Your premier source for legal solutions, including Thomson Reuters Westlaw and law books.
Practice Innovations — Managing in a changing legal environment
Gray Rule
July 2016 | VOLUME 17, NUMBER 3
Gray Rule
The Changing Focus of AI:  Classic AI, Artificial Neural Networks, Biological Neural Networks
»For easy printing – view as PDF


»The Changing Focus of AI: Classic AI, Artificial Neural Networks, Biological Neural Networks
»Thinking Outside the Stacks: New Career Paths for Information Professionals
»Unleash Your Innovators
»Outside Counsel Guidelines: Navigating the Changing Landscape of Client Demands
»Sustainable KM
»Managing the CRM Abyss: Winning the Battle for Data Quality
»Back to Contents

» About Practice Innovations
» Editorial Board
» Past Issues
» Reader Feedback

The Changing Focus of AI:  Classic AI, Artificial Neural Networks, Biological Neural NetworksBy Elaine M. Egan, Head of Research & Information Services-Americas, Shearman & Sterling LLP, New York, NY and Kathy Skinner, Director of Research Services, Morrison & Foerster LLP, San Francisco, CA
Law and legal reasoning appear to be a natural fit for artificial intelligence (AI). Legal analysis relies on interpreting data and applying this data to higher-level concepts, but until now what we termed AI was essentially an advanced programming system. The new focus of AI distinguishes the classic approach to AI with one yielding a fundamental insight into the nature of intelligence. Without a basic understanding of how AI works, how can we envision, implement, and value its role in the legal profession?

The way lawyers think, the way they do business, and the way they interact with clients are central to innovation in the legal community. Competitive pressures in the legal marketplace have challenged law firms to adopt leading edge technology, and in many cases firms have evaluated AI platforms with a serious commitment to investing in AI.

The ability to distinguish the evolution of AI beginning with IBM’s Deep Blue defeating chess legend Garry Kasparov in 1996, Watson’s Jeopardy win in 2011, and Google’s Deep Mind AlphaGo program learning how to play and then win against the world class Go player Lee Se-dol, creates confusion about what AI systems can and cannot do. But these recent game advances are some of the most significant transformations in building smart machines from the Classic AI “expert system” programs to “deep learning.”

We use the term AI to refer to a multitude of programs and computer systems, but on a simple level AI essentially refers to three different approaches in building smart machines: Classic AI, Artificial Neural Networks (ANNs), and Biological Neural Networks (BNNs).

Classic AI is designed to solve problems that a human could perform routinely, such as recognizing images and text. A challenge of Classic AI is that smart machines require access to large amounts of knowledge, which led to the introduction of encoded rules and “expert systems” simulating the decision-making ability of a human domain expert through a series of questions. An example of this would be a computer’s ability to solve problems like diagnosing a medical condition. This approach to AI is geared towards a specified problem and desired outcome.

As limitations to Classic AI became evident, the early theoretical research into Neural Networks (NNs) focused on the idea that certain properties of biological neurons can be extracted and applied to mimic the human brain. ANNs attempt to create this learning system by constructing a program to respond to a problem and then evaluating feedback on how it responds. A computer can optimize response by repeating the same problem thousands of times and adjusting its response according to the feedback it receives.

ANNs are a subset of broader AI category called Machine Learning (ML). ML looks at complete sets of concepts, and extracts statistics and classifies the results. ML is the procedure by which knowledge is acquired through algorithms that “learn.”

With access to faster and faster computers along with vast amounts of data, ANNs have evolved from ML into Deep Learning. This is termed Deep Learning because the structure of ANNs 40 years ago was only two layers deep and ANNs today can be 10 layers or more. Deep Learning ANNs automatically extract characteristics rather than relying on a manual extraction. Generally, Deep Learning applications involve any combination of the following three tasks: speech, image recognition and reading, or generating written language. For example, Microsoft’s Skype recently demoed a version of real-time translation wherein speakers will hear and see both languages simultaneously.

The ultimate AI is the BNNs approach. By studying how the brain works those properties of the brain that are required for intelligence can be determined. The brain contains large networks of interlinked neurons and represents information using sparse distributed representations (SDRs). A good example of the BNNs approach is Numenta’s Hierarchical Temporal Memory (HTM). This is a machine intelligence technology based on learning algorithms. The HTM system learns the structure of streaming data, makes predictions, detects inconsistencies, and continuously learns from unlabeled data. The BNNs system does not need to know what it is looking for and continuously learns as data changes.

Taking the mystery out of even the simplest approach to AI may prepare a lawyer to envision those approaches and branches of AI that are most useful. Identifying those practical tasks that can be automated and are consistently cost effective frees the lawyer to work on the most complex elements of legal practice. Emotional intelligence, client counseling, fact finding, legal writing, court appearances, and negotiation cannot be replicated by computers.

AI models have been embedded in the legal workplace for several years, for the purposes of e-discovery, legal research, compliance, and contract analysis. Recent entrants include outcome prediction tools that mine large data sets to forecast outcomes, such as Lex Machina and LexPredict. But, the most intriguing focus of AI has been in the realm of legal research. Lexis® and Westlaw® users have been exposed to natural language processing techniques for more than a decade, and the added value of visualization features like those in Fastcase and Ravel Law have improved the analysis of result sets.

What has surfaced most often in news and industry meetings is IBM’s Watson and its 2015 upgrade, incorporating three Deep Learning features—translation, speech-to-text and text-to-speech—that will no doubt strengthen its commercial appeal while still building on its Q & A model. Watson is built on natural language processing (NLP) and the way humans understand language.

The world’s largest law firm, Dentons, launched NextLaw Labs in 2015. NextLaw operates independently as a subsidiary of the firm with the objective of innovating through disruption. The ability to accelerate and contribute to the development of legal technology is the driving force behind their commitment to ROSS Intelligence. The founders of ROSS Intelligence built their system on Watson’s NLP. Rather than searching for documents by keywords, queries are formulated in plain English, for example: How does a creditor prepare a proof of claim? The result set is delivered with sections of the Bankruptcy Code along with a confidence rating. The corpus of ROSS data is bankruptcy law, with the intention of adding case law and other third-party materials. Baker & Hostetler has recently emerged as the first law firm to publicly announce that it has licensed the AI product developed by ROSS Intelligence for bankruptcy matters.

In addition to AI technologies like ROSS which facilitate efficiencies in producing legal work, AI and machine learning technologies are also being used by in-house counsel to analyze and benchmark work done by outside counsel. For example, a company called Brightflag provides an AI platform that Yahoo and other companies use to compare legal spend across firms and geographies, and to automatically apply the company’s billing guidelines. Not only does this tool help companies to reduce legal spend, it also provides insight into how to manage matters, including deciding which matters to handle in-house and which to send to outside counsel.

Clearly AI is a growing technology that hasn’t reached critical mass, but for lawyers, AI represents a real opportunity to produce better and better legal work. The AI future is unclear, however, and some of our greatest minds like Stephen Hawking and Bill Gates have expressed deep concerns about AI’s long-term potential for damage to humankind. Perhaps in addition to being beneficiaries of AI to enable more efficient ways of working, lawyers can also contribute to the future AI landscape by helping to define and enforce a regulatory framework which guards against the dangers that Hawking, Gates, and others warn about.

Back to Contents