法律实践变革：人工智能驱动的法律分析与预测

需积分: 9 165 浏览量更新于2024-07-19 收藏 39.16MB PDF 举报

《人工智能与法律分析》是一本由剑桥大学出版社于2017年出版的专业书籍，作者是Kevin D. Ashley教授，他是一位法学与智能系统学教授，同时在匹兹堡大学、学习研究与发展中心担任高级科学家，并在计算机科学方面拥有兼职教授身份。这本书聚焦于人工智能（AI）在法律领域的革命性转变，这一变革起源于像IBM的Watson和Debater这样的文本分析程序，以及它们所依赖的开源信息管理架构。随着科技的进步，新的法律应用逐渐崭露头角。本书特别关注如何通过将计算模型的法律推理与法律文本直接连接，让非程序员也能理解这些技术。书中阐述了AI如何改变法律实践，例如，通过生成针对特定结果的正反论证，预测判决结果，并提供律师可以自行评估的理由，从而增强概念性的法律信息检索能力。这种技术促进了认知计算的发展，使得人类与计算机之间的合作更加紧密，两者各自发挥其擅长的智能活动。通过这种方式，AI不仅能够支持律师们处理大量法律文献，提升效率，还能帮助他们进行深度分析，提高决策的质量。书中所述的应用将法律专业人士引入一个全新的工作模式，即结合人工智能的力量进行法律分析和策略制定，而不再局限于传统的文献查找和逻辑推理。阅读这本书对于对法律界中人工智能发展感兴趣的读者来说，无疑是一份宝贵的指南。它揭示了AI如何重塑法律行业的未来，让法律实践变得更加智能化和精确化。作者Kevin D. Ashley凭借其深厚的专业背景和跨学科研究经验，为读者提供了深入浅出的视角，使读者能够理解和掌握AI在法律领域所带来的革新。

“FM” — 2017/5/31 — 10:36 — page xv — #15

Illustrations

1.1 Heuristic rules dening loss and strict liability (Waterman and

Peterson, 1981) 9

1.2 ATN for offer and acceptance problems with four states: (0) no

relevant legal relations; (1) offer pending; (2) contract exists; (12)

contract exists and proposal to modify is pending (Gardner, 1987, p. 124) 19

1.3 Gardner’s heuristic method for distinguishing hard and easy legal

questions (Gardner, 1987; Rissland, 1990) 22

1.4 Argument diagram of IBM Debater’s output for violent video games

topic (root node) (see Dvorsky, 2014) 24

1.5 Diagram representing realistic legal argument involving violent video

games topic 25

1.6 Architecture of text analyzer for legal documents including contracts.

Dashed boxes show components for semantic analysis and conceptual

information retrieval 29

2.1 Normalized versions of two alternative interpretations of the Louisiana

statute and corresponding Prolog rules (bottom) (Allen and Engholm,

1978) 42

2.2 IRC section 354 and a normalized version (right) (see Allen and

Engholm, 1978) 43

2.3 Flow chart for propositionalized IRC section 354 (see Allen and

Engholm, 1978) 44

2.4 BNA provisions as represented in rules (Sergot et al., 1986) 47

2.5 BNA program output (excerpts) (Sergot et al., 1986, p. 376f) 49

2.6 Norm graphs for concluding “Legality” in section 4 (1) FDPA and

“Effective Consent” (see Oberle et al., 2012, pp. 305–6, Figs. 13 and 14) 59

of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781316761380

Downloaded from https://www.cambridge.org/core. University of Florida, on 03 Nov 2017 at 09:07:18, subject to the Cambridge Core terms

“FM” — 2017/5/31 — 10:36 — page xvi — #16

xvi Illustrations

2.7 Sample BPMN diagram of simple insurance claim process with

business rule annotations (see Table 2.1) (Koetter et al., 2014,

Fig. 2, p. 220) 60

2.8 Petri net representing simple producer–consumer resource allocation

problem (see Kafura, 2011, p. 8) 66

2.9 Compliance system report of traces, rules, and tasks responsible for

noncompliance (excerpts) (see Governatori and Shek, 2012) 69

2.10 Statutory network diagram comparing Pennsylvania (PA) and Florida

(FL) statutory schemes re public health emergency surveillance:

Circles indicate public health system actors and partners in FL

and PA. Grey links indicate relationships present in both states;

white links indicate legal relationships present in PA but not in FL

(Sweeney et al., 2014) 71

3.1 Secrets-Disclosed-Outsiders dimension in Ashley (1990) 82

3.2 Hypo-style three-ply argument for the Mason case (see Ashley, 1990) 85

3.3 Hypo argument model with Venn diagram (Ashley, 1990) 86

3.4 Hypo claim lattice (Ashley, 1990) 87

3.5 Example of CABARET’s process for analyzing Weissman v. IRS, 751

F. 2d 512 (2d Cir. 1984) (Rissland and Skalak, 1991) 89

3.6 CATO Factor hierarchy (Aleven, 1997, 2003) 92

3.7 CATO argument downplaying/emphasizing distinction (Aleven, 2003) 92

3.8 GREBE semantic net representation of Vaughn case (Branting, 1991,

1999) 93

3.9 GREBE matches structure of Vaughn case to Jarek problem (Branting,

1991, 1999) 94

3.10 In-furtherance-of employment cases retrieved by GREBE for Jarek

problem (Branting, 1991, 1999) 96

3.11 Excerpts of GREBE’s argument for Jarek problem (see Branting, 1991,

1999) 97

3.12 Which argument better accounts for teleological concerns? (Berman

and Hafner, 1993) 99

3.13 Theory constructed from factor and value preferences (Bench-Capon

and Sartor, 2003) 101

3.14 Argument as theory construction from factor and value preferences

(see Bench-Capon and Sartor, 2003) 102

4.1 Projection of capital gains tax cases onto two dimensions (see Mackaay

and Robillard, 1974) 109

4.2 Bail decisions data (a) from which decision tree is constructed (b) 110

4.3 Example features from Supreme Court database [S], the Segal–Cover

Scores [SC], and feature engineering [FE] (Katz et al., 2014) 112

4.4 IBP domain model (Ashley and Brüninghaus, 2006) 116

4.5 IBP algorithm (Ashley and Brüninghaus, 2009) 117

of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781316761380

Downloaded from https://www.cambridge.org/core. University of Florida, on 03 Nov 2017 at 09:07:18, subject to the Cambridge Core terms

“FM” — 2017/5/31 — 10:36 — page xvii — #17

Illustrations xvii

4.6 Example of IBP’s prediction for MBL case 118

4.7 IBP vs. other prediction methods: results (Ashley and Brüninghaus, 2009) 119

4.8 Excerpts from a theory learned by AGATHA with Mason case as cfs

(Chorley and Bench-Capon, 2005a) 121

5.1 A Carneades argument diagram (Gordon, 2008b,c; Gordon and

Walton, 2009; Walton and Gordon, 2009; see Ashley, 2012) 131

5.2 Classical deduction (Gordon, 2008b,c; Gordon and Walton, 2009;

Walton and Gordon, 2009) 133

5.3 Classical deduction cannot prove proposition and its opposite

(see Gordon, 2008b,c; Gordon and Walton, 2009; Walton and

Gordon, 2009) 133

5.4 Scheme for arguments from defeasible inference rules (see Gordon,

2008b,c; Gordon and Walton, 2009; Walton and Gordon, 2009) 134

5.5 Arguments (pro and con) with defeasible inference rules (Gordon,

2008b, c; Gordon and Walton, 2009; Walton and Gordon, 2009; see

Ashley, 2012) 136

5.6 When the rules run out (see Gordon, 2008b, c; Gordon and Walton,

2009; Walton and Gardon, 2009) 136

5.7 Carneades case-based argument (con) (Gordon, 2008b, c; Gordon

and Walton, 2009; Walton and Gordon, 2009; see Ashley, 2012) 138

5.8 Carneades case-based argument (pro) (Gordon, 2008b,c; Gordon and

Walton, 2009; Walton and Gordon, 2009; see Ashley, 2012) 139

5.9 Attacking arguments as per VAF (see Atkinson and Bench-Capon,

2007, p. 113) 140

5.10 VJAP domain model (Grabmair, 2016) 150

5.11 Values protected by trade secret law: interests of plaintiffs in property

and condentiality (Grabmair, 2016) 152

5.12 Values protected by trade secret law: interests of general public in

the usability of publicly available information and in fair competition

(Grabmair, 2016) 153

5.13 VJAP Program output for Dynamics Case (excerpts) (Grabmair, 2016,

pp. 59–60) 155

5.14 Statement and argument structure for reasoning about a Restatement

issue with trade-offs in VJAP (Grabmair, 2016, p. 51) 157

5.15 DLF partial rule tree for vaccine decisions, showing three causation

conditions of Althen (see Walker et al., 2011) 162

5.16 DLF extracted reasoning chains (excerpts) (see Walker et al., 2011) 163

6.1 Sample ontology for contract formation 172

6.2 Excerpts from e-Court ontology showing expansion of “reasoning

object,” “agent,” and “juridical role.” Links are is-a unless otherwise

noted (see Breuker and Hoekstra, 2004; Breuker et al., 2004; Van

Engers et al., 2008) 175

of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781316761380

Downloaded from https://www.cambridge.org/core. University of Florida, on 03 Nov 2017 at 09:07:18, subject to the Cambridge Core terms

“FM” — 2017/5/31 — 10:36 — page xviii — #18

xviii Illustrations

6.3 Ontology in the DALOS system (excerpts) (see Francesconi et al., 2010) 179

6.4 Model of legal argument with hypotheticals (Ashley, 2009b) 188

6.5 Case frame for Property-Interests-in-Quarry Microworld (P = Plaintiff,

D = Defendant) 191

6.6 Factor frame for Property-Interests-in-Quarry Microworld

(P = Plaintiff, D = Defendant) 192

6.7 ILC frame for Property-Interests-in-Quarry Microworld (P = Plaintiff,

D = Defendant) 196

6.8 Test frame for Property-Interests-in-Quarry Microworld (P = Plaintiff,

D = Defendant) 197

6.9 Policy/Value frame for Property-Interests-in-Quarry Microworld

(P = Plaintiff, D = Defendant) 197

6.10 Judge McCarthy’s Test in Popov v. Hayashi 201

7.1 Three-dimensional vector space model 217

7.2 BN for the family-away problem (see Charniak, 1991, p. 52) 219

7.3 Inference network retrieval model (see Turtle, 1995, p. 33) 220

7.4 Architecture of SPIRE (Daniels and Rissland, 1997a) 228

7.5 Retrieval for “videocassette” (see Rose and Belew, 1991) 230

8.1 History Project system architecture (see Al-Kofahi et al., 2001) 249

8.2 SVM examples. Black dots are positive instances, gray dots are

negative, and the white dot is “unknown” (see Noble, 2006, p. 1566) 252

8.3 Statutory analysis tool (Savelka et al., 2015) 256

9.1 Sample input/output of xmLegesExtractor (Francesconi, 2009, p. 66) 263

9.2 Multiclass SVM hyperplane example (see Francesconi and Passerini,

2007, p. 12) 264

9.3 Sample statutory sentences annotated for logical parts: antecedents

< A >, consequents < C >, and topics < T > (Bach et al., 2013) 269

9.4 Term suggestions for annotation templates (see Yoshida et al., 2013) 274

9.5 Partitioning statutory provision into subtree chunks 278

9.6 Decision tree for classifying statutory text units as relevant (rel.) or

irrelevant (irrel.) to the School of Public Health study 279

9.7 The box plot summarizes the results of all the experiments on each

of the tasks for Florida (FL). Each box plot describes performance in

terms of an F1-measure within a single experiment. The tasks included

identifying: AA (acting agent), PR (prescription), AC (action), GL

(goal), PP (purpose), ET (emergency type), RA (receiving agent), CN

(condition), TF (time frame) (Savelka and Ashley, 2015) 282

10.1 Excerpt of argument tree structure extracted automatically from a

case (see Moens et al., 2007, Fig. 5) 292

10.2 Overview of SMILE and IBP (see Ashley and Brüninghaus, 2009) 295

10.3 Schematic of LUIMA-Annotate (Grabmair et al., 2015) 301

10.4 LUIMA annotation environment 308

of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781316761380

Downloaded from https://www.cambridge.org/core. University of Florida, on 03 Nov 2017 at 09:07:18, subject to the Cambridge Core terms

“FM” — 2017/5/31 — 10:36 — page xix — #19

Illustrations xix

10.5 Annotation with GATE Teamware of factors and components in a

trade secret case (Wyner and Peters, 2010) 309

10.6 WebAnno annotation of trade secret factors in the Mason case

(see Yimam et al., 2013) 310

11.1 LUIMA pipeline architecture (Grabmair et al., 2015) 317

11.2 LUIMA-Search: sample query (top) and sentence entry in Lucene

database index (bottom) 319

11.3 AP of LUIMA Versions v. Baseline for Eleven Queries and MAP

(Grabmair et al., 2015) 325

11.4 NDCG of LUIMA Versions v. Baseline for Eleven Queries and Average

NDCG (Grabmair et al., 2015) 326

11.5 Queries for cases with propositions playing particular argument

roles (from the V/IP Domain). Bold-faced terms represent existing

sentence-level argument-role types. Italicized terms represent legal

rule requirements 330

11.6 Queries for legal factors and argument roles (from the trade

secret domain). Bold-faced terms represent existing sentence-level

argument-role types plus additional legal factor and value-related types

(see Section 11.5.3). Italicized terms represent legal rule requirements.

Underlined italicized terms represent legal policies or values 331

11.7 New argument-role sentence-level types 333

11.8 Argument mining for the Mason case opinion. Annotations

(with WebAnno) are: trade secret misappropriation legal fac-

tors, core LUIMA sentence types, proposition/premise or

proposition/conclusion, and ARGUMENT SCHEMES 337

11.9 Query input data scheme. Nodes represent successive levels of a

DLF-style rule tree and reasoning chain. Here user seeks cases with

evidence and an evidence-based nding that “MMR vaccine can

cause intractable seizure disorder and death,” in connection with a

legal ruling on the “medical theory causally connects” requirement

of the Althen rule on causation-in-fact 341

11.10 General (left) and domain-specic (right) statutory types 344

11.11 Statutory conceptual queries from the Public Health Emergency domain 345

11.12 Texas LENA Statutory Network (Ashley et al., 2014) 348

12.1 Two templates for targeted legal hypotheses 357

12.2 Sample legal hypotheses CCLAs should target (from the trade secret

misappropriation domain) 358

12.3 Sample legal hypotheses CCLAs should target (from the V/IP

domain). Issues in italics. 358

12.4 RST-Tree for Mason excerpt showing some attribution information 371

12.5 Annotation environment for rst-year law students 376

of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781316761380

Downloaded from https://www.cambridge.org/core. University of Florida, on 03 Nov 2017 at 09:07:18, subject to the Cambridge Core terms

剩余450页未读，继续阅读

WindStand

粉丝: 35
资源: 367

法律实践变革：人工智能驱动的法律分析与预测

China's Rise in ArtificialIntelligence PDF 英文版

Paradigms-of-Artificial-Intelligence-Programming.pdf

Artificial Intelligence with Python 英文原版PDF

Big Data and Visual Analytics-Springer(2017).pdf

Joshi -- Artificial Intelligence with Python -- 2017 -- code.7z

Artificial-Intelligence-A-Modern-Approach-3rd-Edition (1).pdf )

introduction.to.artificial.intelligence.-.10.-.learning.from.oberservations(ppt).pdf

The Elements of Artificial Intelligence Using Lisp - Steven L. Tanimoto.pdf

Artificial-Intelligence-with-Python.pdf

1920in01-study-of-development-blueprints-and-growth-drivers-of-artificial-intelligence-in-selected-places-20191023-c.pdf

最新资源