SHYSTER Legal System LQ

SHYSTER: A Pragmatic Legal Expert System JAMES DAVID POPPLE A thesis submitted for the degree of Doctor of Philosophy of The Australian National University April 1993 © James Popple 1993 Doonesbury (page 131) copyright G. B. Trudeau Reprinted with permission of Universal Press Syndicate All rights reserved Earlier versions of some parts of this thesis have appeared in: Proceedings of the Thirteenth Australian Computer Science Confe © Australian Computer Science Association 1990 ence Advances in Computing and Information: Proceedings of the International Conference on Computing and Information © International Conference on Computing and Information 1990 The Australian Computer Journal © Australian Computer Society Inc. 1991 National Library of Australia Cataloguing-in-Publication entry Popple, James David, 1964~ SHYSTER: a pragmatic legal expert system. Bibliography. Includes indexes. ISBN 0 7315 1827 6. 1. SHYSTER (Computer file). 2. Law— Methodology Data processing. 3. Expert systems (Computer science). I. Australian National University. Faculty of Engineering and Information Technology. Dept of Computer Science. IL. Title. 340.11 Except where otherwise indicated, this thesis is my own original work. James Popple 29 April 1993 shy-ster \'shisto(r)\ n -s [prob. after Scheuster fl1840 Am. attorney frequently rebuked in a New York court for pettifoggery] : one who is professionally unscrupulous esp. in the practice of law or politics ... Webster’s Third New International Dictionary (1961)! shyster (‘faisto(r)) ... [Of obscure origin. It might be f. SHY a. (sense 7, disreputable) + -ster; but this sense of the adj. is app. not current in the U.S.] ... ‘A lawyer who prac in an unprofessional or tri ially, one who haunts the prisons and lower courts to prey on petty criminals; hence, any one who conducts his business in a tricky manner’ (Funk’s Stand. Dict. 1895). Also attrib. or adj. Orig. and efly U.S. slang ... manner; The Oxford English Dictionary (1989)? shyster. An unscrupulous lawyer (note that the definition presumes the existence of scrupulous ones) ... The term does not come from— supposedly a lawyer noted for shyste acter, Shylock; ... or from any of the various meanings of shy (e . shyster evolved from the underworld use of shiser, a worthless from the German sc , who cannot control his bodily functions) ... sted in various dictionaries: the surname Scheuster, from the name of the Shakespearean char- to be shy of money). Rather fellow, which derived in turn e, excrement, via scheisser, an incompetent person (specifically, one A Dictionary of Invective (1991) Abstract Most legal expert systems attempt to implement complex models of legal reas- oning. But the utility of a legal expert system lies not in the extent to which it simulates a lawyer’s approach to a legal problem, but in the quality of its predic- tions and of its arguments. A complex model of legal reasoning is not necessary: a successful legal expert system can be based upon a simplified model of legal reasoning. Some researchers have based their systems upon a jurisprudential approach to the law, yet lawyers are patently able to operate without any jurisprudential insight. A useful legal expert system should be capable of producing advice similar to that which one might get from a lawyer, so it should operate at the same pragmatic level of abstraction as does a lawyer—not at the more philosophical level of jurisprudence. A legal expert system called SHYSTER has been developed to demonstrate that a useful legal expert system can be based upon a pragmatic approach to the law. SHYSTER has a simple representation structure which simplifies the problem of knowledge acquisition. Yet this structure is complex enough for SHYSTER to produce useful advice. SHYSTER is a case-based legal expert system (although it has been designed so that it can be linked with a rule-based system to form a hybrid legal expert system). Its advice is based upon an examination of, and an argument about, the similarities and differences between cases. SHYSTER attempts to model the w in which lawyers argue with cases, but it does not attempt to model the way in which lawyers decide which cases to use in those arguments. Instead, it employs statistical techniques to quantify the similarity between cases. It decides which cases to use in argument, and what prediction it will make, on the basis of that similarity measure. ay vii viii Abstract SHYSTER is of a general design: it can provide advice in areas of case law that have been specified by a legal expert using a specification language. Hence, it can operate in different legal domains. Four different, and disparate, areas of law have been specified for SHYSTER, and its operation has been tested in each of those domaii Testing of SHYSTER in these four domains indicates that it is exception- ally good at predicting results, and fairly good at choosing cases with which to construct its arguments. SHYSTER demonstrates the viability of a pragmatic approach to legal expert system design. Acknowledgements Thanks ... TO the members of my supervisory panel—Robin Stanton, Roger Clarke, Peter Drahos and Malcolm Newey: for their counsel and encouragement. TO Robin Creyke and Phillipa Weeks: for giving so freely of their time and legal expertise. TO Peter Bailey, David Cullen, Kate Lazenby-Cohen and Kevin Popple: for their comments on various parts of various drafts. TO Graham Jefferson, Chris Johnson, Brendan McKay, Gavin Michael, Bob Moles, Neil Tennant and Trevor Vickers: for their suggestions and advice. TO Sally Begg, Jeannie Haxell, Bev Johnstone and Jacquie Wilson: for adminis- trative assistance. TO David Hawking and the programming staff: for technical assistance. TO Richard Walker and Markus Zellner: for TeXnical assistance. TO Zdzislaw Loboz, Rachel Onate, Dharmendra Sharma, Wang Xiaoji and Yang Jian: for sharing, with me, different offices at different times. TO Diane Hutchens: for all her help during my time with PARSA. TO my parents—Marli and Kevin: for their unfailing love and support. AND TO Paula Fearn: for whom, but for whom ... search was supported by an Australian National University PhD Scholarship, funded by the Centre for Information Science Research. Additional financial support was provided by Sigma Data Corporation. ix Abstract Contents Acknowledgements Figures Chapter 1: Introduction 1.1 The aims of this thesis 1.2 SHYSTEReciescs eee os eee meinem cece emia os ee pra sate 1.3 ‘The structure of this thesis». c...1 yeyy ca ese ya dye wees ee aee Oeiees ¥ Chapter 2: Legal analysis systems 2.1. Introduction 2.1.1 2.1.2 2.1.3 2.1.4 2.2 Jurisprudence ............ 2.2.1 2.2.2 2.2.3 2.2.4 2.2.5 2.2.6 2.2.7 Sources of law .. The doctrine of precedent The structure of this chapter Terminology The importance of jurisprudence ........ Scientific and mechanical jurisprudence . . Judgment machines ................. Petrifaction of the law Clear rules and clear cases Legal realism and rule scepticism A/JUPISprudential CONSENSUS? exer ore: ere oxen maces & 2.3. Jurimetrics and the behaviourists 2.3.1 2.3.2 Kort xi vii ix xxi xii Contents 23:3 Nagel.and Schubert «2: woaes o4..cage Me cta. ds aeeta Dake Aca 25 2.3.4 Haar, Sawyer and Cummings .............6. 6.00 e cece eee eee 25 23.5 PYEICHON secu memasan eae ma ean mi marie aR eee 25 2.4. Rule-based systems: vs si g3005.25 (a wds $4 We wh Maes ds Ode Ra Ha ka ee ale ales 27 2.4.1 McCarty 29 2.4.2 Bench-Capon, Kow: : ‘ ‘ “ .. 32 2.4.3 Gardner 33 2.4.4 Susskind 35 2.4.5 Knowledge acqui 36 2.4.6 Fact represétitationiwisses veces ox eae: secre eee extern owns 37 2.4.7 The inadequacy of rules for case law 38 2.5 Case-based systems 2.0... 66.6 cece eens 39 2.5.1 Nearest neighbour analysis .............. 0060s c cece cence eens 41 .5.2 FINDER . Al 2.5.3 HYPO... 42 2.5.4 The inadequacy of semantic networks...............0.0.0.0055 44 2.6 Hybrid systems: o< cnes: seein seein ox ees oose eae eee owes 44 2.7 Conceptual models: deep and shallow 46 2.8 Conclusion ... 48 Chapter 3: A pragmatic approach to case law 51 Bek, UNPODUCHOM ears sceserarecaie eacmenenets erence scale omteseenes caren aesrecaense coarser 52 3.2 Design criteria 52 3.2.1 Users and output ... 52 3.2.2. A pragmatic model of legal reasoning.....................0005 53 3.2.3 Knowledge representation ............0 00.006 c eee e cece 53 3.2.4 Generality of application 54 3.2.5 Prediction and argument. . 54 3.2.6 Evaluation of advice...... 54 S:2IC A Hybrid Structure ncncesse caerace sceensensin caxoenerens erence srarmecneas epeesee 55 3.3. A model of legal reasoning 55 3.3.1 Private and public law. 55 3.3.2 The functions of legal reasoning... ...........0 60.0 e cece eee 56 3.3.3 Adopting a model of legal reasoning : : : R -. 56 3.4 Experts, users, and expert users .......0.6 60.6 c ccc c eee e cece eee eens 58 3.5 Knowledge representation .......... 60.6. c cece eee ee 58 3.5.1 Areas.... 59 3.5.2 Attributes . 5 5 F ; a -. 60 3.5.3 Leading cases and attribute values ...............0.00000 0000s 60 3.54. Ideal POitits: cic ccmnsas canuu ma ean armen marin caer eae 61 3.6 A specification language ...... 26.6.6 cece eens 61 Contents 3:7 ‘“Weighting:attributes; 23 sco sa.0%;.22 oeems Sakae NEW a HRY OE AG ae Y 3.8 Detecting attribute dependence 3.8.1 | Functional dependence 3.8.2 Stochastic dependence 3.9 Calculating distances ...............0.0.0055 3.9.1 Similarity measures Distance measures Association coefficier Correlation coefficients . 3.9.2 Weighted similarity me 3.9.3 Choosing a similarity measur 3.9.4 Infinite distance 3.10 Nearest cases and nearest results... .............0.00005 S101 Nearest Cases oncccncaniis ayonoonenets eavenca sexes pene os ge 3:102 Nearest results: 2wes% ua.acc ge 4280s dh HORS Sa kegs PRE ee Bee 3.10:3 -Equidistanee? a. m.cn imines carne sarrcacen seamen aa eave eines © 3.11 Writing a report 3.11.1 Arguing with the instant case ........ sures 3.11.2. Arguing with instantiations 3.11.3 Arguing with hypotheticals........... 3.12 Safeguards ........ 00... e cece eee eee 3.12.1 é 3.12.2 3.12.3. Centroids .... : 3.12.4 Attribute direction . 3.12.5 Equidistance 3.12.6 Instantiations 3.13 Testing and evaluation 3.13.1 Choosing a test domain. . 3.13.2. Evaluating SHYSTER’s opinion 3.13.3 The paucity of test cases ...... 60.0.6. e cee eee cece zu BABA ‘Gsierated 1E8t8 cca ssawesa eases va sane ex sama canna eoaree 3.13.5 Reflexive tests ©... 00.06. cnn 3.14 Conclusion 3.14.1 Comparisons with other approaches .. . ae 3.14.2 A Sisyphean journey? «2.2.2.0... 066 c ccc cece eee Chapter 4: Implementing SHYSTER 41 Antrodctidtises: mews ex ews exwen eres vane wees oo Hee eaten 4.1.1 Internal representation 41.2 ‘Qutpiit files: s1.e8 exes ceees secs eee eres 42 ‘The SHysTeR module s 2.08. eee wows eye eee eR eee EYEE Y 43 The Starures module's :26 0% s3.0%;.02 omens S200 oe Meee as HORSE OE Ska ae YL xiii 75 85 xiv Contents 44, The Gases module: ove sn uyac5.ns ees 2 keh Sew. ds UE DE kG ee Bees 107 4.5 The ToKENIzER module . . 108 4.5.1 Tokens 108 4.5.2 | Comments and whitespace . . 108 4.6 The ParsER module ... 109 4.6.1 Hierarchy 109 4.6.2 Areas.... 111 4.6.3 Attributes 112 Local attributes ...... 112 External attributes . 112 4.6.4 Leading cases............ 113 4.6.5 Ideal points ... 114 4.7 The Dumper module .. 114 4.7.1 Attributes 116 Local attributes 116 External attributes ... 116 4.7.2 Leading cas 117 4.7.3. Ideal points ... 117 4.8 The CHECKER module 117 4.8.1 Detecting dependence . 118 4.8.2 Writing matrices of probabilities .....................0.0.0055 119 4.9 The ScaLes module 120 4.9.1 Calculating weights . 120 4.9.2 Writing tables of weights 120 4.10 The ApjusTER module . 120 4.11 The Consutrant module. . 120 4.12 The ODOMETER MOdUIE:. csccnias sacs oa seams eee earees sareK caren 121 4.12.1 Calculating distance : 122 4.12.2 Writing tables of distance: 122 413 The-ReporTER module «4 20:2 coeds 65. We oe Saas ds OS Oa DE oda ee ale wes 124 4.13.1 Arguing with the instant case 124 Arguing for the nearest result . . 124 Arguing for other results . 126 Variations ......... 127 4.13.2. Arguing with instantiations . 127 4.13.3 Arguing with hypotheticals. .. 127 4.13.4 Safeguards 128 4.14 Conclusion ... 130 Chapter 5: Case studies 131 5.1 Introduction. ... 20.6... nce ene ee 132 5.2) ‘Asimulation.of BINDER cscs wera caeucein omteceains eareines aiareuacerass eaters 132 S21 The-law: s ice seavn.ne weeNE Sh kege Neate yee Dem ge Aree 132 5.3 5.4 Contents 5.2.2 The FINDER specification ......... 66.6.0 c cece cece eens R@SUtB ic: war cence samen wae wexrer Attributes Leading Cases vac: eecman veer oy a Attribute dependence Weightss: s:.iceou waves gee Base 5.2.3. Test case Parker v. British Airways 5.2.4 Conclusion The authorization of copyright infringement .............0..0..0.20008 5.3.1 The law 5.3.2 The AUTHORIZATION specification Resultse s:esusa eeage ues Pee Attributes Leading cases .............0.0004 Attribute dependence WIRE), mcasoce aomenne econ ane 5:3:3, “Test cases aus s3 ices wa,ava. ae yee oa CBS Inc. v. Ames Records and Tapes Ltd WEA International Inc. v. Hanimex Corporation Ltd ... CBS Songs Ltd v. Amstrad Consumer Electronics ple ... Australasian Performing Right Association Ltd v. Jain .. Hypothetical case 1 Hypothetical case 2 we 5.3.4 Generated tests... 20.6... Generated 1. Generated test 2 . Generated test 3. 5.3.5 Conclusion Employees and independent contractors....... 5.4.1 The law 5.4.2 The EMPLOYEE specification RESUS ies ox see: ancora eae eet Attributes Leading cases Attribute dependence Weights ..............0..00.5000- §:4:3' “Test Ca8@8sc2 ox eieens cece pew ey aati amen exe Narich Pty Ltd v. Commissioner of Pay-roll Tax Stevens v. Brodribb Sawmilling Company Pty Ltd Re Porter; Re Transport Workers Union of Australia . ae Building Workers’ Industrial Union of Australia v. Odco Pty Ltd. Hypothetical case 3 Hypothetical case 4 xv Contents xvi 5.44 Generated tests. 200.44 s6s)6s 08 dees ok Wa we Meee de bee oa we 166 Generated test 4 .. Generated test 5 .. Generated test 6 .. 5.4.5 Conclusion 5.5 The implication of natural justice 5.5.1 5.5.2 The NaTuRAL specification . . Natural area .. . Affected area . Expectation area. . Leading cases ...... Attribute dependence Weights ... 5.5.3 Test cases..... Twist v. Council of Municipality of Randwici Salemi v. MacKellar . . Heatley v. Tasmanian Racing and Gaming Commission ..... 183 Cole v. Cunningham Ackroyd v. Whitehouse (Director of National Parks & Wildlife Service) .. 184 Hodgens v. Gunn 185 Western Australia v. Bropho ..... 0.0.0.6 0 cece cece eee 186 Ainsworth v. Criminal Justice Commission ...............5 187 5.5.4 Generated tests....... 188 Generated test 7 188 Generated test 8 .. 188 5.5.5 Conclusion . . 189 5.6 Conclusion ... 189 5.6.1 Test 190 5.6.2 Generated t 192 5.6.3 Reflexive tests 193 Chapter 6: Conclusion 195 6.1 Introduction. ... 20.66. eee 6.2 Evaluating SHYSTER 6.2.1 A useful, working system. . 6.2.2 A general approach 6.2.3 Good advice... 6.2.4 Limitations ... 6.2.5 Possible enhancements... .. 0.0.0.6 0 66.0 c cece ccc ee eee 200 6.3. Future research 6.4 The contribution of this thesis ..... 2.0.0.0 6 66.0 c cece cence cece 202 Contents Appendix A: Case law specifications A.1l Introduction......... A.2.— The FINDER specification a Hierarchy: son gages MR Ree RE RE Ee ER Finder area RéSUlts! cana au means axconivns navman sarne Attributes Cases in which the finder wins ........ Cases in which the finder loses ...... 0. 6.6.. 06 sees een eee eee A.3 ‘The: AUTHORIZATION’SPpeCifiCabion 01. conn caeceae mamrenier wants earcearee « Hierarchy Authorization area Opening RéSUtS) ccssscs wr nacee oscar seer exer we ATEPIDUGES! 6 cu means axmoniam marmin cu meats saeete earns SHOT 8 Cases in which the acc Cases in which the accused did not authorize the infringement ... Cases in which the accused is liable (directly or vicariously) for CHOUBTINGEMENE cone eres acces ac EE ene 6 sed authorized the infringement ......... A@ ‘The BMpLover: specification: 92,005.05 Grats S8.08G8e RAs RD BE. ge 8 Hierarchy Employee area Opening Results 2.0.0.0... 0 0c e cece eee eee Attribute: Cases in which the worker is an employee Cases in which the worker is an independent contractor ... A.5 The Natura specification .. 0.000.000 0c Hierarchiy: iit Shae NEAL WET BLIGE NEEL SE BOE BERGE 8 Natural area Opening ReSUlt) caccccss ou micas axcrninns marmene saree Attribute: ee in which a duty to observe natural justice is implied ...... ses in which a duty to observe natural justice is not implied... AfECIEURAIED cans wr aemas meee cae meee sae eee ene & ReSUlt) caccccss ou micas axcrninns marmene saree Attributes Cases in which the decision affected the property, right, interest, status, or legitimate expectation of the applicant we Cases in which the decision did not affect the property, right, inte status, or legitimate expectation of the applicant xvii 205 - 206 . 206 206 -. 206 -. 206 .. 207 . 209 211 216 .. 216 .. 216 . 216 x 297% 217 219 222 xviii Contents Expectation area: 07,505.06 Brae She MEL ee MR SEE BS See, Opening ... Results .. Attributes acy saai.ne yea SE ews NEA ds AeA Sake ge AeA: Cases in which the applicant had a legitimate expectation which was affected by the decision .................... 276 Cases in which the applicant did not have a legitimate expectation which was affected by the decision.......... 280 Appendix B: Example reports 283 B.l Introduction. ........ 000.2 cece eee eee 284 B.2_ Report file for Parker v. British Airway: 284 Finder area 284 Instant case ..... 284 Hypothetical 1... 286 Hypothetical 2.......0.0 0.00000 ce ccc cece cee ce cess veces sees 288 B.3 Report file for APRA v. Jain .. 289 Authorization area 289 Instant case 289 Hypothetical 1... 292 Hypothetical 2... 293 Hypothetical 3. 294 B.4 Report file for Re Porter; Re TWU.. 296 Employee area 296 Instant case . 296 Instantiation 4... 298 Hypothetical 1... 299 Hypothetical 2... 301 B.5 — Report files for Ainsworth v. CJC... 302 Expectation area . 302 Instant case ... 302 Hypothetical 1... 304 Affected area 304 Instant case 304 Hypothetical 1... 306 Natural area 307 Instant case ... 307 Hypothetical Lo... 00.0.0 0000 0c cece cece eee cee cece eee ee ees 309 Appendix C: A complete example 311 C.1 Introduction. ... 20.66. ccc ene 312 G2 Log file for BWIUG. 0460 saci secs ox nanan oeamiwn mere waren sane 312 C.3 EMPLOYEE specification file .... 2.2.0.0... c ccc cece eee nn eens 316 Contents xix C.4 Dump file for EMPLoyEE specification (IATpX input) ............0.0..05 328 C.5 Dump file for EMPLOYEE specification (IATgX output) [see §A.4] ......... 331 C.6 Probabilities file for EMPLoYEE specification (IATgX input) .............. 332 C.7 Probabilities file for EMPLOYEE specification (IATgX output) . . 333 Employee area 333 C.8 Weights file for EMPLOYEE specification (MTgX input) ................05 334 C.9 Weights file for EMPLOYEE specification (IATgX output) . . 335 Employee area a C.10 Distances file for BWIU v. Odco (IXTgX input) ...... 2.06.0 e eee eee eee C.11 Distances file for BWIU v. Odco (I&TgX output) . . Employee area Instant case Instantiation 1 Instantiation 2 Instantiation 3 Instantiation 4 Hypothetical 1 Hypothetical 2 see C.12 Report file for BWIU v. Odco (IXTGX input) ...... 06.0.6 eee eee eee C.13. Report file for BWIU v. Odco (IXTgX output) .. Employee area Instant case Hypothetical 1 we Hypothetical 2.0.2.0. 00 0.00 Appendix D: Reflexive tests 353 D.1 Introduction . 354 D.2.— The FINDER specification 354 Hannah v. Peel ... 354 Bridges v. Hawkesworth r P . 354 Armory v. Delamirie... . 355 Moffatt v. Kazana “ . 355 City of London Corporation v. Appleyard (1).. . 355 City of London Corporation v. Appleyard (2).. . 356 South Staffordshire Water Co. v. Sharman.... . 356 Elwes v. Brigg Gas Co 356 D.3. The AuTHORIZATION specification . . 357 University of New South Wales v. Moorhouse 357 Australasian Performing Right Association Ltd v. Canterbury- Bankstown League Club Ltd .. 20.0.6 6 6.0 c cece cece nce 357 Winstone v. Wurlitzer Automatic Phonograph Co. of Australia PEG LM cvesorass scxcarnnien spatmsesanate wecntans sarseacnees aiokoonenene eareee searmecmnaes ¢ 358 Mellor v. Australian Broadcasting Commiss 358 KE Contents Falcon v. Famous Players Film Co. 0.0.00. 060 cc cnn RCA Corporation v. John Fairfax and Sons Ltd .........0.6060 000% Performing Right Society Ltd v. Ciryl Theatrical Syndicate Ltd . 4 AGM Records Inc. v. Audio Magnetics Inc. (UK) Lid .............. Australasian Performing Right Association Ltd v. Miles ............ 360 D.4 The EmPLoyee specification ..... 0.0.0.0. 6 cece cece cece 361 Zuijs v. Wirth Brothers Pty Ltd ...... 06.006 c ccc cence 361 Cam and Sons Pty Ltd v. Sargent 0.0.6.6. 6066s 361 Federal Commissioner of Taxation v. J. Walter Thompson (Australia) Pty Ltd... 0.6.0 cence 362 Australian Timber Workers Union v. Monaro Sawmills Pty Ltd 362 Ferguson v. John Dawson & Partners (Contractors) Ltd ...... . 362 Stevenson Jordon and Harrison Ltd v. Macdonald and Evans (2) .... 363 Performing Right Society Ltd v. Mitchell & Booker (Palais de Danse) Ltd Humberstone v. Northern Timber Mill: “ Queensland Stations Pty Ltd v. Federal Commissioner of Tazation . 364 Price v: Grant Industries Pty Ltd ccx.cscwrscs aiaxesenerers wareinca soainecaises epeteses Australian Mutual Provident Society v. Chaplin .........0.0.000055 Massey v. Crown Life Insurance Co. ......0.6 0606000 Stevenson Jordon and Harrison Ltd v. Macdonald and Evans (1) .... 365 Ready Mixed Concrete (South East) Ltd v. Minister of Pensions and National Insurance ... 2.26060. 366 D.5 The Narurat specification ©0202... 00.0. c ccc cece nee ee 366 FAI Insurances Ltd v. Winneke «0.0.6.0 0 00 ccc ccna 366 Haoucher v. Minister of State for Immigration and Ethnic Affairs ... 366 ANNEtE 0. MéCOihs casei meena wr nace srewaen mac ox wes 367 Kid v. West .. 6.6 cent nee 367 Commissioner of Police v. Tano. 367 Marine Hull © Liability Insurance Co. Ltd v. Hurford 367 Macrae v. Attorney-General for New South Wales . . 368 Attorney-General of Hong Kong v. Ng Yuen Shiu..............0055 368 Durayappah v. Fernando South Australia v. O’Shea on Bread Manufacturers of New South Wales v. Evans Minister for Arts Heritage and Environment v. Pei . 369 Nashua Australia Pty Ltd v. Channon... 2.0.6.6 0 ccc ccc cee 370 Council of Civil Service Unions v. Minister for the Civil Service..... 370 McInnes v. Onslow Fane: .s. ccccs 05 vas ha dds Wont a a be ae 370 Notes 375 Cases 403 Statutes 409 Bibliography 411 * * * + + FF HH FOF * Figures Figures marked with a star are extracted from SHYSTER output. 3.1 Forms of functional dependence ..... 2.0.0.6. 000 cece cee e eee eee 67 3.2 Example attribute values ........ 68 3.3 Probabilities for example attributes 69 3.4 Two-way association table 73 3.5 Example distances ......... z 78 3.6 SHYSTER’s algorithm for choosing cases ...... 81 3.7 SHYSTER’s algorithm for using c 82 3.8 SHYSTER’s algorithm for using cases: 83 3.9 SHYSTER’s algorithm for using cas 84 4.1 SHYSTER’s command line switches ........... 20.0.0. e cece cee eee ee 106 4.2 Keywords in SHYSTER’s case law specification language . 108 4.3 EBNF description of SHYSTER’s case law specification language . 110 4.4 — Attribute values for Employee area 115 4.5 Extract from probabilities for Employee area... . 119 4.6 Similarity measures 121 4.7 Distances for BWIU v. Odco we. 123 4.8 Extract from log file for BWIU v. Odeo... 2.6.6. 129 5.1 Probabilities for Finder area 137 5.2 Weights for Finder area 138 5.3 Distances for Parker v. British Airways 141 5.4 Probabilities for Authorization area................-. 146 5.5 Weights for Authorization area 147 5.6 Distances for CBS v. Ames 150 5.7 Distances for Narich v. CPT 162 5.8 Relationship between areas in NATURAL specification . .. 173 5.9 Probabilities for NATURAL specification ..........6 660.0 c cece cence 177 xxi xxii 5.10 * 5.11 5.12 5.13 D1 D.2 D3 DA Weights for NATURAL specification Distances Summary Summary Summary Summary Summary Summary Figures for Twist v. Randwick, Natural area . of test c of generated t ng. . of reflexive testing of FINDER specification of reflexive testing of AUTHORIZATION specification of reflexive testing of EMPLOYEE specification . of reflexive testing of NATURAL specification . . 178 181 191 193 356 360 365 371 Chapters Introduction Dicke the Butcher: “The first thing we do, let’s kill all the Lawyers.” William Shakespeare (1591) Henry VI, Part 24 Attending a Cabinct when there was a tendency to Mutiny in the Fleet, Sir Thomas Troubridge, ... who wi ked his opinion what was best to be done. He said let me hang a hundred Lawyers, and we shall hear no more of the business. I asked what: he could mean—what were these People that he called Lawyers. He replied, Fellows that can read and write. They are the Fellows, that I call Lawyers, and make the whole of the Mischief. a most excellent Officer, was as Lord Eldon (1827) Lord Eldon’s Anecdote Book? To his colleagues, [he] was seemed a bold crusader, in between. He was in the early stages of a PhD thesis. a lonely and solitary nut. To an objective outsider he might have uging war against the forces of darkness. The truth was somewhere Alan Plater (1985) The Beiderbecke Affair® 4 Chapter 1: Introduction 1.1 The aims of this thesis The history of the development of legal expert systems has, for the most part, been characterized by the development and implementation of complex models of legal reasoning. This thesis aims to show that a legal expert system need not be based upon a complex model of legal reasoning in order to produce useful advice. It advocates a pragmatic approach to legal expert system design based on the way in which lawyers deal with the law on a day-to-day basis. It argues that a system based upon a simple model of legal reasoning can still produce good advice, where that advice is evaluated by reference to the accuracy of its predictions and to the quality of its arguments. Furthermore, such a sys- tem, with its simpler knowledge representation structure, makes commensurately simpler the process of knowledge acquisition. These arguments are made theoretically, and then by example. A legal expert system which is based on a pragmatic approach to the law has been developed by the author. The development and testing of that system, called SHYSTER, is described in this thesis. 1.2 SHYSTER SHYSTER was developed to demonstrate that a useful, working legal expert sys- tem could be based upon a pragmatic approach to the law. SHYSTER is of a general design, so that it can operate in different legal domains. It was designed to provide advice in areas of case law that have been specified by a legal expert using a specially developed specification language. SHYSTER is a case-based legal expert system. Its knowledge of the law is acquired, and represented, as information about cases. It produces its advice by examining, and arguing about, the similarities and differences between cases. By contrast, a rule-based expert system represents the law using rules. A hybrid system uses both rule-based and case-based techniques. SHYSTER has been de- signed so that it can be linked with a rule-based system to form a hybrid legal expert system. Although SHYSTER attempts to model the way in which lawyers argue with cases, it does not attempt to model the way in which lawyers decide which cases to use in those arguments. It uses statistical techniques to quantify the similarity between cases, and chooses cases on the basis of that similarity measure. SHYSTER’s representation structure was designed so as to be as simple as possible while complex enough to allow SHYSTER to produce good advice. This simple structure greatly simplifies the process of knowledge acquisition. $1.3 The structure of this thesis 5 1.3. The structure of this thesis The body of this thesis is divided into six chapters: e Chapter 2 discusses previous work of relevance to the development of legal analysis systems, especially legal expert systems. The value of jurisprudence to legal expert system development is also discussed, and the adoption of a pragmatic approach (as opposed to a jurisprudentially pure approach) is recommended. This approach holds that legal expert systems should operate at the same level of abstraction at which lawyers operate on a day-to-day basis. e In chapter 3, a pragmatic approach is proposed for developing legal expert systems. It incorporates a simple model of legal reasoning, and uses a simple knowledge representation structure. Comparisons are made between this approach, and approaches adopted by other legal expert system developers. This approach is adopted for the development of SHYSTER. Specific design criteria for SHYSTER are detailed, and methods of testing and evaluating the system are discussed. e The implementation of SHYSTER is explained in chapter 4. The twelve modules that comprise SHYSTER are described, and demonstrated using examples. e Four different specifications have been written for SHYSTER, and these are used as the basis of case studies in chapter 5. Each specification represents a different area of case law. Several different methods are employed to test SHYSTER and these specifications. e In chapter 6, conclusions are drawn about SHYSTER and its approach to case law. Some enhancements to SHYSTER are suggested, and avenues of future research are identified. Finally, the nature of the contribution made by this thesis is discussed. There are four appendices to this thesis. Appendix A contains each of the four specifications used to test SHYSTER in chapter 5. Six example reports, demonstrating the use of each of these four specifications, are given in appendix B. Each report is SHYSTER’s opinion on one of the test cases used in chapter 5. A complete example of SHYSTER’s input and output files for another of those test cases is given in appendix C. Appendix D gives details of one of the methods of testing, as applied to each of the four specifications. The thesis concludes with the endnotes to the chapters and appendices, a table of cases, a table of statutes, and a list of bibliographical references. Legal analysis systems ead the It > far In the yahoo world of computer public relations, it is often the loudest mouths which l unsuspecting computer user into that lonely canyon of empty pockets and broken prom: now seems ... that parts of the academic world are fast approaching that decibel level achieved only by computer salesmen. Thus ... it seems that space can only be booked on the band-wagon if one is prepared to make more outrageous claims than the next man. And such claims are being made for the application of AI to the law. Philip Leith (1986)? The computer scientists, encouraged by the modern positivists, fail to recognize ... that law, positive morality and ethics are inseparably connected parts of a vast organic whole. Judgments are involved at every stage of the legal process and machines cannot make judgments. In stating that legal rules can be applied without further judgment; that they apply in an all or nothing fashion; that legal decision making follows the form of the syllogism or that it is a pattern- matching routine, the modern positivists, joined now by the computer scientists take us along a dangerous road. Robert N. Moles (1987) Definition and Rule in Legal Theory: sment of H.L. A. Hart and the Positivist Tradition® Aussi, lorsqu’un homme se absolu, songe-t-il d’abord les lois. On commence, dar Thus when a man takes on absolute power, he first thinks of simplifying the law. In suc ate one begins to be more affected by technicalities than by A étre plus frappé des inconvénients the freedom of the people, about which one no longer particuliers, que de la liberté des sujets at all. dont on ne se soucie point du tout. Montesquieu (1748) De VEsprit des lois” a cares 8 Chapter 2: Legal analysis systems 2.1 Introduction The range of computer applications in the law is wide. It extends from general applications, of use to lawyers, to applications designed specifically for the law. This thesis is concerned only with a subset of those systems that make use of artificial intelligence (AI) techniques to solve legal problems. Legal AI systems can usefully be divided into two categories: legal retrieval systems and legal analysis systems.'° Legal retrieval systems allow lawyers to search through databases containing details of statutes and decided cases. AI techniques may be employed to simplify this task: e.g. by searching for keywords which have not been input by the user but are deduced to be equivalent to, or sufficiently related to, the input keywords.!! Legal analysis systems take inform- ation about a set of facts and determine the ramifications of those facts in a given area of law. Mehl claims that there is no fundamental difference between these two cat- egories—that the difference is one of degree only.!? Shannon and Golshani suggest that the difference between systems based on a “conceptual model of legal ana- lysis” and text retrieval systems is that the latter do not “understand” any area of the law. Similarly, Susskind says that “knowledge-based” systems, as opposed to “database” systems are capable of “applying their knowledge of the law to the problem data presented to them.” !4 A better distinction—one which avoids the vexed question of whether any AI system can be said to really know or understand anything—can be made by reference to the output of the system. The output from a legal analysis system is such that, if it had been produced by a human, that human would be said to have legal expertise. By contrast, the output from a legal retrieval system could be produced by a human possessed of no legal expertise; such output is used in, and is not the product of, legal analysis.!° This thesis is concerned only with legal analysis systems. * * * Legal analysis systems can be divided into two categories: judgment machines and legal expert s A judgment machine is a machine designed to replace a human judge. Such machines were first proposed over forty years ago, though no such proposals have been made in the last decade. Writings on judgment machines are discussed in §2.2.3 because they are of historical interest, and because the idea of a judgment machine rais sues which are also relevant to the second category of legal analysis systems: legal expert systems. A legal expert system, as the term is used in this thesis, is a system cap- able of performing at a level expected of a lawyer. AI systems which merely assist a lawyer in coming to legal conclusions or preparing legal argument are not here considered to be legal expert systems; a legal expert system must exhibit some §2.1 Introduction 9 some legal expertise itself. This definition does not exclude systems that can only be used by legal experts; several systems—including SHYSTER—have been developed for exclusive use by lawyers.!® 2.1.1 Sources of law The law in so-called “common law countries”—e.g. Australia, Britain, Canada, USA—is derived from legislation and from case law. Legislation, also referred to as statute law, consists of statutes and delegated legislation. In Australia, statutes are made by Federal Parliament, or a State Parliament, and enacted when given royal assent by the Governor-General, or a State Governor. Delegated legislation (rules, regulations, ordinances, by-laws, etc.) is made by a person or body to whom legislative power has been delegated by statute. Case law, or the common law, is judge-made law: judicial resolutions of spe- cific disputes. The sources of case law are the published reports of the cases as heard before various courts. Legislation takes precedence over case law;!” parliaments can override judge- made law by legislative enactment. However, judges have the task of determining the meaning of legislation. A legal analysis system that seeks to deal with the law in a common law country must account for both statute law and case law. 2.1.2 The doctrine of precedent The development of case law is based upon the principle of stare decisis which holds that courts should apply the doctrine of precedent. Morris et al. summarize the general rules of the doctrine of precedent as follows: e each court is bound by decisions of courts higher in its hierarchy; e a decision of a court in a different hierarchy may be of considerable weight, but will not be binding; e only the ratio decidendi of a cas is binding; e any relevant decisions, although not binding, may be considered and fol- lowed; and 8 precedents are not necessarily abrogated by lapse of time.! (The ratio decidendi of a case is, literally, the reason for deciding; rationes are “pronouncements of legal principle necessary for the judge’s decision on the es- tablished facts of the case”.!® They are different from pronouncements of legal principle which may be illustrative or clarifying, but are not strictly relevant to the case in issue. These are called obiter dicta, and are not binding on other courts.) 10 Chapter 2: Legal analysis systems Courts have a tendency to follow similar cases, even when not strictly bound to do so. According to Cross: It is a basic principle of the administration of justice that like cases should be decided alike. This is enough to account for the fact that, in almost every jurisdiction, a judge tends to decide a case in the same way as that in which a similar case has been decided by another judge.?° There is considerable argument about the extent to which judges actually apply the doctrine of precedent.?! Some legal theorists contend that the doctrine is simply part of the public discourse that judges use to justify their decisions. Stone claims that “the degree of certainty and stability in the law secured by the doctrine of stare decisis is far less than it appears to be,” and that one of the important social functions of the doctrine is “to maintain at a maximum the feeling and appearance of certainty and stability.”?? Yet, judges of the highest court in Australia have been known to follow previous cases which they believed to be wrongly decided.”3 What is clear is that judges use previously decided cases to justify their decisions—if not to reach them—and that lawyers use previously decided cases in legal argument. Any legal analysis system which deals with case law assumes the application of the doctrine of precedent, at least to this extent. 2.1.3 The structure of this chapter This chapter reviews previous work of relevance to the design of legal expert systems. Jurisprudence and its value to the development of such systems is discussed (§2.2) and the field of “jurimetrics” and some behaviouristic research is examined (§2.3). The bulk of this chapter concerns the development of legal expert systems. This is discussed under three headings: rule-based, case-based, and hybrid systems. Rule-based systems are examined first (§2.4), and special attention is paid to the work of four projects (§2.4.1-§2.4.4). The problems of knowledge acquisition and representation, and fact representation are also discussed (§2.4.5 and §2.4.6). Rule-based systems have been used to represent statutes and cases. However, as explained in §2.4.7, rule-based systems are fundamentally inadequate for repres- enting case law. Case-based systems are examined (§2.5) and the work of three projects is focused upon (§2.5.1-§2.5.3). Systems that use rules in order to represent case law are considered to be rule-based systems; only systems which adopt case-based reasoning methods are considered to be case-based. Applying this distinction is usually straightforward, though not always.?4 Attempts have been made to use semantic networks to represent case law but, as explained in §2.5.4, they are not suited to the task. Hybrid systems (§2.6) employ both rule-based and case-based methods. §2.2. Jurisprudence aia The search for conceptual models of legal reasoning is discussed (§2.7). Finally, conclusions are drawn from the literature as to the best approach to the design of legal expert systems (§2.8). (The development of legal expert systems raises legal questions itself. For example, who—if anyone—is liable for “bad” advice provided by such a sys- tem? These questions are beyond the scope of this thesis, but are examined elsewhere.”°) 2.1.4 Terminology For consistency, the following three terms are used wherever possible in this thesis for a variety of terms that are used throughout the literature. An attribute is a legally important fact; some use the words “descriptor,” “dimension,” “factor,” “feature” or “variable.” The instant case is the situation about which the expert system is interrogated: this is sometimes called the “actual case,” the “current fact situation,” the “hypothetical case” or the ” Leading cases make up the case base of a legal expert system; some call these the “reference set” or the “training set.” “new cas 2.2 Jurisprudence For centuries, lawyers and philosophers have written on, and argued about, the nature of human laws. This field is called jurisprudence. It would seem sensible for the developer of a legal expert system to have regard to jurisprudential the- ory before designing her/his system. However, as Susskind complained in 1987, most of the published research on various legal expert systems makes no use of jurisprudential resources.” 2.2.1 The importance of jurisprudence Some expert system researchers have been doubtful as to the value of jurispru- dence; Niblett claims that “a successful expert system is likely to contribute more to jurisprudence than the other way round”.”’ Susskind disagrees: It is beyond argument ... that all expert systems must conform to some jurisprudential theory because all expert systems in law necessarily make assumptions about the nature of law and legal reasoning. To be more spe- cific, all expert systems must embody theories of legal knowledge, legal science, the structure of rules, the individuation of laws, legal systems and sub-systems, legal reasoning, and of logic and the law (as well perhaps as elements of a semantic theory, a sociology, and a psychology of law), theor- ies that must all themselves rest on more basic philosophical foundations. If this is expert systems should be explicitly articulated using (where appropriate) , it would seem prudent that the general theory of law implicit in 12 Chapter 2: Legal analysis systems the relevant works of seasoned theoreticians of law. Perhaps the reason that there is, as yet, no overwhelmingly successful system is that the vast corpus of apposite jurisprudential material has not yet been tapped in the construction process a Although it is true that all legal expert systems necessarily make assump- tions about the nature of law and legal reasoning, it does not follow that they must conform to some jurisprudential theory. A lawyer must have a model of the law (maybe unarticulated) which includes assumptions about the nature of law and legal reasoning, but that model need not rest on basic philosophical found- ations. It may be a pragmatic model, developed through experience within the legal system. Many lawyers perform their work with little or no jurisprudential knowledge,?’ and there is no evidence to suggest that they are worse, or better, at their jobs than lawyers well-versed in jurisprudence. Susskind concedes that it is possible to build a legal expert system “without jurisprudential insight”, but suggests that such a system would be of very poor quality: Because successful legal knowledge engineering presupposes so profound a familiarity with the nature of law and legal reasoning, it is scarcely ima- ginable that such a mastery could be gained other than through immersion in jurisprudence.*° Harris, himself a jurisprudent, takes a different view: People acquire those technical skills of legal reasoning and legal argument- ation which make up the concept of ‘good lawyer’ by immersing themselves in substantive legal subjects. Jurisprudence has to do, not with the law- yer’s role as a technician, but with any need he may feel to give a good account of his life’s work—either to fellow citizens, or to himself, or to any gods there be.*! If Harris is right—and the existence of good (though jurisprudentially illiterate) lawyers suggests that he is—then the importance of jurisprudence to legal expert system design is questionable. A legal expert system need only operate at the same level of abstraction as does a lawyer, rather than at the philosophical level of a jurisprudent. The fact that many lawyers have mastered the process of legal reasoning, without having been immersed in jurisprudence, suggests that it may indeed be possible to develop legal expert systems of good quality without jurisprudential insight. This does not mean that legal expert systems designers should completely ignore jurisprudential literature. However, as a legal expert system need not conform to any jurisprudential theory, a pragmatic approach to expert system design may be preferable to a jurisprudentially pure one. Susskind, as Niblett complains, “gives the role of jurisprudence in the design of legal expert systems a greater significance than it deserves.” §2.2. Jurisprudence 13 It is well beyond the scope of this thesis to cover “the vast corpus of apposite jurisprudential material”,?? however several areas of jurisprudence of relevance to expert system design are discussed below. 2.2.2 Scientific and mechanical jurisprudence The dictionary defines jurisprudence as: The science which treats of human laws (written and unwritten) in general; the philosophy of law.+ Pound refers to the “scientific character” of the law: Sir Frederick Pollock gives us the clew when he defines the reasons that compel law to take on this scientific character as three: the demand for full justice, that is for solutions that go to the root of controversies; the demand for equal justice, that is a like adjustment of like relations under s; and the demand for exact justice, that is for a justice whose operations, of action. In other words, the marks of a scientific law are, conformity to reason, uniformity, and certainty.°° within reasonable limits, may be predicted in advance This approach to the law has been much criticized. Frank contends that uncer- tainty is inherent in the legal process, and that seeking certainty in general legal principles is simply “an expression of infantile emotional attitudes which have persisted into adulthood.”*6 In his 1908 paper, Mechanical Jurisprudence, Pound distinguishes scientific jurisprudence from mechanical jurisprudence: Roman law in its decadence furnishes a striking example [of mechanical jurisprudence]. The Valentinian “law of citations” made a selection of jur- isconsults of the past and allowed their writings only to be cited. It declared them, with the exception of Papinian, equal in authority. It confined the judge, when questions of law were in issue, to the purely mechanical task of counting and of determining the numerical preponderance of authority. Principles were no longer resorted to in order to make rules fit cases. The rules were at hand in a fixed and final form, and cases were to be fitted to the rules.°” By contrast, Pound says administration of justic Loevinger also argues a role for science in law, but sees little science in juris- prudence. In 1949 he proposed an alternative to jurisprudence which he termed jurimetr scientific law is “a reasoned body of principles for the 38 The next step forward in the long path of man’s progress must be from jurisprudence (which is mere speculation about law) to jurimetrics—which 14 Chapter 2: Legal analysis systems is the scientific investigation of legal problems ... The inescapable fact is that jurisprudence bears the same relation to a modern science of jurimet- ric strology does to astronomy, alchemy to chemistry, or phrenology to psychology. It is based upon speculation, supposition and superstition; it is concerned with meaningless questions; and, after more than two thousand years, jurisprudence has not yet offered a useful answer to any question or a workable technique for attacking any problem.°? This proposal spawned a new field of study, some work from which is discussed in §2.3 below. Loevinger calls those who fear the dangers of mechanized jurisprudence “quix- otic and uncomprehending” “°—yet he was not a mechanical jurisprudent himself (as explained in §2.3). Others, however, saw the development of computers as an opportunity to develop a judgment machine: a machine that could replace a judge. 2.2.3 Judgment machines In 1955, Lasswell predicted that: When machines are more perfect [sic] a bench of judicial robots ... can be constructed. The machine would apply a system of “weights” to allegations of fact made by parties to a controversy, and also to the justifications advanced in support of the claims put forward by participants. Litigation can proceed by counsel for the plaintiff and the defendant pressing buttons that translate their cases into the physical signs built into the machine. Many results would be “no decision.” However, the machine could be designed to settle a controversy of this kind with a “random” operation (by lot).44 He was not altogether serious: It is a challenging task for legal historians to assist in constructing a robot whose weights would give substantially the same result as those produced by the [US Supreme] Court at various periods. The task would not be too difficult for some justices on some issues. But a robot facsimile of the less repetitive members of the Court would provide a genuine challenge to the 42 engineers. However, some have taken the idea of a judgment machine very seriously indeed. In 1949, Frank wrote that a “logic machine” might “disclose all possible available alternative legal rules,” although judges would still have to exercise the “the sovereign prerogative of choice” between the rules on the basis of the judges’ “conscious or unconscious notions of policy.”‘? Similarly Mehl wrote, in 1959, that a machine could perform some of the functions of a judge, but that a role for humans would remain because “the solution to a legal problem may depend upon extra-rational factors, involving the whole of human experience”. §2.2. Jurisprudence 15 As recently as 1977, it has been seriously suggested that a judgment machine could replace a human judge. D’Amato proposed that a machine could take the relevant facts of a case as input and produce a number in the range —1 to 1 (where a positive number indicates a victory for the plaintiff). Given the multiplicity of factors, he claims, a result of zero would be extremely unlikely,’ although it is not clear why this would be so. These facts would be determined by a jury, but the law would be decided by the machine. Somewhat grudgingly, he allows for some vestige of human control: an appeal court could review all of the machine’s determinations in a certain numerical range (e.g. —0.05 to 0.05) within which the cases would be so close that a re-examination might be required. The review court’s subsequent decision would then be incorporated into the system.*® The idea of human judges being replaced by machines has been trenchantly criticized. According to Weizenbaum: The very asking of the question, “What does a judge ... know that we cannot tell a computer?” is a monstrous obscenity. That it has to be put into print at all, even for the purpose of exposing its morbidity, is a sign of the madness of our times. Computers can make judicial decisions ... They can flip coins in much more sophisticated ways than can the most patient human being. The point is that they ought not be given such tasks. They may even be able to arrive at “correct” decisions in some cases—but always and necessarily on bases no human being should be willing to accept.47 But D’Amato sees advantages in replacing human judges by machines: Would we lose a judge’s “judgment,” and how important would such a loss be to our legal system? Surely computers do not make “judgments” the way humans do, and so we would lose the “human” aspect of legal judgments. But what specifically do we lose when we lose the humanne: of judgments? Is human judgment just a euphemism for arbitrarin« discretion, or bias?4® By contrast, Stone stresses the importance of legal change springing from “devi- ance, tentativeness, and even indecision in judgment.” It is these phenomena above all which promote a judge’s sensitivity to new ideas or to newly perceived social situations and stir Hamlet-like introspec- tion. On these there rest some of the main foundations of the social good we call “justice.” 49 Proponents of the idea of automated judges claim that such systems would reduce the cost of the legal system, find inconsistencies in the law, and provide a level of certainty in the law which does not exist at present.°° D’Amato claims that a judgment machine would allow people to live under the rule of law and not under the “rule of persons.”°! 16 Chapter 2: Legal analysis systems D’Amato’s argument is based upon “two bold initial premises.” He assumes that, from a jurisprudential point of view, the law has been made completely determinable. He also assumes that, if a computer can be programmed to make judicial decisions, human “discretion” will have been completely removed.*? His views on judgment machines are extreme, and have not been widely ac- cepted. For the most part, expert system designers have focused on building systems which provide advice to lawyers or to laypeople, rather than usurping the role of judges. However (as is discussed in §2.4 below) many expert systems designers have made D’Amato’s first assumption—that the law is completely determinable—often tacitly. 2.2.4 Petrifaction of the law The idea of a judgment machine removing uncertainty and ambiguity in the law raises the possibility of the petrifaction of the law. This possibility is also relevant to legal expert system design. D’Amato claims that, once programmed, the law would become settled. The computer would stop “progress,” although the legislature could always step in if anomalous results were being produced. Similarly, Kayton believes that the use of propositional logic could identify legal ambiguity and provide guidelines for its resolution. He asks: Would not each concrete demonstration of ambiguity and its resolution tend to rigidify the law and eventually destroy the very flexibility which has made the common law viable? ... Despite the dangers of general- ization, it is submitted that ... [this question] should be answered in the negative. Information brought to bear in a rational pursuit is always better than ignorance or confusion. The elimination of ambiguities, rather than ossify the law, would produce an optimum condition for the purposeful, intelligent, and efficient development of the law.>4 Schubert, another early researcher in this field, is not convinced of the possibility, or the importance, of certainty in law. He writes: ... the ideal of certainty in law is tolerable only in the context of an em- pirical world in which forces inducing change are so manifold that the attainment of the goal is never possible.®® Tyree argues that a deterministic computer judge could “overrule himself” if cases are added as they are decided.*° But as Pound pointed out more than ninety years ago, the problem is more than mere ossification of the law: The effect of all system is apt to be petrifaction of the subject systemat- ized. Perfection of scientific system and exposition tends to cut off indi- vidual initiative in the future, to stifle independent consideration of new problems and of new phases of old problems, and to impose the ideas of one generation upon another.°” §2.2. Jurisprudence 17 This problem is inherent in expert system design—in the law and in other domains. There are dangers, for example, in the use of a legal expert system by judges. Stone considers the use of predictive techniques in the law and warns that: . reliance at the judgment seat on new predictive techniques would ser- iously threaten the judge’s central concern with justice. For if the results thus predicted for him do affirmatively guide him in present decisions, each judge will tend to vote somewhat more consistently with his past record. The margin of deviation would accordingly disappear in deference to pre- dicted patterns, aided by the human tendency to follow the less agonizing because already trodden path .. 5° Judges, Stone says, have a duty to strive wholeheartedly for justice “at the mo- ment of judgment.”°? Although an expert system may be of some use to judges of lower courts, there is a critical border between a machine serving the legal order and the dangers of subversion of that order. For whenever an appellate judge faces the duty to do justice in thi: he must address himself to it with his own present experience and insights. It would be a corruption of justice for him to shortcut this duty by resting on predictions of his future decisions based on his own past behavior—still is the average past behavior of a group of judges.° nse, more so if the bas So, as Stone points out, the “pioneers and experts of the new techniques” must face “the limits of the contributions they can make.”®! 2.2.5 Clear rules and clear cases H.L.A. Hart was the most famous of the positivist legal theorists. His major work, The Concept of Law, was published in 1961. Hart realizes that because the law is expressed in natural language, it is subject to considerable semantic indeterminacy. This he terms the open texture of law. However, he claims that it is possible to use rules deductively to solve “clear cases”: that is “those in which there is general agreement that they fall within the scope of a rule.”® He also contends that: ... the result of the English system of precedent has been to produce, by its use, a body of rules of which a vast number, of both major and minor importance, are as determinate as any statutory rule. They can now only be altered by statute, as the courts themselves often declare in cases where seem to run counter to the requirements of the established precedents.®% the ‘meri 18 Chapter 2: Legal analysis systems As Susskind writes, one ramification of Hart’s analysis is that: ... all expert systems in law whose inference procedures are solely deductive will function exclusively in the clear case domain, and will be of no aid in the solving of “problems of the penumbra.”°! But what constitutes a “clear case” is itself far from clear. As Hart concedes: . it is a matter of some difficulty to give an exhaustive account of what makes a ‘clear case’ clear or makes a general rule obvious and uniquely applicable to a particular case.° Moles disputes the very existence of clear rules and clear cases. He performs a detailed analysis of the application of an example of an ostensibly clear rule in a British statute: a provision which prescribes the circumstances in which an injunction should be issued in domestic violence cases.°° Moles summarizes the effect of three cases as follows: (1) Before B v. B®’ we are concerned with a statutory provision which on its face appears to be clear and comprehensible (at least to non-lawyers) and it would appear that an injunction should issue in cases such as those we have looked at. (2) After Cantliff, the provision has been considered twice by the {English] Court of Appeal within two weeks. All of the six judges who considered the matter are in agreement that the injunction should not issue and that the rule is clear. (3) The matter is further considered by a specially constituted Court of Appeal of five,®? who decide that the rule is clear, but different from that in (2), and that the injunction should issue.” Moles also analyses the judicial application of the rule of precedent—which, with this statutory rule, “must be regarded as amongst the clearest available to us”’!— to demonstrate that “our experience of the legal system bears little relationship to Hart’s account of it”.” Leith developed a legal expert system which operated on “clear rules” in the law.’ He later recanted, saying that: ... the very idea of a clear rule is inherently confusing and is not observable in the real operation of the judicial process ... judicial creativity is not an aberation [sic] of the legal process, but (as Moles ... suggests), the very heart of the law." As is explained in §2.4 below, many expert system designers have adopted a view of the law that allows for clear rules and clear cases—without considering the possibility that the law may not be like that at all. §2.2. Jurisprudence 19 2.2.6 Legal realism and rule scepticism In the 1920s and 1930s, a movement called legal realism developed in America.” Realists rejected the importance that mechanical jurisprudence placed on rules. However, the rule scepticism of the American Realists is meek by comparison with the strong indeterminacy thesis. Drahos and Parker explain that: According to Karl Llewellyn a characteristic feature of Realism was the rejection of simple (by which he meant general) rules and the substitution of more detailed classificatory schemes which better captured the specific nature of judicial rule-making ... On Llewellyn’s account, rule scepticism emerges as a set of doubts about the veracity of legal actors’ claims to be following the legal rules they say they are. This is not rule scepticism in the strong sense of denying the existence of rules, however. Legal actors may simply be following some other rules. The rule scepticism of American Realism could perhaps be more accurately described as rule cynicism.” Kripke, by contrast, is a true rule sceptic. His argument is an example of the Wittgensteinian Paradox. Consider the two functions “plus” (+) and “quus” (@). The + function is the mathematical function, addition. The @ func- tion is defined as follows: nm, fury, ife,y < 57; zou { 5; otherwise. Suppose that Kripke has used + in the past, but always with values of a and y smaller than 57. He performs the computation 68 + 57 and gets a result of 125. Yet it could be said that when he thought he was using the “plus” function in the past, he was in fact using “quus.” As Kripke explains: ... in this new instance, I should apply the very same function or rule that I applied so many times in the past. But who is to say what function this was? In the past I gave myself only a finite number of examples instantiating this function. So perhaps in the past I used ‘plus’ and ‘+’ to denote ... ‘quv Who is to say that [‘®’] is not the function I previously meant by ‘+7?75 There is no justification for Kripke answering 125 rather than 5. There is no way of determining (from his past behaviour—even his past thoughts) whether by “plus” Kripke meant + or @. Rules are derived from a finite number of examples; for any given rule, there is always an alternative rule which also explains those examples. Mathematicians could counter that the meaning of + is well defined. But Kripke argues that: ... scepticism about arithmetic should not be taken to be in question: we may assume, if we wish, that 68 + 57 is 125 ... I cannot doubt coherently that ‘plus’, as I now use it, denotes plus! Perhaps I cannot ... doubt this 20 Chapter 2: Legal analysis systems about my present usage. But I can doubt that my past usage of ‘plus’ denoted plus ... . There is no objective fact—that we all mean addition by ‘+’, or even that a given individual does—that explains our agreement in particular cases. Rather our license to say of each other that we mean addition by a brute fact that we generally agree. is part of a ‘language game’ that sustains itself only because of the 79 Drahos and Parker write that: The consequences of this argument, if valid, are shattering. Rules turn out to be no more than leaps in the dark and the whole notion of rule following seems illusory.°° The implications of Kripke’s argument have been realized by legal theorists.*! As Drahos and Parker point out: Saying that there are no rules to follow, only social practices, means that propositions about the law are potentially open to wild fluctuations. The argument also puts paid to any possibility of a correspondence theory of truth in law.5? They propose a solution to this problem: The Kripkean argument does not prevent people from saying that they are following rule X. Rather it stops them from being able to justify the existence of that rule by reference to some objective meaning. This still leaves the practice, as opposed to the justification, of successful rule following to be accounted for.*? Their solution is to view the law as a set of rules and conventions: “a type of rule used to fix or interpret the meaning of other rules.”*4 They claim there is a distinction between rule knowledge and rule understanding, the latter being “a matter of absorbing conventions relating to rule use.”®° A lawyer requires knowledge of the rules, and understanding as to how to apply those rules prop- erly. Drahos and Parker argue that the problems facing designers of legal expert systems “flow from the difficulties of representing rule understanding rather than rule knowledge.” Rules can be used to represent rule knowledge; the problem s “how to represent with tolerable accuracy ... conventions which confer rule understanding.” °° But they concede that, as with other rules, conventions have to “run the gauntlet of scepticism”.*” 2.2.7 A jurisprudential consensus? The only major examination of the role of jurisprudence in the development of legal expert systems is Susskind’s 1987 book Expert Systems in Law: A Juris- prudential Inquiry.8® As mentioned in §2.2.1 above, he argues that all expert systems must conform to some jurisprudential theory. §2.2. Jurisprudence 21 He also sets out to find consensus in jurisprudential theory. As there is far too much literature in the field for one person to cover it all, he limits his choice of material: “the vast majority” of sources that he uses are British analytical juris- prudential writings since the mid-fifties and early sixties, “the impetus for which”, he concedes, “was derived very largely from the publications of H. L. A. Hart.”®? Susskind concludes that “there are no theoretical obstacles, from the point of view of jurisprudence, to the development of rule-based expert systems in law of limited scope.”°° He also claims that the divergence of views within jurisprudence has been overstated because legal theorists tend to focus on the differences. He claims that there is a jurisprudential consensus, “albeit of mundane and limited application”.?! Susskind’s work provided the theoretical justification for not only his own development work,®? but for much of the subsequent work of other developers of legal expert systems. It could also be said to have provided, retrospectively, legal theoretical justification for most of the work on rule-based expert systems that preceded his book. However, Susskind’s approach to jurisprudence is fundamentally flawed. As Moles points out: {Susskind] said that he would carry out a survey of the jurisprudential literature. He acknowledged, of course, that it would not be possible to survey the whole of the jurisprudential literature. In fact, he determined that law was a system of rules by “surveying” only those whose avowed position was based on the fact that the law was a system of rules ... I would venture to suggest that this is in fact a misuse of the survey technique ... Susskind was perfectly familiar with the work of Hart and his followers, and was well able, therefore, to find any number of books and articles which supported the “law as rules” view. He then developed his position ... on the basis of what this purported consensus within jur- isprudence had to say. ... Of course, Susskind was telling certain sections of the AI and Law community what they wanted to hear, and hence their enthusiasm for it.” Moles also notes Susskind’s admission that “the most rigorous of these writings constituted the source materials with greatest potential given the overall purpose of the project.”®! Because the first objective of Susskind’s work was to design, develop and implement an expert system in law,®® Moles cites this admission as evidence that Susskind prejudged his survey. In the light of this criticism, it is ironic to note Susskind’s own caution that: ...a little jurisprudential knowledge can be a dangerous thing! It is tempt- ing for the jurisprudential neophyte to become an ardent devotee of a par- ticular school of thought within legal theory and to go on from there to implement all and only the teachings of that school. This course of action 22 Chapter 2: Legal analysis systems should be avoided at all costs. Familiarity with a wide range of works should be achieved prior to commitment to any particular jurisprudential posture.9° Clark, another critic of Susskind, says that: The shortcomings of the book coincide with the limits of positivist legal theory, and even then some doubt must remain as to whether expert sys- tems in law mark a revival of the kind of “mechanical jurisprudence” which Hart opposed so vigorously from within the confines of positivist legal thought.°7 But the strongest criticism of Susskind comes from Leith: ... [Susskind] believes that he can come to some sort of compromise with the various theoretical positions taken by the renowned thinkers of the field and produce “a general theory” (an indication, I might suggest, of {his] poor theoretical conceptions) ... ... Susskind sees jurisprudence as providing a variety of theoretical models which can be modelled mathematically and translated into com- puter programs. ... of neces y formal specification requires a formalisation of law; there can be no “informal models” which are mysteriously formalised into a computer model ... Therefore, in order to use formal specifications, Susskind must provide a formal specification of law which can then be incorporated into the rule-format of a computer program ... And, of course, formal specific- ations of law are renowned for their theoretical and practical inadequacy. Legal formalism can, surely, hardly be the compromise he wishes might arise from the conflicting positions of Kelsen, Hart, Dworkin et al.°* ... if he really does believe that his informal theoretical models can be transformed into formal theoretical models without loss of their informal attributes, then I must suggest that he has really little understanding of the discipline of computer science.°? Leith’s comments are consistent with the point made in §2.2.1 above: lawyers and legal expert systems operate at a lower level of abstraction than the philosophical level of jurisprudence. Susskind’s claim to have found a consensus in jurisprudence—even one of mundane and limited application—is absurd. As shown in §2.2.5 above, the views of just two jurisprudents (Hart and Moles) are completely irreconcilable. Similarly, Hart’s views are totally at odds with those of Kripke (as explained in §2.2.6). Susskind states his theory, and develops an expert system which conforms to that theory. His mistake is to claim that his theory is definitive, in that it represents a jurisprudential consensus. $2.3 Jurimetrics and the behaviourists 23 2.3 Jurimetrics and the behaviourists As mentioned in §2.2.2, the term “jurimetrics” was coined by Loevinger. He defines “jurimetrics” as “the study of law and legal problems by scientific methods and concepts, the employment of science in law to the extent that it is applicable or adaptable.”! Loevinger is neither a mechanical jurisprudent nor a positivist: To begin with we must be clear that science offers us neither ultimate nor certain answers to legal problems. The dream that science might someday tell us which of several competing interests is the more important is a vain one. Science essays no such answers in any field. Science does not assign social or ethical values. Science may, indeed, provide data from which social or ethical judgments may be made; but the judgments will remain with man. ... There is no prospect of any process that will preclude consideration of social desirability or wisdom. The opportunity will always be available to argue that precedent should not be followed, and that considerations of policy, or expediency, require a different rule or a special result ... ... Science does not and will not offer us any law machines that give automatic answers to specific questions put to them, whether as to partic- ular cases or as to ultimate legal issues such as the relative importance of interests that may be in conflict. By the same token, science will provide us with no formulae or calculus that will give us certainty either of predic- tion, analysis or answers to ultimate questions such as which interest is to be preferred or which desire has greater social value.” Although these comments are eminently reasonable, some of Loevinger’s turns of phrase are exasperating. In 1949 he claimed that putting the law on a “rational basis” was the “indispensable condition” of the survival of the human race.* In the light of such a ridiculous statement, and after his trenchant criticism of the value of jurisprudence (quoted in §2.2.2 above), it is not surprising that, as Gardner says: The attitudes Loevinger and his colleagues expressed were never adopted by the legal profession generally ... As a movement within the legal pro- fession, jurimetrics has not been much heard from since the early 1970s.4 Nevertheless, some of the work of his colleagues in predicting judicial decisions is worthy of comment here. A number of researchers in the early 1960s focused on the statistical analysis of the behaviour of judges. On the basis of this analysis, they claimed that they could predict the future behaviour of individual judges, and of courts.° Such predictions were justified on the basis of the application of the doctrine of precedent. As Lawlor writes: s are akin to the lence. Without such 's impossible.® Even if they are man-made, the principles of stare all embracing assumption of uniformity of natural a principle to guide us, prediction of legal decisions 24 Chapter 2: Legal analysis systems 2.3.1 Kort Kort uses mathematical expressions to represent judicial decisions.’ He identifies attributes which are of importance and represents decided cases in the following form: ; Ayu = Vj i=1 where n is the number of attributes, Aj; is the value of the ith attribute for the jth case (attribute values being 0 or 1), w; is the weight of the ith attribute, and V; is the number of votes of judges favourable to the party seeking redress in the jth case. This approach is justified on the basis that the decisions of split courts do not necessarily constitute two opposite extremes, but represent certain degrees of support for one party.® By solving these simultaneous equations, weights are obtained for the attrib- utes. The sum of the weights of those attributes present in the instant case is, according to Kort, the number of likely judicial votes in favour of the party seeking redress. 2.3.2 Lawlor Lawlor uses logical expressions to represent previously decided cases.? He ana- lyzes right-to-counsel cases heard before the US Supreme Court over thirty years, and finds that in all of these cases each judge behaved consistently with his own “personal stare decisis.” He identifies legally significant attributes and builds a logical expression which represents the behaviour of each judge. A program uses these logical expressions to predict the likely outcome, given a composition of the Court specified by the user. Lawlor’s system successfully predicted the US Supreme Court’s overruling of Betts v. Brady in Gideon v. Wainright.! But it predicted a 5:4 majority; the Supreme Court’s decision was unanimous. Applying “traditional” stare decisis (i.e. considering the court as a whole, rather than the decisions of individual judges) Lawlor’s system did not predict the Supreme Court’s change. Given the same cases, Kort’s approach did not predict the change either. This failure draws criticism from those who disagree with such a “logarithmic approach to justice.”!? Wiener complains that “advocates of the computer” rest their arguments on an assumption that courts will adhere to the doctrine of stare decisis, which does not always hold.!* In defence of the behaviourists, Kayton refers to their failure to predict the decision in Gideon v. Wainright and writes: Should we have expected otherwise? Of course not! A reversal is by definition a logical inconsistency. That which by stare decisis had been called black is now called white.!4 $2.3 Jurimetrics and the behaviourists 25 The developers of these prediction systems accept that they rely on the doctrine of precedent. Hence, a reversal of previous authority will always be beyond the predictive capacity of these systems—as it is beyond the predictive capacity of most lawyers. 2.3.3. Nagel and Schubert Nagel and Schubert!® examine the personal attitudes of US Supreme Court judges towards various political and economic relations, and claim that these “off-the- bench” attitudes (as Nagel calls them) affect the judges’ decision-making. Correlations were ... made between responses to specific items and various decisional propensities. For example, there was a high and statistically significant correlation between disagreeing with [one questionnaire item] (“Our treatment of criminals is too harsh; we should try to cure, not to punish them”) and being above the average of one’s court with regard to the proportion of times one voted for the prosecution in criminal cases. Off-the-bench judicial attitudes thus do seem to correlate in a meaningful way with on-the-bench judicial decisions.!® 2.3.4 Haar, Sawyer and Cummings In the mid-1970s Haar, Sawyer and Cummings made use of regression analysis to build a predictive model for zoning amendment cases in Connecticut.!” Re- gression analysis is a statistical technique for analyzing the relationship of a set of independent variables to a dependent variable. For Haar et al. the dependent variable is the outcome of the cases, the independent variables are attributes. They identify 167 attributes which “appeared to be important” to the courts, 40 of which were deemed significant using a y? (“chi-square”) test for association. This is “too many to use in a regression analysis”,!® so Haar et al. employ two different methods to reduce this number further: grouping attributes on the basis of “experience, knowledge, and intuition”,!® and factor analysis. 2.3.5 Prediction The behaviourists’ predictive research has been criticized because they do not attempt to model legal reasoning.”° But, in not so doing, the behaviourists just reflect the influence upon them of the American Realists. They adopt Holmes’s analysis that for any individual “the law is simply a prediction of the way in which the public force possessed by the government will act upon him.”?! Loev- inger claims that some method of legal prediction is “indispensable” ;?? Lawlor says that the ultimate goal of all scientific methods is reliable prediction of future events and “(rjeliable prediction is also one of the ultimate goals of law.” Suss- kind (certainly no behaviourist) also stresses the importance of the prediction of judicial decisions.?* 26 Chapter 2: Legal analysis systems Tapper writes that: Although the statistical techniques employed by some of these workers have been criticized, there can be no real doubt that this work provides a successful approach to the analysis of the decisions which have been reached in the p: and at least as satisfactory a method of predicting future decisions as can be arrived at by native wit and unaided intuition.?° However, he criticizes the conclusions that have been drawn using these tech- niques: Either the behaviourist is to content himself with observing the objective phenomena, in which case he can conclude nothing as to motivation, or he is to ascribe motivation to the phenomena, in which case he ceases to be objective. It is precisely at the point at which decisions of the [US] Supreme Court, for example, are characterized as pro-civil liberties or anti-labour that doubts arise as to the real objectivity of the studies. The same general line of argument may be advanced against the fact- oriented approach ... Here the point at which the study loses its objectivity is in the characterization of the facts present in the case.?° (Although this may be a valid criticism of a behaviouristic study, it is not an argument against the use of predictive techniques where the characterization of facts has been performed—subjectively, admittedly—by a legal expert, as is done for SHYSTER.) Gardner is also critical of the jurimetric programs: ... programs, done from a political science viewpoint rather than a legal one, in which the data concern legally irrelevant matters such as the ideol- ogy and social background of individual judges.?” By “legally irrelevant matters” Gardner means those matters to which judges do not—and/or should not—explicitly have regard when coming to their decisions. But if a system is designed to predict decisions then any attribute which assists in prediction should be included—regardless of whether regard ought to be had to that attribute. These “legally irrelevant matters” are not necessarily irrelevant to the prediction of judicial decisions. Stone dismisses concern about “jurimetric” prediction: “the behavioralists [sic] may have no conscious designs on the integrity of the decisional process. The judgment of justice may be of no concern to them.”?5 He defines the “judgment of justice” by reference to a situation where a judge makes a decision which: ... does not merely declare the existing law but decides what justice re- quires that the law should be. (... [W]e include tacitly here the reversal of earlier decisions, that is, creative decisions which unsettle and resettle law: this is a fortiori creative.) It is this kind of creative judgment which we have here termed “a judgment of justice.”?9 $2.4 Rule-based systems 27 The behaviourists are, Stone says: . only observers looking at what has already been done in judgment. And when in the course of prediction they turn their attention to future judgments, it is to ask not what the judge should do to further justice but only what kind of decision he will give if he acts consistently with values attributed to him on the basis of his past decisions.%° In other words, their work attempted to be predictive not normative. When a lawyer gives advice, she/he is expected to make some prediction as to the likely result. This prediction need not be emphatic: it may be no more than a tentative statement about the strength of a person’s legal position. Hence, legal expert systems should have some degree of predictive capacity. Like the behaviourists’ research, a legal expert system’s predictive power is a projection based on past cases. There is—and can be—no allowance for changing social mores. This is not a serious limitation upon such a system’s utility: predicting a judge-made change in the law is beyond all but the very best lawyers. However, as discussed in §2.5 (especially §2.5.2) below, the justification that a legal expert system provides for its prediction is also important. 2.4 Rule-based systems There are many examples of rule-based legal expert systems, the most important of which are discussed in detail below: the work of McCarty (§2.4.1), Bench- Capon, Kowalski and Sergot (§2.4.2), Gardner (§2.4.3) and Susskind (§2.4.4). The first proposal for a rule-based legal expert system was made by Buchanan and Headrick in 1970. They complained that interdisciplinary work between lawyers and computer scientists had “floundered on the misconceptions that each has of the other’s discipline”,*! and suggested that the computer modelling of legal reasoning would be a fruitful area for research.*? Their proposal asserts that “{iJn the absence of any reason to speculate on how they carry on their work, [lawyers] now apply complex sets of rules without being aware of the rules themselves.”*? They make no reference to jurisprudential writ- ing, but they do make it clear that their approach is based upon two assumptions about human problem-solving in general: “(1) problems can be broken down into a set of subproblems, and (2) the solution to any subproblem requires a series of decisions that are governed by decision rules.”34 The literature is replete with examples of projects in which the assumption that lawyers work with rules is unstated or unsupported. Maggs and deBessonet, for example, tacitly make this assumption.*° The law, they claim, can be ex- pressed as rules using propositional calculus. A program which implements these rules could answer questions from a user, and allow the checking of statute law for redundancy and contradiction. (It is not clear what role case law plays in their system, or indeed in their model of the law.) 28 Chapter 2: Legal analysis systems Popp and Schlink’s JUDITH system*® uses rules to represent parts of the Ger- man Civil Code. The principles behind JUDITH are “strikingly similar” to those of the MYCIN system (an expert system dealing with bacterial infections®” )—so similar that, its developers claim, it would be possible to create a legal knowledge base for MYCIN and a medical knowledge base for JUDITH.*® Michaelsen and Michie’s TAXADVISOR system®® is implemented using MYCIN.“° TAXADVISOR was designed to assist lawyers to advise clients on taxation and estate planning.’ Meldman’s system uses two different kinds of rules: general rules which define the elements of the claim, and specific rules extracted from cases.’? Things and relations are used to represent the “everyday world of human affairs”,*? and are classified hierarchically into categories. A fact comprises two things and a relation between them; facts are assembled into situations. These situations are compared with the situation of the instant case, and the system determines the extent to which the instant case falls within or near the law of intentional torts (e.g. assault and battery). Waterman and Peterson’s LDS system“ is a rule-based system for the field of product liability. It does not determine whether liability exists, but is designed to assist legal experts in settling product liability cases. LDS is a typical example of a system which developed without any legal theoretical justification. After admitting that there is no “deep model of the legal process”,*° Waterman and his colleagues proceed on the unstated assumption that any such model must involve rules: One might expect that the large body of legal rules and regulations that have been accumulated and formalized in the legal domain would make expert system development easier. Unfortunately, this is not the case. Instead, this characteristic of the domain, having rules that already exist, has led to trouble ... First, the formal rules that define and regulate legal activity are often ambiguous, contradictory and incomplete. And second, there exists a body of informal rules or procedures about how to access, interpret and use the ‘formal’ rules. Without these informal rules the formal rules can not be used in any efficient or cost-effective way.*® AT This body of rules, they write, “needs to be mapped into code’ expert system can be built. before a legal Bing’s SARA system*® is designed to analyze discretionary decisions. It uses rules to represent legal norms: some strict and some discretionary. These discre- tionary norms are weighted using correlation techniques. SARA allows a lawyer “to back up his qualitative legal reasoning by quantitative indications.”“” Stamper’s LEGOL language” also uses rules to represent legal norms. Pattison and Ciesielski use a rule-based system to review contracts.°! SoftLaw’s STATUTE, a commercially successful system, uses rules to represent statutes, regulations and departmental guidelines in taxation, social security and veterans’ affairs law.>? $2.4 Rule-based systems 29 2.4.1 McCarty McCarty has been described as “the father of AI and law”? His TAXMAN project>4 was concerned with the development of a computational theory of legal reasoning, using corporate tax law as an experimental problem domain. He claims that a computer-based legal consultation system must be able to represent the “facts,” at some comfortable level of abstraction, and the “law,” which would consist of a system of “concepts” and “rules.” These concepts and rules are relatively abstract (i.e. they subsume large classes of lower-level factual descriptions), and they have normative implications (i.e. they specify which actions are permitted and which are obligatory). Legal analysis, in its implest form, would then be a process of applying the “law” to the “facts”. Put this way, the paradigm seems to be an ideal candidate for an artificial intelligence approach: the “facts” would be represented in a lower-level semantic network, perhaps; the “law” would be represented in a higher-level semantic description; and the process of legal analysis would be represented by a pattern-matching routine.®> However, McCarty concedes, the representation of facts in such a system is more difficult than in other expert systems, because “the facts of a legal case typically involve all the complexities of daily life: human actions, beliefs, intentions, mo- tivations, etc.”°° Even if the facts can be represented, he writes, the rules will often be problematic: Some rules, usually those embodied in statutes, have a precise logical struc- ture, and this makes them amenable to the existing artificial intelligence techniques. But it is a commonplace among lawyers that the most im- portant legal rules do not have this form at all: instead they are said to have an “open texture”; their boundaries are not fixed, but are “construc- ted” and “modified” as they are applied to particular factual situations. A sophisticated legal consultation system would not be able to ignore these complexities, but would have to address them directly.°” McCarty also makes the startling claim that the “simplest” problems for first- year law students are the hardest for an AI system because the student draws upon ordinary human experience: Paradoxically, the c: system are those cast s that are most tractable for an artificial intelligence usually involving commercial and corporate mat- most complex. There is a simple reason why this is so. A mature legal system in an industrialized democracy is composed of many levels of legal abstractions ... Because of their technical complexity, the legal rules at the top levels of this conceptual hierarchy are difficult for most lawyers to comprehend, but this would be no obstacle for an artificial intelligence system.°* ters, which a lawyer fin 30 Chapter 2: Legal analysis systems McCarty chose the area tax law for his system because “commercial abstrac- tions, in fact, are artificial and formal systems themselves, drained of much of the content of the ordinary world” and, by legal standards, well structured.®? The field of corporate tax law, he says, is “very near the apex of the hierarchy of commercial abstractions” °°— “whatever that means”, comments Moles.®! * * c In TAXMAN I, McCarty’s first prototype system, the basic “facts” of a corporate case are captured in a relatively straightforward representation (e.g. a corporation issues securities). Below this level is an expanded representation of the meaning of various entities (e.g. a security interest) in terms of their component rights and obligations. Above this level—presumably above both levels, although this is not made clear—is the “law” (statutory rules which classify transactions as taxable or non-taxable ete.).° McCarty found that, although the rules are complex, the underlying represent- ations are manageable. He concludes from his early work that “the construction of an expert consultation system in this area of the law is a feasible proposition.” McCarty sees the development of legal expert systems as an opportunity to contribute to jurisprudence. Although the jurisprudential literature includes “many illuminating examples and many valuable insights about the structure and dynamics of legal concepts”, he complains that “taken as a whole” it is “no- toriously imprecise”.®* The TAXMAN system adds a strong dose of precision and rigor to these discussions of linguistic and conceptual problems. Its critical task is to clarify the concepts of corporate reorganization law in such a way that they can be represented in computer programs.°°* Moles is extremely critical of rule-based expert systems designers, and Mc- Carty in particular. McCarty takes a positivist approach to the law, and Moles complains that: The computer scientists have taken the de-humanizing aspects of modern positivism to their extreme. ... The computer scientists, encouraged by the modern positivists, fail to recognize the point which Austin® correctly emphasized throughout his work—that law, positive morality and ethics are inseparably connected parts of a vast organic whole.” Most damning is Moles’s statement that: The sad thing is that [McCarty] has not shown the slightest awareness of the nature of the legal enterprise. Far from having emphasized any difficulties, he shows that he simply does not understand what they are.® $2.4 Rule-based systems 31 McCarty does identify two major limitations to the approach taken in TAX- MAN I. Firstly, he concedes that the factual descriptions, although manageable in the corporate domain, would be too complex in (for example) the average con- tract or tort problem. Secondly, the higher-level conceptual representations are not adequate for all domains because judicially created concepts are incurably open-textured, have a dynamic structure with the capacity to evolve and adapt to new situations, and the evolution of these concepts is governed by a sense of purpose.®® Having recognized these problems, McCarty set out to solve them with TAXMAN IL. * * * TAXMAN II” uses “prototypes and deformations” to represent a legal concept by specifying the prerequisite conditions for that concept, a set of cases (real and hypothetical) in which that concept does or does not apply, and a set of transformations for getting from one case to another. If a given case (representing a legal concept) can be transformed into the instant case, while still satisfying the prerequisite conditions for the concept, then it can be used in argument as an example of that concept. Ashley sees McCarty’s work as a significant advance, but points to several shortcomings of TAXMAN II: The work is largely an exercise in knowledge representation. McCarty does not set forth a control or process model that clarifies how a program would actually generate a legal argument ... The reported research involves a hand simulation of the arguments in one US Supreme Court case"! and work done on hand simulations of several subsequent cases. ... [TAXMAN II] has no mechanisms for comparing cases in terms of how on point they are, for distinguishing cases, or selecting the best pre- cedents ... McCart 4 y’s model assumes a much neater domain than exists in law. He assumes that in reality, legal cases are consistently allocated as positive and negative exemplars of concepts. They are not. He assumes that ther near match between concepts and the features of a case that are relevant to the concept. There is not.”? isa McCarty’s definition of legal primitives has also been criticised. Moles writes: McCarty appears not to appreciate that ‘corporations’, ‘securities’, ‘prop- erty’, ‘dividends’ and so on are not subsumed ‘beneath the law’, but are each the products of complex legal analy The question of whether certain transactions are taxable or not is intimately tied into that legal analysis.” Ashley makes a similar point: The problem with such primitives, if they are taken seriously as a means for defining concepts, is that they assume what is to be shown. Far from being a primitive, that someone has a right or a duty in a given fact situation is an arguable legal conclusion that must be justified by citing authorities. 32 Chapter 2: Legal analysis systems 2.4.2 Bench-Capon, Kowalski and Sergot Bench-Capon, Kowalski, Sergot and their colleagues use PROLOG to model stat- utes.”° Their approach to legislation is the most extreme of all expert system developers, due to their attitude towards knowledge acquisition. They write: The formalisation of legislation by means of rules has almost all the charac- s of an expert system. It differs, however, in one important respect. ical expert system, before knowledge can be formalized, it has to be elicited from the subconscious of an expert. Eliciting this knowledge is generally regarded as the main bottle-neck in the construction of expert systems. It is entirely absent, however, in the case of legislation which is already formulated and written down. Thus the use of expert system tech- niques for representing legislation has virtually all the advantages of expert systems without the attendant disadvantages of eliciting the knowledge.” This statement is nothing less than astounding. Even if it is accepted that statute law can be represented using PROLOG clauses, it is a bold claim indeed to assert that constructing those clauses requires no expertise. As Moles says, “{c|learly these researchers do not distinguish between the writing (which is the legislation) and the meaning of that writing.” Bench-Capon et al. have worked with several statutes. Most famous is their work with the British Nationality Act 1981 (UK).”* Consistent with their ap- proach to knowledge acquisition, their PROLOG representation of the British Na- tionality Act was implemented in two months by a student, “without any expert legal assistance.””? They also represented the Supplementary Benefits Act 1976 (UK) and its reg- ulations using a similar method, and made a similar claim about the importance of legal expertise: For our project, the accuracy of the representation was not a critical con- sideration at this [early] stage. Our formalisation could therefore be un- dertaken with no expert legal assistance ... In general, accuracy of the formalisation is, of course, critical, particularly if one were constructing a representation to be used in practice. Predictably—and correctly—this approach has been strongly criticized. Moles points out that: This is to assume ... that the problem with regard to “accuracy” is merely a matter of changing the detail of content. It fails to appreciate that an expert may have a great many useful things to say about how one goes about the process of interpretation. The expert advice will therefore have implications for the method being employed and the way in which the knowledge is structured.*! $2.4 Rule-based systems 33 Kowalski and Sergot make this assumption explicit: Access to an expert adviser might well have chang