Artificial Intelligence in Law: Facts, Futures & Risks Michael Mills PRESENTATION TITLE
Why are we talking about AI? 2
3
What is AI? 4
Artificial intelligence is the study of how to make real computers act like the ones in the movies. Anon. 5
Artificial intelligence is the science and engineering of making intelligent machines. John McCarthy 6
AI diagnoses diseases 7
AI translates text 8
AI advises on the law 9
AI finds cats 10
AI wins games 11
12
It s all about inference IF some fact or pattern of facts is observed with some degree of accuracy and certainty THEN we can say some other things are X or Y true / false permitted / prohibited prudent / risky relevant / not about A / about B with some degree of confidence AND we can explain why with some degree of clarity 13
Inference methods Logical inference human experts, heuristics, explicit representation in rules & algorithms, deterministic, transparent Statistical inference big data, machine learning (unsupervised, supervised, reinforcement), probabilistic, opaque Hybrid combine logical + statistical 14
Logical inference IF Company State of Organization = Delaware AND Entity Type = Corporation, THEN Applicable Law = Delaware Corporation Law IF Company Headquarters = California, THEN Applicable Law = California Corporations Code IF Applicable Law = DE-CL and Board of Directors < 6, THEN E&O Policy Risk Level = Medium IF Applicable Law = CA-CC, THEN E&O Policy Risk Level = High IF No Company Subsidiary Organized in Kansas, THEN Kansas Tax Law = N/A 15
Statistical: unsupervised learning Find patterns without training 16
Statistical: supervised learning Train with labeled data 17
Statistical: reinforcement learning Self-train with rewards over time 18
Why now? Compute Power Big Data The Cloud AI as a Service 19
AI in Law 20
Watson could pass a multistate bar exam without a second thought. Robert Weber, GC IBM, 2015 21
Legal AI 22
Fastcase Bad Law Bot 23
Casetext CARA 24
Ravel judge analytics 25
Lex Machina 26
LawGeex contract analysis 27
Knowledge management 28
Advice from the B Schools 29
Expertise as a product 30
Expertise as a product line 31
Self- service guidance for clients 32
Technology & Access to Justice 33
34
The Future 35
The hype cycle 36
37
38
If a typical person can do a mental task with less than one second of thought, we can probably automate it using AI either now or in the near future. Andrew Ng, 2016 39
What to do next? Understand the tools Don t think about AI find problems to be solved Analyze ROI though I is uncertain & R speculative Start with the simple & obvious Be prepared to fail fail fast, move on 40
41
AI computing power 2007 2015 100,000,000,000 90,000,000,000 80,000,000,000 70,000,000,000 60,000,000,000 50,000,000,000 40,000,000,000 30,000,000,000 20,000,000,000 10,000,000,000 0 CPU 2007 GPU 2008 CPU Cloud 2011 GPU Cloud 2015 42
43
We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten. 44
AI & Risk Management 45
Questions What methods do you use now to evaluate & assure the quality of human processes & work product? How will you evaluate & assure the quality of AIaugmented processes & work product? Like humans, technology makes mistakes. More than is so with humans, with technology the mistakes are measurable, predictable, and transparent. What level of error is acceptable? Who decides? Client or firm? Case-by-case? Firm standard? What is the standard of care? What must/should you disclose to clients about the firm s use of AI? 46
More questions What must/should you disclose to the firm s professional liability insurers? When a technology makes mistakes that fall below the standard of care, who is liable? Must the firm insure specifically against this sort of liability? Does disclosure to and consent of the client suffice? Can a technology provider be deemed engaged in the practice of law? How will you account for the costs & benefits of AIimproved processes? What impact will they have on pricing and billing models? What impact will they have on performance measurement and compensation models? 47