Building Sustainable Free Legal Advisory Systems: Experiences from the History of AI & Law

tags

DataLex

Notes

The desired result, we argue, is the development of integrated legal decision-support systems, not ‘expert systems’ or ‘robot lawyers’.

NOTER_PAGE: (2 . 0.49925149700598803)

previous iteration of computing practices that were going to ‘make everything different’ such as Japan’s ‘Fifth Generation’ project.

NOTER_PAGE: (4 . 0.20059880239520958)

expert system boom continued until the late 90s,2 by which time enthusiasm for the Internet attracted people’s interest and greed (the first ‘.com’ boom) in substitution.

NOTER_PAGE: (4 . 0.2971556886227545)

such software and applications must be maintainable from their internal resources

NOTER_PAGE: (5 . 0.6459580838323353)

focusing solely on legal expert systems, with development of the LES shell for procedural (decision network) inferencing

NOTER_PAGE: (6 . 0.13847305389221556)

integration of inferencing (primarily rule-­‐based and to some extent case-­‐based expert systems), hypertext and text retrieval, with some document generation capacity as well.

NOTER_PAGE: (6 . 0.4191616766467066)

Support stopped in 1995, when a commercial publisher terminated DataLex’s licence to include case law content.

NOTER_PAGE: (6 . 0.6010479041916168)

automated generation of large-­‐scale automated hypertext mark-­‐up of legal documents.

NOTER_PAGE: (7 . 0.125)

integration of inferencing (knowledge-­‐bases and dialogues) with hypertext and text retrieval was further developed, and methods of ‘collaborative inferencing’ (distributed, multi-­‐author knowledge-­‐bases) were pioneered

NOTER_PAGE: (7 . 0.20808383233532934)

heuristics to improve text retrieval and hypertext mark-­‐up

NOTER_PAGE: (7 . 0.30538922155688625)

heuristics for automated construction of an international case and journal citator (LawCite)

NOTER_PAGE: (7 . 0.3196107784431138)

Law is not ‘just another problem domain’

NOTER_PAGE: (7 . 0.7956586826347306)

law is an unusual and difficult problem domain, because legal expert systems do not usually involve modelling either (i) heuristics of how experts make decisions, or (ii) causal models of physical systems.

NOTER_PAGE: (8 . 0.07335329341317366)

‘legal expert systems’ always included what we call ‘formal’ advisory systems, based primarily on representing formal legal doctrine in rules (according to expert interpretations), with lawyer’s heuristics playing only a minor role.

NOTER_PAGE: (8 . 0.4528443113772455)

both these options are of little use by themselves.

NOTER_PAGE: (8 . 0.531437125748503)

is the system which is being built intended to provide justifications for its answer based on underlying legal sources

NOTER_PAGE: (9 . 0.17065868263473055)

give the ‘correct’ provision of useful answers without legal reasons

NOTER_PAGE: (9 . 0.2095808383233533)

the intention is to build systems which can justify their answers/recommendations

NOTER_PAGE: (9 . 0.28742514970059885)

for many legal problems, there are no sufficiently comprehensive ‘training sets’

NOTER_PAGE: (9 . 0.3510479041916168)

Despite nearly 40 years of research into case-­‐based legal reasoning, there is probably not much role for AI representations of case law

NOTER_PAGE: (9 . 0.5187125748502994)

Predicting outcomes of litigation using correlations with factors that have no direct relationship to formal reasoning is increasingly possible using ‘big data’ analytics (what used to be called ‘jurimetrics’), but is not likely to have any sustainable use by free legal advice services.

NOTER_PAGE: (9 . 0.7140718562874252)

attempt to alert users of inferencing systems when case law reasoning is required

NOTER_PAGE: (9 . 0.7919161676646707)

the best starting points for AI-­‐based legal systems, at least for free access legal advice providers of limited means, are still the representation of legislation, or procedural problems which depend to a large extent on legislation.

NOTER_PAGE: (10 . 0.1220059880239521)

a very significant proportion of legal claims that are in fact resolvable by statutory provisions do not have any relevant ‘corpus of legal cases’

NOTER_PAGE: (10 . 0.3023952095808383)

discussions of isomorphism45, the need for declarative representations and similar matters

NOTER_PAGE: (10 . 0.531437125748503)

ensure that users always take into account all relevant statutory provisions

NOTER_PAGE: (10 . 0.5823353293413174)

Users organisations should maintain their own knowledge-­‐bases

NOTER_PAGE: (10 . 0.625748502994012)

‘knowledge acquisition bottleneck’ has always been the largest practical problem in the construction of legal expert/advisory systems.

NOTER_PAGE: (10 . 0.7612275449101796)

construction of legal expert systems is done directly by a lawyer without the use of a knowledge engineer intermediary.

NOTER_PAGE: (11 . 0.07559880239520958)

considerable design constraints in both software and knowledge-­‐ bases

NOTER_PAGE: (11 . 0.157185628742515)

Use declarative knowledge representations where possible

NOTER_PAGE: (11 . 0.19535928143712575)

without any particular order of representation being required, or any order of processing specified.

NOTER_PAGE: (11 . 0.23577844311377247)

isomorphism is easier to achieve; knowledge-­‐base development is faster; knowledge representation is more transparent; and less maintenance of knowledge-­‐bases is required.

NOTER_PAGE: (11 . 0.4446107784431138)

limitations of the completeness of such representations because of such issues as the open texture of legislation.

NOTER_PAGE: (11 . 0.5553892215568863)

Isomorphic representations are desirable

NOTER_PAGE: (11 . 0.6002994011976048)

well defined correspondence between source documents and the representation of the information they contain

NOTER_PAGE: (11 . 0.6205089820359282)

 source change can be related to a defined fragment of the knowledge base

NOTER_PAGE: (11 . 0.6699101796407186)

Quasi-­‐natural-­‐language knowledge-­‐bases avoid repetitive coding

NOTER_PAGE: (12 . 0.12425149700598803)

‘textual baggage', detracting from isomorphism, is eliminated

NOTER_PAGE: (12 . 0.30164670658682635)

removes ambiguities better than natural English

NOTER_PAGE: (12 . 0.3068862275449102)

transparency during use is increased

NOTER_PAGE: (12 . 0.3196107784431138)

transparency for validation is increased

NOTER_PAGE: (12 . 0.3203592814371258)

increased explanatory power

NOTER_PAGE: (12 . 0.3383233532934132)

relative ease with which domain experts can check rules

NOTER_PAGE: (12 . 0.35404191616766467)

With a quasi-natural language knowledge representation, coupled with automated hypertext links from its terms, there need only be one representation in order to achieve links to the statutory text, access to decision aids of different types, and generation of explanations

NOTER_PAGE: (12 . 0.4603293413173653)

Propositional representation is enough for most tasks

NOTER_PAGE: (12 . 0.5411676646706587)

predicate calculus inferencing system

NOTER_PAGE: (12 . 0.561377245508982)

increased the complexity of the dialogues with the user

NOTER_PAGE: (12 . 0.625)

not as easy for developers to understand what inferencing steps would be taken

NOTER_PAGE: (12 . 0.6422155688622755)

knowledge representation could be regarded as ‘deceptively simple’.

NOTER_PAGE: (12 . 0.7065868263473054)

YSH (or WYSH) knowledge-­‐base was somewhat less isomorphic and less resembling natural (legislative) language, but had the advantage that the steps that would be involved in drawing inferences were more apparent

NOTER_PAGE: (12 . 0.7365269461077845)

Ashley points out the advantages of predicate logic, but his example of a Prolog representation of the British Nationality Act, while very informative, is also able to be represented in propositional logic, and it is not clear under what conditions he considers predicate logic essential.

NOTER_PAGE: (13 . 0.13697604790419163)

Semi-­‐expert systems and users collaborate

NOTER_PAGE: (13 . 0.21482035928143714)

The aim in building legal expert systems is not to build a ‘robot lawyer’, which simply extracts unproblematic facts from a user and then comes to a conclusion.

NOTER_PAGE: (13 . 0.23353293413173654)

some degree of interpretation of the questions asked

NOTER_PAGE: (13 . 0.27095808383233533)

minimal level of interpretative skills

NOTER_PAGE: (13 . 0.28742514970059885)

Inferencing is not enough for decision support

NOTER_PAGE: (13 . 0.6407185628742516)

interpretation issues cannot be eliminated

NOTER_PAGE: (13 . 0.6751497005988024)

inferencing systems cannot be ‘closed’

NOTER_PAGE: (13 . 0.690868263473054)

inferencing systems must be as open as possible to all relevant legal resources, primary and secondary.

NOTER_PAGE: (14 . 0.07634730538922156)

Legal expertise can and should be captured by multiple means

NOTER_PAGE: (14 . 0.3293413173652695)

what counts as a useful level of legal expertise is relative

NOTER_PAGE: (15 . 0.23877245508982037)

should not be ‘closed’: it must be integrated with text retrieval, hypertext and other tools which allow and assist the user to obtain access to whatever source materials are necessary

NOTER_PAGE: (15 . 0.375)

reasonably high level of isomorphism

NOTER_PAGE: (15 . 0.5194610778443114)

reasonably close to natural language

NOTER_PAGE: (15 . 0.5538922155688623)

Desirable improvements for sustainable legal advisory systems

NOTER_PAGE: (16 . 0.07559880239520958)

More supportive editing environments for knowledge-­‐bases

NOTER_PAGE: (16 . 0.312125748502994)

visual programming, drag-­‐and-­‐drop programming and sophisticated integrated development environments.

NOTER_PAGE: (16 . 0.4782934131736527)

NLP layer on top of existing systems.

NOTER_PAGE: (16 . 0.5606287425149701)

for a lawyer to be able to go immediately to some ‘first draft’ of part of a knowledge-­‐base, with at least some of the rule syntax correct, would speed knowledge-­‐base construction and reduce the knowledge acquisition bottleneck.

NOTER_PAGE: (16 . 0.7043413173652695)

vastly more legal sources available for free

NOTER_PAGE: (17 . 0.21032934131736528)

Platforms to assist development and maintenance of free advisory systems

NOTER_PAGE: (18 . 0.13922155688622756)

Data-­‐driven and data-­‐oriented feedback tools

NOTER_PAGE: (18 . 0.3023952095808383)

Renewed attention to transparency and ethical operation of legal analytics

NOTER_PAGE: (18 . 0.5359281437125749)

Free legal advisory systems (or legal information institutes) are less likely to have a strong proprietary interest in the knowledge-­‐bases they employ than are commercial law firms, and are therefore more likely to be willing to open them to scrutiny.

NOTER_PAGE: (18 . 0.7020958083832336)

may add little to what available human expertise, textbooks and checklists can provide

NOTER_PAGE: (19 . 0.26646706586826346)

relatively rare applications of the technology that really do justify the costs involved in developing them

NOTER_PAGE: (19 . 0.3001497005988024)