Hype v Reality

Professor Katie Atkinson offers a reality check on legal tech: what’s workable now and on the horizon?

Artificial Intelligence (AI) is no longer a topic of discussion restricted to the laboratories of academia and technology companies. In addition to the large volume of media reports on general developments in AI, a variety of professional domains are looking to apply AI to their specific tasks. The legal profession is one such domain. Although interest shown in AI by legal professionals is currently highly prevalent, researchers in the field of AI and law have been quietly conducting research on this topic for over 30 years.

While in the 1970s groups of researchers were already investigating the application of AI to legal reasoning, the field became firmly established in its own right through the First International Conference on AI and Law (ICAIL) held in Boston, Massachusetts in 1987. The conference is supported through the work of the International Association for AI and Law (IAAIL), which was founded in 1991 and was followed by the introduction of the Artificial Intelligence and Law journal in 1992. The ICAIL conference has been held biennially ever since 1987, with the most recent taking place in London in 2017. This conference attracted a record number of attendees, including an unusually large number of participants from law firms, reflecting the current attention on the topic. So why is there a sudden interest when researchers have been working on AI and law for decades, with relatively little interest previously shown outside narrow academic circles?

From laboratory to the law

The interest in AI and law has gone hand in hand with the raised profile of the general topic of artificial intelligence. The field of AI has experienced its own ‘boom and bust’ cycles in the past decades, but we are currently at a stage where the fruits of AI research are not only being seen, but are now actually being deployed in a variety of application areas. While some large scale applications of AI, such as self-driving cars, frequently grab the headlines, there are a whole host of examples where AI technologies are successfully being deployed. Huge strides have been made with intelligent personal assistants on smartphones that can perform natural language processing and generate appropriate responses to queries posed in natural language. Machine learning is used in a wide variety of everyday applications from recognition of individual faces in photo collections to producing personalised recommendations on shopping and entertainment applications. Information retrieval systems such as those used in IBM’s Watson are now being applied in the medical domain, with one success area being support for identification of cancer treatments (ibm.co/2p99IB2). These are just a few examples of how AI is currently being deployed in varied applications. So what is happening with AI in the legal domain?

The legal domain, like many other domains, covers a variety of tasks and different techniques are being used to address different tasks. E-discovery tools are now commonplace, with research continuing on advancements making use of various AI techniques (see, for example, the workshops associated with the ICAIL conference over the past ten years: bit.ly/2kGkvWb). Machine learning is being used in a variety of ways to support legal work, for example through tools to assist with contract analysis (such as kirasystems.com) and risk management (such as www.legalrobot.com). Plenty of tools are now available to support the automated production of documents, in addition to tools to review and analyse existing documents – see, for example, the list of companies from the CodeX Center’s Legal Techindex (techindex.law.stanford.edu).

Computational models of argument

In addition to the successful AI-based products that have made it to the market, a whole wave of innovations continues to be investigated in the academic community. One such relevant topic is the field of computational models of argument, which is the focus of my own research. In this field, AI researchers have developed techniques to represent arguments and the interactions between them, specifically how arguments attack and also defend one another. Additionally, meticulous methods have been defined to reason about sets of attacking arguments to determine the mutually justifiable ones, taking into account objective facts and subjective preferences. All this is captured through formal models that can be easily translated into a computerised form that allows for automated reasoning about arguments. Given the pervasiveness of arguments in legal work, law has been a focus area of research on computational models of argument. (For an excellent review of work on computational models of argument in law, see: H Prakken, and G Sartor, 2015, ‘Law and Logic: A Review from an Argumentation Perspective’, Artificial Intelligence, Vol 227: 214-245.)

There are numerous groups worldwide who have been working on computational models of argument for a variety of different purposes, such as e-democracy applications, natural language processing, modeling dialogues and argument mining from written texts. A body of recent research conducted at my own institution, the University of Liverpool, has focused on argumentation for case-based reasoning and builds on influential prior systems developed for reasoning with legal cases. These earlier systems are the HYPO system (see: K Ashley, 1990, Modelling Legal Argument: Reasoning with Cases and Hypotheticals, Bradford Books/MIT Press, Cambridge, MA) and the CATO system (see: V Aleven, 1997, Teaching case-based argumentation through a model and examples, PhD thesis, University of Pittsburgh). The latest research provides a methodology for capturing reasoning about legal cases within an argumentation model, which can be transformed into a software programme that reasons about the cases in a given domain. Initial experiments have shown a high success rate of the decision-support software being able to produce the same outcome – find for plaintiff or find for defendant – as the human judges came to in the real life cases. (See L Al-Abdulkarim, K Atkinson and T Bench-Capon, 2016, ‘A methodology for designing systems to reason with legal cases using abstract dialectical frameworks’, Artificial Intelligence and Law, Vol 24(1): 1-49.)

The LegalTech space

Fundamental research on the theory that underpins AI solutions is the staple of academic conferences. However, in the past couple of years there has been an explosion in the number of more commercially oriented conferences in the LegalTech space. Such events are cropping up across different continents and attract a high volume of legal practitioners eager to find out about technology solutions that can be applied in their own company. Furthermore, there is also an increase in the number of law firms developing their own technology in-house and, understandably, these new endeavours make for attractive media headlines. The current activity in the LegalTech space, and the media attention this is now receiving, is creating a lot of excitement around the topic of AI for legal applications. So is the hype is warranted?

Scrutinising developments

As described above, there are plenty of success stories that show how AI is being used here and now. However, retaining a solid grounding in the reality of what is and isn’t currently achievable will provide a sound basis for progress to the next level as the technologies mature further. Of course, there are also a host of non-technical issues surrounding the use of AI in law. Scrutability of automated reasoners is an important property for building up trust by adopters and end users. The academic community places great emphasis on solid evaluation of new research and peer-reviewed conference and journal articles provide a mechanism for this important evaluation to be documented and disseminated. Then there is the task of transferring research into fielded applications and currently practitioners seem more open to this than ever. This desire for further collaboration has been demonstrated recently through the Workshop on AI in Legal Practice that was held in conjunction with the ICAIL 2017 conference. The aim of the workshop was to bridge the gap between legal professionals and AI and law researchers. The topics discussed were wide ranging and covered discussions about the need to develop shared languages and datasets, hurdles around intellectual property concerns, the alignment of different stakeholders’ interests and how to keep the conversations grounded amongst all the hype (for a summary of discussions, see: bit.ly/2snuLWn).

My own perspective is that there has never been a more exciting time to be involved in the field of AI and law – innovative technology solutions are being developed to assist with various aspects of legal work and there is currently unprecedented interest from the legal community. Examining rigorous, documented evidence of solutions in this space offers a way to navigate the hype that is surrounding the field. In addition to the technical solutions, there are also many social issues surrounding the adoption of AI in legal work: automation of jobs, access to justice and regulation of AI technologies, are just a few of the important ones that will require engagement with a range of stakeholders. Opportunities to shape this landscape are plentiful.