Machine learning and the challenges of digital transformation in the law
0 likes329 views
The presentation by Sebastian Ko addresses the integration of AI in legal services, highlighting the challenges and opportunities present during digital transformation in the law sector. Key themes include the potential benefits of AI for cost-efficiency and risk management, the importance of data quality and volume for effective machine learning, and the implications of AI adoption on legal professionals and clients. Despite advancements, barriers such as regulatory constraints, risk-averse cultures, and the need for informed tool selection continue to hinder the adoption of AI technologies in the legal field.
Machine learning and the challenges of digital transformation in the law
- 1. Machine Learning and the Challenges of Digital Transformation in the Law Presentation by Sebastian Ko Webinar for CFRED, CUHK Faculty of Law May 29, 2020
- 2. Speaker Sebastian Ko, lawtech entrepreneur • Led Asian eDiscovery review business for global ALSP (PE-funded, ex NASDAQ), and led international corporate projects. • Legal experiences in private practice (financial services regulation, dispute resolution) and as in-house regional counsel. • Past research includes probabilistic reasoning in law (joint Science & Law program). • Chairman of the InnoTech Law Hub (WP), Law Society of Hong Kong. • Member of the MIT Technology Review Global Panel. • Int. Assoc. for Contract and Commercial Management, Council Member (Asia) • Qualified as Hong Kong solicitor and U.S. attorney. hk.linkedin.com/in/sebko Sebastian Ko 2020 © All opinions are my own and in no way representative of any of my affiliations. 1
- 3. Framing questions for this presentation • How should legal services develop to benefit from A.I.? • How is A.I. applied in the law? • Industry & technology trends • Building blocks, drivers & barriers • Implications for: • Lawyers and law students • Clients and the public • Tech providers Sebastian Ko 2020 © 2
- 4. Global headlines Sebastian Ko 2020 © 3
- 5. Recent trends: Tech in business Supply • Availability of key technologies • Computational power (integration of 1000s or 1Ms of data points) • Self-learning machines (specific tasks) • Natural language processing (computational linguistics; unstructured data) • ↓ Costs + ↑Investments • Building technology solutions • Running tests and deployment of solutions • Business models (cloud, SaaS, PaaS etc.) Demand • Big Data “5 Vs” – data explosion • Volume, Velocity, Variety, Veracity, Value • Socialization of technology use: personal à professional • Sophisticated in-house IT expertise Sebastian Ko 2020 © 4
- 6. Common use cases: Are they ready for law? • Speech recognition and machine transcription • Personalization (AUI; recommendation engines) • Email SPAM filters • Fraud detection (banks) Why don’t we ask Siri (Apple) or Alexa (Amazon) for legal research help, yet? Sebastian Ko 2020 © 5
- 7. A.I. adoption by firms and in-house teams Charts by Kira Systems, April 2020 Thomson Reuters In-house Technology Survey 2017 2019 Sebastian Ko 2020 © 6
- 8. Where is A.I. applied today? But why? • Why are uptakes in these areas? • Why are there many products? Or not enough? • Where next? LawGeex, 2019 Sebastian Ko 2020 © 7
- 9. Definitions, Examples and Challenges Sebastian Ko 2020 © 8
- 10. Key concepts and values • Automation: cost-efficiency and risk management • A.I. = ensemble of tech. that makes assumptions, tests and learns autonomously • ↑↑ Value of data integration • Data analytics: Gaining insights and patterns • Descriptive analytics: Tells you what happened in the past (e.g. summary stats) • Diagnostic analytics: Helps you understand why something happened in the past • Predictive analytics: Predicts what is most likely to happen in the future (based on historical data) • Prescriptive analytics: Recommends actions you can take to affect those outcomes Automation Analytics & A.I. Supervised & Unsupervised Learning; Deep Learning Yates, 2019 Sebastian Ko 2020 © 9
- 11. Example: Third-party funders of claims • Predictive analytics; a decision- making aid • Better information for p values • Additional factors: Has the judge eaten lunch yet? (PNAS, 2011) • Litigation funding is US$3 billion asset class (New Yorker, 2016) M. Victor, 2015 (litigation example) Sebastian Ko 2020 © 10
- 12. Examples: Rule-based, supervised and unsupervised learning models Classifying clauses (Kira)Cluster wheel of concepts (Brainspace) Sebastian Ko 2020 © 11
- 13. A.I. areas of applications and adoption by industries McKinsey, 2019 Sebastian Ko 2020 © 12
- 14. Speed of adoption: Technical challenges Collection Source: Human vs. machine; Real vs. synthetic Unstructured data Data sensitivity (incl. transfer restrictions) Processing Data volume, quality Cleaning & Unitization Standardization, normalization, protocols Training / Learning Selecting the right algorithm / model Unsupervised vs. Supervised; Reinforcement; Deep Analysis Error acceptance / tolerance Validity of insights from model's results Adapting to new cases Present & Audit Visual design, easy-to- understand Interpretable Explainable (Non-exhaustive lists) Sebastian Ko 2020 © 13
- 15. Massive volume of quality data needed General • Great models are useless without data • “Massive” = 10k to 100k cases… and shrinking • A.I. smarter with fewer lessons needed • Real vs. synthetic data • “Quality” of data • Words vs. numbers • Language(s) • Context – industry-specific etc. • Data preparation • File types • Text cf. MS Word vs. social media • Filters (time, custodian etc.) depending on case context Legal Sector-Specific • Sheer data volume held / processed • Law firms vs. FAANG • LegalTech Corporations • Ability to share and pool data • Client ownership of data • Data sensitivity • Cross-border data transfer Extract and abstract relevant data points Sebastian Ko 2020 © 14
- 16. Legal, regulatory, and ethical constraints • Sensitive data and cross-border data transfers • US HIPAA • CN Cybersecurity Law • SG Banking Secrecy Provisions • Data-sharing models • Attorney-client privilege; practice- ownership; marketing rules • SaaS & other business models • Bias in A.I. decision-making and profiling processes • EU GDPR, Art. 22 • US Algorithmic Accountability Bill How is the client-lawyer-vendor relationship impacted? Sebastian Ko 2020 © 15
- 17. Finding the right models • How to represent human-created text in computer models to find relevancy and other criteria? • Existence of keywords • Co-occurrences with e.g. metadata and other artifacts • Linguistic distances • Model achieves an acceptable level of …? • Accuracy, reliability / consistency • Statistical standards • Legal standards (see Triumph v Primus (2019), E&W) Sebastian Ko 2020 © 16
- 18. Legal informatics issues • Classification into computer-friendly structures • Manual ways: headnote searches, and WestLaw KeyCite • Natural Language Processing: Better classification of legal text and queries • Research on ontologies, argumentation theory and computational logic How to factor into relevancy? • Words with both legal and common meanings (e.g. director – company vs. film) • Precedential value (common law) • Open-ended terms that change meanings over time (common law) • Intention: a case favouring your side • No. of citations (preceded Stanford page- ranking methods, used by Google) Relativity, search setup and search hits report Sebastian Ko 2020 © 17
- 19. Model validity and blackbox decisions • Q: "What is covering the windows?” • A.I. “There are curtains covering the windows.” • Bed > Bedroom > Bedrooms have curtains covering windows (training set) • Does the machine understand the substantive meaning? • Is it learning the right lessons? Virginia Tech A.I. eye-tracking research (2016) Sebastian Ko 2020 © 18
- 20. Whitebox and statistical interpretations Predicting apartment prices 100s to 1000s of trees Everlaw, Predictive Coding (SVM) Sebastian Ko 2020 © 19
- 21. Speed of adoption: Non-technical factors Adoption Barriers Risk-averse culture Laws & regulations Funding, business model Informed tool selection Steep learning curve Expert support Solutions to be adapted to legal context • Understand which processes need automation • Infrastructure / legacy systems • Change management practices / implementation • Skilled resources – service providers and consultants Sebastian Ko 2020 © 20
- 22. Implications on Legal Sector and Justice Sebastian Ko 2020 © 21
- 23. Buyers and users • Sales pitch vs. actual adoption, and who cares? • Luxury vs. necessity • Lawyers vs. clients • Lawyers vs. other consultants • Reshaping the legal world for mechanical savants • Legal process outsourcing (huge growth after GFC, 2008) • “Legal operations” as a discipline • Fundamental concepts about the legal profession are questioned New ways to deliver legal services? Sebastian Ko 2020 © 22
- 24. Builders, purveyors, beneficiaries, other stakeholders • Private sector-led innovations • Export by e.g. US and UK (legal cultures) • How to get the “tool smiths” to make better tools? • Silo developments • Users are often not the beneficiaries • Tech providers; increasingly key players in the legal sector • LegalTech as niche sector • General tech providers weighing in • Consumer protection • “Equality of arms” • Lob-sided growth in commercial practices vs. other areas • Spurred by tech influence • Assessing metrics (ROIs etc.) vs. more abstract notions (justice?) • Innovation policies • Regulatory sandbox • Privatized justice • Stagnation of jurisprudence (in certain pockets?) Sebastian Ko 2020 © 23
- 25. The tech and legaltech mindsets Fundamental Rights / Constitutional Substantive Law Procedural & Evidentiary Industry-Specific Regulations Professional Ethics General (e.g. Privacy) / Transactional Rights & Justice Obligations Efficiency Risk Manage Revenue ↑ & Productivity ↑ Cost ↓ Security ↑ Consistency ↑ & (Human) Error ↓ Predictability ↑ & Auditability ↑ Integration ↑ & Connectivity ↑ ? ? ? ? Sebastian Ko 2020 © 24
- 26. Conclusion Massive development gaps End users ≠ end payers Highly regulated, risk- averse users / buyers Niche sector, investments Legal & tech experts, knowledge sharing Overzealous vendors, Product-market fit lacking • Raising tech baseline, supporting SME firms’ tech adoption • Public funding • Guidance and education • Develop building blocks of legal A.I. “public goods” • Open legal data • Open-source algorithmic law models • A.I. transparency and interpretability • Consumer protection • Cross-disciplinary talent Sebastian Ko 2020 © 25