Ben Gardner, data & information architect, Linklaters
Summary
Is AI a threat or an opportunity? It is both, to those focused on the routine it is a threat, to those focused on innovating/bespoke it is an opportunity, as pointed out by John O. McGinnis & Russell G. Pearce[1]. The advice therefore should be to understand these new technologies and explore what opportunities they create. This should not be limited to iterative improvements to the current process but more importantly to identifying opportunities for transformational change.
Introduction
This last year has seen Artificial Intelligence (AI) emerging as a very hot topic within the mainstream media (see Elon Musk, Stephen Hawkins, Bill Gates, etc.)[2]. Many of the articles have invoked the spectre of a new generation of ‘Robots/AI’ raising up to take skilled jobs, previously the preserve of the professions (see Susskind book)[3]. Within the legal industry a similar trend has been seen with the emergence of tools, start-ups and large law firm investment in AI. The impact of all this activity has been encapsulated by the replies to a question in the recent Altman Weil Survey which showed that a significant increase in the belief in the probability that timekeepers of all levels will be replaced by AI is held by senior partners at law firms[4]. But is this prediction realistic? If we are to comprehend the impact of AI then we need to look beyond the hyperbole of the moment and understand what AI is and more importantly what it is not. In this paper I will first define the term Artificial Intelligence and elucidate the technologies underlying current tools. Next I will explore how AI is currently being applied with respect to the law both within law firms and in the start up legal service industry. Finally I will address the question “AI a threat or an opportunity?”
What AI is and is not
The first thing to recognise is AI is a hype term and one which popular culture has ‘anthropomorphised’ and ‘demonised’. Our first thoughts when we think of AI are examples from science fiction, HAL 9000 (2001 Space Odyssey), Skynet (Terminator) or more recently Ava (Ex Machina). These are examples of Artificial Generalised Intelligences, a sentient conscious self-aware entity with the capacity to reason, learn, emote, plan, communicate, etc. But this is not the AI we are contemplating here. Instead we are looking at examples of Narrow or Weak AI. This is a non-sentient form of AI that simulates the performance of activities that we perceive as requiring intelligence. For example consider chess grandmaster Garry Kasparov vs IBM’s Deep Blue. Kasparov used his experience and intuition to select his moves while Deep Blue used a brute force approach to calculate all possible moves and then select the one with the best probabilistic outcome. The important point is that the way this problem is solved computationally is not the same as the way we solve the problem. The observed outcome might be the same but the approach can be very different.
Weak AI solutions are built by combining a range of techniques from different fields of computing. Most recently it has been advances in Natural Language Processing (enabling computers to derive meaning from unstructured data i.e. text or speech), Machine Learning (giving computers the capacity to learn without explicit programing) and Modelling (a formal description of a system that can be understood and utilised by a computer to make predictions) that have led to the resurgent application of AI. Broadly speaking Weak AI is defined by two approaches based on the type of models utilised, symbolic or statistical. Symbolic modelling was the dominant paradigm from 1950s through to mid-1980s and uses formal logic to construct a representation of an area of knowledge. The models are hand crafted and developed in a manner akin to a computer program. The most successful example of this approach are expert systems (see below) which drove the previous commercial interest in AI during the late 1980s. More recently statistical modelling techniques have evolved and these are the driving force behind the current generation of AI application. In this case large data sets are used, in conjunction with machine learning algorithms, to construct statistical models based on the patterns found within the data set being analysed. It is important to recognise that this approach often requires very large data sets to establish the models and breakthroughs in this space have been enabled via the scale of data available via the internet. Application of this approach to AI can be found throughout everyday life from spell checkers and spam filters through to self-driving cars and recommendation engines. Stephen Wolfram has described the essential difference between the two approaches as “[The statistical approach to] AI is about [simulating] what individual brains do, rather than [the symbolic approach] reproducing and automating … what the whole civilisation knows about”[5].
More recently we have seen the emergence of hybrid AI applications which use a combination of statistical and symbolic approaches to create the desired end user experience. Examples of hybrid AI include Smartphone Assistants (Siri, Cortana and Google Now) and solutions built with AI cloud services (IBM Watson, Microsoft Azure and Amazon). These next generation of AI tools use a combination of technologies and their development has been heavily tied to the evolution of the internet. It is likely over the next few years that major advances in AI will continue to be driven by our demands to manage and exploit information accessed via the internet.
Artificial Intelligence in Legal
In the previous section we considered AI from a broad technical perspective and compared the underlying approaches. In this section we will pivot our view and look at AI from a capability perspective and aim to understand how and where AI tools are being used within the legal profession. We will start by examining examples of AI in the areas of expert systems and text analytics. This will be followed by a review of the trends within the AI legal start up space and consider potential future developments.
Expert systems
As mentioned previously expert systems have been around commercially since the mid-1980’s and while they may not grab the headlines that the current generation of AI tools do they are still a powerful tool. In their simplest form an expert system is a collection of rules, ‘if this, then that’ statements, which can be used to guide a user through a complex decision making process. Within the legal industry Neota Logic[6] and VisiRule[7] are two established providers of expert systems. An example implementation might be a system which helps scale a compliance process or one that triages litigation claims. Consider the following scenario, a banker needs to ensure she is compliant with a regulatory framework. Rather than consulting a lawyer for each transaction they are prompted by the expert system to answer a series of questions that leads them to a pre-drafted opinion; yes proceed, no stop or seek further clarification[8]. This approach allows a compliance process to be applied at a scale that would not be achievable if each transaction required an actual lawyers input. The expert system helps sort the ‘wheat from the chaff’ by allowing the simple decisions to be answered in an efficient, cost effective fashion while flagging those which require expert consideration.
Text Analytics
Text analytics aims to allow computers to extract meaning from unstructured content. The field covers a wide range of techniques combining both statistical and symbolic approaches to AI. While a simple form of text analytics is used within established eDiscovery tools it is the commercialisation of advanced text analytics that has allowed the emergence of new products such as Ravn[9] and Kira[10]. These solution go beyond simple finding documents with matching keywords, and synonyms, and allow the identification of contextual meaning both of the sentence structure and of the entities discussed. Further through the use of machine learning algorithms these solutions can be trained to identify larger blocks of text, for example whole clauses within a contract, and the wider structure of a document. This allows them to not only analyse individual documents but to also cluster documents based on similarity across a corpus of documents. The application of these capabilities within a legal context can readily be seen in support of due diligence and document comparison. For example consider a due diligence exercise as associated with the purchase of a bundle of loans. A modern text analytics tool can allow the full corpus of contracts to be analysed rather than manually reviewing a subset. All contracts can be clustered, identifying common templates and flagging outliers/anomalies. Text identification can then be applied to each cluster to accurately extract entities from the unstructured text i.e. names, dates, addresses, interest rates, terms, borrower, break clause, etc. Not only can this analysis be performed across the whole corpus but it can be done in a fraction of the time it currently takes a group of lawyers and paralegals to do on just a subset of documents. Further it is important to recognise that because this analysis is performed in a consistent manner by a computer the output is in a standard format. This means the output can be further analysed to produce dashboards and business intelligence reports to visually describe the profile of the whole corpus of documents.
Legal Tech Start-ups
An examination of legal tech start-ups provides an interesting perspective into how AI technologies are being used to innovate new services. In general terms the start-ups can be split into two broad groupings. The first with established products that are centred around extracting, analysis and creating insights based on patent/court records, while the second group, still in pre-launch phase, appear focused on analysing/interpreting the law and regulations. The first group includes examples such as Lex Machina[11], Juristat[12], Ravel Law[13] and CLAWS[14]. These companies use text analytics to extract information from unstructured sources, patent and court records, to construct a persistent database that can be mined to provide insight, similar to classical business intelligence reporting. Juristat and Lex Machina have been primarily focused on intellectual property space, specifically providing insight into the performance of patent examiners, the state of IP within a market or the profile of a competitor’s patent portfolio. Ravel Law & CLAWS in contrast analyses court documents to provide similar insight into judges and cases.
The second group of start-ups are in a pre-launch phase includes Judicata[15], Ross[16] and industrial/academic partnerships such as the Governance, Regulatory & Compliance Technology Centre (GRCTC)[17]. Judicata and Ross are using AI tools to analyse areas of law, Californian employment law and Canadian bankruptcy law respectively. Based on the limited information available in the public domain it appears these two companies are attempting to build symbolic models describing these two areas of law. These models can then be used to build recommendation engines but focused on helping aid research of these areas of law. These solutions should be seen as smart legal research aids rather than AI lawyers, think competitors with tools such as LexisAdvance or WestlawNext. The other member of this grouping is the GRCTC who are developing symbolic representations of financial regulations. Similar to Judicata and Ross these models could be used to aid lawyers working with these regulations however the end goal in this case is to apply the models to drive efficiencies within the governance, regulatory and compliance processes within the financial sector through automation of monitoring/auditing.[18]
AI a threat or opportunity?
So far in this article we have looked to get an understanding of what AI is, the main approaches to AI and the emerging generation of tools/service being developed using AI technologies. In this section we will consider what the impact of these developments could be for lawyers.
One way to predict the impact is to look back in the past and consider what the effects of previous technological advances have been. For example a 10-attorney firm today can do way more with Westlaw, mobile phones, and email than a 20-attorney firm could do with a full library (and twice the support staff) 25 years ago. Does that mean that 10 attorneys at that firm were “replaced” by Westlaw/mobile phones/computers? Or have those resources been redeployed? The answer would be yes they were replaced if the quality of the brief remains constant and that demand for legal product is inelastic. Let’s consider these two aspects.
Firstly does the quality of a brief remain static as the tools for research improve? The answer is clearly no. A brief that was “good enough” in the past might be utterly insufficient ten years from now, if the opposing side can exploit your omissions or research mistakes. In fact the risk is that it will be a Red Queen race where both sides are using a series of better and better tools merely to keep up with each other instead of making comparative gains.
Secondly is the demand for legal product inelastic? Again the answer to this is no. Whenever a new technology is introduced it will eliminate some tasks but at the same time open up new opportunities. Just as Richard Susskind described how legal services evolve from Bespoke to Standardised to Systematised to Packaged to Commoditised[19], technology is an agent for change and drives this cycle. It is easy to think of established processes, towards the commoditised end of the cycle, to which AI tools could be applied to drive continuous improvement and deliver efficiencies, i.e. reduce the need for large numbers of lawyers/paralegals to perform manual document comparison/review tasks during due diligence. However to innovate we have to think differently and re-phrase the question. Take the example of due diligence, AI not only enables the review of all documents, rather than a representative sub set, but also, just like Lex Machina, Juristat & Ravel Law, it can be used to construct a database of information extracted from across the corpus of documents analysed. The challenge is to think what could this data be used for? What insight could be extracted? What new service could be offered? And ultimately How can this add value to the client?
So is AI a threat or an opportunity? It is both, to those focused on the routine it is a threat, to those focused on innovating/bespoke it is an opportunity, as pointed out by John O. McGinnis & Russell G. Pearce[20]. The advice therefore should be to understand these new technologies and explore what opportunities they create. This should not be limited to iterative improvements to the current process but more importantly to identifying opportunities for transformational change.
This comment covers much of the content covered by Ben at a Legal Geek talk on AI. You can read Ben’s slides here.
[1] John O. McGinnis & Russell G. Pearce “The Great Disruption: How Machine Intelligence Will Transform the Role of Lawyers in the Delivery of Legal Services” – http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2436937
[2] Elon Musk “AI is our biggest existential threat” – http://webcast.amps.ms.mit.edu/fall2014/AeroAstro/index-Fri-PM.html; Stephen Hawkins “AI could end mankind” – http://www.bbc.co.uk/news/technology-30290540; , Bill Gates “AI is a threat” – http://www.bbc.co.uk/news/31047780
[3] Richard Susskind & Daniel Susskind “The Future of the Professions: How Technology Will Transform the Work of Human Experts” – Oxford University Press 2015
[4] Altman Weil Survey “2015 Law firms in transition” – http://www.altmanweil.com/dir_docs/resource/1c789ef2-5cff-463a-863a-2248d23882a7_document.pdf
[5] Interview with Stephen Wolfram on AI and the future (July 2015) https://gigaom.com/2015/07/27/interview-with-stephen-wolfram-on-ai-and-the-future/
[6] Neota Logic http://www.neotalogic.com/
[7] LPA – Logic Programming Associates http://www.visirule.co.uk/
[8] EU Collateral Directive Advisor – https://demo.neotalogic.com/a/eucda
[9] Ravn http://www.ravn.co.uk/
[10] Kira https://kirasystems.com/
[11] Lex Machina https://lexmachina.com/
[12] Juristat https://juristat.com/
[13] Ravel Law https://www.ravellaw.com/
[14] CLAWS http://www.claws.io/
[15] Judicata https://www.judicata.com/
[16] Ross http://www.rossintelligence.com/
[17] Governance, Risk & Compliance Technology Centre http://www.grctc.com/
[18] Tom Butler “From Problems to solutions for GRC in Financial Industry” – http://www.grctc.com/wp-content/uploads/2015/12/GRCTC-Symposium-Prof-B-paper.pdf
[19] Richard Susskind “Tomorrow’s Lawyers: An Introduction To Your Future” – Oxford University Press 2013 & http://www.legaltechnologyjournal.co.uk/content/view/21/
[20] John O. McGinnis & Russell G. Pearce “The Great Disruption: How Machine Intelligence Will Transform the Role of Lawyers in the Delivery of Legal Services” – http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2436937