritize-d cir- of PI can be cl~i~~~g~dfrom true COUIlt?rpiWt of changing P2 arbitrarily. Generalization form of pointwise circumscri~~tion A(P, 2) A VXZl[Pa: A A(Xy(Py be denoted 0lll.y posil,ivcb occ1ircticcs tlitV(’ to false even at the price is It will I’2 is the pointwisct by C,( gCIlcsrillly, A 2 # y), z)]. A(P, Z); Z). Zk second-order to (3) iT all owllmiws Because f0~tIllll~. (4) of the (4) of I’ it1 /l( I’, Z) are forms next goal of poinlwisc tivating on the some more We start general with a mo- example. Consider which is to introduce circuliiscripliorl. a l)lock tnblc n simple catI or OIL version of the blocks world, in bc irl 0111,~ ou(’ of two i)IiL(*cs: cit,hcr ttlc floor. KNOWLEDGE We want to tlcscribc REPRESENTATION the / 407 effect of one table. This stants, particular action, can be done using ONTABLE block of bloclcs before B which and after What on the predicate and ONTABLEI, the configurations There putting two unary con- above to the circumscription new notation? values of ONTABLE represent ues corresponding the action. constant x > (ONTABLE x f ONTABLE x) (5) list and then must ONTABLE Here AB is the “abnorma1it.y” axiom expresses what they are. This tion for solving under axiom McCarthy normally, formula the frame consideration: The calls first the “com- remain where (McCarthy basic 1986). property The of the action in the new configuration AB? to be able in 2, the part fixed, be varied The purpose in the process of our axiom new configuration of blocks; AU with a circumscription set is to characterize hence ONT,4BLEl (global it is natural allowed or pointwise) predicate of minimizing the to cir- to vary. Such gives This is exactly rameters which changes pens if it was not can bc shown sanic on thr that expert; and table prior ~i~clltnscl-iI)tion to the dots or both ONTABLE fixed; in difFerent Let us change in this examplr SitJJtltiOJl CakJJhS dit,ion to not Irad the formal to tlic of (McCarthy for I)locks, supposed ‘l’lt(~tT prcdicatc to have a situation III t,hc new notation, 1969). used 111 ad- ilI‘(’ tW0 Instcnd a~14 ON’Z’A BLEI, ON7’11BLE, term as its second which argument. (5) and (6) bccornc -lAB z > (ONTABLl+, St,) G ONTAU,!,~(x, identically rather true varyiltg tflli111 get the then P nor a Consider a which 2. has no pa- Intuitively, of 2 on which we propose desired rffect, becomes it 2 may the following f- S,,); ONTABLE other than (3). ( I). V Making In the cxnmplc and WC cn~l by taking 1s ’ allowed (7) 2 as a parameter Z is ON7’,4llLE, for instance, Xys(s The (7) t 0 t,reating it; (7) hcotnc5 scctiotl, V to be to vary in situations S,,. counterpart of (7) for pointwisc circumscription is is We can allow for x to affect which cvcn more the ditional flexibility, for more of the and %. Intuitively, value not nccdcd domain on (This ad- at x. is essential Let T/’bc a X-expression equals the sulli of 1,hc arit,ics tlic function the set of all values we will write form of ~ir~utllscriI)t,ioll it possible of the in this example, 1/ reprcscnts of 2 into by making part .P is minimized c’olnplc>x applications.) XZ?LV( T, 74 w I 10s~ arity S,)) ilcxibility choice 2 may vary when T/(x, u); accordingly, / SCIENCE = gx)). one symbol, z, 2) A A(p) z) A p < P]. Calsc is equivalent fro111 the prrvious IIC~ 408 as 2, circumscription, IT I/ is idcnt,ically every and of the same arity > (fx sikrta- two sit8uations f1 on the table. ONTdl1Z;E,, q, lr .we write (“p and q are equal constant. domain to be If p, then of only arity neither of the A(P, 2) A ‘++WV( of the wc ittl ro~ltt(‘~ n scc*ottd sort S,, and S, , which rcprc~sctlt predicates language formalism and llaycs for SilflitliOtts. we have now one binary and ways. by the act ion of placing of two tmary It t,o the ON?‘Al1LE,, now slightly vnriahlcs tion constatlt,s, treat event. /I I, I<, arc b0l.h and move closer Vllriill)lPS Vilriil,l)lCS, separated we must consists its val- values useful. arit,y symbols or a function contains the part be same and vary. the oldy is H, and tlkis only hap- rcs~~lt. if ONT/1/1 /,I{,, alit1 ON’I’,jl varied of its location 2 other for Vx(lrx V of the same X-expression specifies stands that We function formula: II). what we wol~ltl intuitively block will of the > (pz ZE qz)) constant For global AB x z (x = B A lONTABL& notation EQ1.( f,g) first of minimization. of its domain on which 7”‘). If f, g are function as r then of blocks, it in or function for each and allow the symbols EQ,,(p, 4) for VO(TX outside include predicate to specify, remain following (4), it in 2, and then all of its values predicate The of that in the ‘process ues must Assume should cumscribe like would B is on the table. What fixed are predicate the use of circumscrip- problem the will be cir- (6). objects exemplifies expresses which of (5) and John law of inertia”: (6) predicate in the conjunction monsense second B. remain (3), or function we have to either all values the val- Definitions for any predicate in the language, 2, described like to vary the s) for s = S1, and have to s = So fixed. may vary, or not include and cumscribed We would do not allow us to do that; are two axioms: TAB corresponds in this is whicli of P nlaps of u satisfying Vx for AuV(x, u), The We will denote V is identically this formula by Cp(A(P, true then Cp(A; If V is identically Z/V) Cp (A; Z/V) false then Z/V). 2); If Cp(A; 2). is equivalent to becomes set along with a circumscriptiorl policy, a metamathematical statement vary, scription CP(A). tion 6. More some values possibility: mized predicate We start schema how to perform of 2 allowed esting to vary. we may proposed policies which priorities with circumscription There vary some is another values inter- of the mini- As another (Hanks here allows example, of these ai the case when 2 is empty. The # V is a X-expression (8) expresses the values value of P A. than Cp(A) the basic signing into at different arbitrarily then change loosing its the prop- circumscript,ion V is identically it when shows to the Applying the tasks the basic forms the P of global (2) to Pa V Pb gives Using 5 x = a) v Vx(Px the form of pointwise in this section, we can higher priority to the circumscription 5 2 = b). circumscription express task the idea kind. Applying Vx(Px and To this end, introduce second a binary P at b; this formula shows language of pointwise at b is given this case, An interesting matsion 011 priorities which is iucludtd circumsrriptive P may reasoning “minimization priority. at point expresses that It is rasy a in the minimizing P to see that, in (9). feature tant but is rcprcscntc>tl example by the axiom iu the datnl~asc~ along Ihc>or,y is tISlIiLI1~ is that with tl1ollgIlt, inforV(b,n), l’u V 1’6. ask can of some (2) to A(P) (1)) kind the instants temporal apparently what capture the second ysis shows that reasoning non- of preferring of time”, disjunctive cannot of formal idea which term. would Their anal- of this kind is impor- be captured by the existing formalisms. problem Extend is clearly the theory The additional similar to the one discussed by thcsc “policy” axioms: (0 5 i < j 5 3). axioms points “later “gives prefcrcnce satisfies than about A of as an axiom method moving is a11 One expressing that, 1 leacls cxccp tion. by side on the also using to Imagine Vi’C tlic 11ow (see like two (Mc- to use the effect moving case a diffireasoning the of is “111~ law of mo- in which iL1 t.Clll[)l KNOWLEDGE resolve to tlescribc norn~nlly, can true. to formalize of the axioms a2 -f a3 mottcl circumscription calculus us that. table. used WC would to a situation tells with a “bc11,er” attempts 12). Sit~lliLtiOIi a block. axiom world If a model tllcll at the at x; in this way, it Pa2 false and Pa3 be cm1 Section of location past”. of (IO) in recent 198G), formalism minimized by making thcl blocks Carthy other tcrui uncovered P may be varied tell us that z”when to tlic I,llc> lirst This culty tion” of this S as the G (x = a1 v x = a3)]. at earlier leacl to selecting V, and b, a). The V( be varied condition circumscription a higher (8) implies that at b. This when it is minimized constant and of time, “abnormality” E (x = al v 2 = a;z)] v(ai,aj) result prcdicatc P as an and a (9) of Pa V Pb be tl le conjunction TPao. x) > Py, McDermott monotonic introduced zi x = a). = a3 A y = a& any of circumscriptions be constructed let A(P) in be gives a b ove. of assigning of minimizing will lead to the stronger posed Let A(P) 3), of a0 , . . . , a3 as instances relation, The Vx(Px 7.1. = ao)V (x = a2 Ay = a) A s(y, v Vx[Px of as- of minimizing circumscription We can think successor Hanks use effect the problem Section (0 < i < j 5 aj Vx[Px false. how we can to create or pointwise (1) of form of pointwise example points. and term It is, generally, priorities circumscription to change without of circumscription different has no pa- second by Cp(A;P/V). (8) following form The : V(X, y)}, on {y and turns The new y) which P. it is impossible We denote stronger contain true to false, at x from erty hyV(z, not that circumscrip- metamathematical axioms: G [(x = qAy ~Px does than to of circum- new S(x,y) and are allowed form us to describe 1985), V(x Here The consider and McDermott the conjunction is rameters predicates are. axioms rather by P itself. with the expressions. on Priorities Now we know describing and what a block when blocks t0 x to a 2 is at 2. pl;L’.C Rn- x is riot clear A and It 011 REPRESENTATION R side top Of / +O’-) B and then to move B second action action B is not clear. will be “normal” second will be after the first Unfortunately, the the first 1. action and the circumscrib- the possibility remains clear, result. In this of the blocks and the second alternative , and the first that unchanged, action model, is not. to a minimal leads the Each value the first 2. There more of the B action is two models AB; circumscription of than prcdicat The erence “bad” will be given ideas in the full paper. of the previous definition section, above. This wise circumscription The solution which most symbol be a 0-ary stant). We want and Sz,..., this is a all the forms definition of point- section. mathematical function say, Si. S,, correspond The pointwise where symbol, to minimize list, We hope relatively should each S; is (in particular, i.e., an object con- one of the predicate (Thus sym- 5’1 corresponds to 2 in the notation circumscription to P used be- of 5’1 in A with S; A The present cases determining Here S stauds for ,!?I,. . . , S,, icate and function icate atld predicate frrllctiott without Sl , * * * , s,, and arity Of Si. We denote variables ~~~lIlStiLlltS parameters whose (11) arity A x Y), Z 32,. s is a list 31,. V; (; which is the to the = 1,. does arity not of SI pred- . . ,72) is a contain plus the by Cs,(A; Sl/Vl,. . . ,S,/V,,). If K is identically true then we will drop /Vi in this notation. If V, is idrutically false Illen we can drop the term Si/Vi altogether. 8. point wise following atlvautnges 410 / SCIENCE to circumscription over the traditiounl globill hits the approach. proposed for formalizing of commonsense of pointwise and error, to extend principles regard- policies (such a higher priority when Fu- circumscription of general of time many reasoning. as, for into min- in temporal the policy is clearly reason- is selected unsatisfactory. the existing of circumscription methods for to more general in applications. to Michael Gelfond, Nils Nilssotr, for useful Benjamin Raymond Grosof, Reitcr and Yoav discussions. References S. and nrrn Ucthult, #/430, Yale I,ifwltil.x, 1985, McDermott, Log&, 011 Arlificial I’rcw. $111 [II- 11~telligezlce 1, 121-127. J., Circumscription lY’iLSOJlillg, Artificial M&artlly, - a form of non-monotonic hlk~fipJtce *J., A pplicntiotts (1980), 13 27-39. to formal- of ~irc.tttnscription lmowlrtlgc~, ~‘OlIllJJOltS~‘J1S~’ J3. iLtt(l brlrgll Reasoning YALEU/CSD/RR c,irc.rlttts~riotiott, C:ohfcrellcc McCarthy, ixirtg Temporal Report (I 985). (:otttl)lll,ittg Joirlt D., Tccbuical 1Tuivcrsity V., t~~~llili.iOll~ll ilrtiliichl Iu.kl!igcnce 28 89-118. McCartlty, i1pprOdl to of circumscription situation, J. and Hayes, from 1,ltv st,;m(lpoittt The point and actions. of assigning the res.llt from useful in appli- Acknowledgements (198(i), Conclusion of meta- time instants by trial needed Hanks, , . , s,, of pred- corrcspouding S; . .>I. about of circumscription at earlier Sltoharu (11) @Y(QY forms I aui grateful i=l the selection may vary powerful on applications .Joh u McCarthy, A(S) A V’ZS~[S~ZA A EQV,,(Si, Si) the form It is also important to vary on I< is, by definition, &owed that the principle forms minimized instead flexibility reasoning lead to the discovery in many including policies is sufficiently complex work stance, symbol more “modular”: of the by axioms, additional to formalizing in this paper ing). become definitions. Circumscription cations ture of for each policies, which provides imization or a function policies is defined can be described ing the choice Case it can from we need covers general pref- Details uses also the what A(S1, . . . , S,) b e a sentence, Let fore). The is given in the next General a predicate bols by “giving example. so that of circumscription introduced 7. can be eliminated as in the previous policy Circumscription 5. model to the past”, the circumscription es. 4. point, is formula. is no need to define of priorities, only gives a disjunction. circumscription of pointwise one predicate. a separate to the normal second case by a first-order 3. Circumscription action so that basic The expressed does not allow us to prove this assertion. the positions corresponds Intuitively, to the law of motion, “abnormal”. It does not eliminate “normal” else. because In ot(her words, relative ing abnormality leaves somewhere will be unsuccessful, Midtie, 1 Iliiversity I’., Some of :~rt,ilic*i;d I>. (bhls.), l’ress, M il(’ philosophical it~ttlligc~ttcc, I Jill<’ I~tliub~~rgli, IJJt(‘l/igX’JlCt’ I%!)), problems itt: Meltzer, 4 463502. (l~tiill-