Artificial Lawyer recently caught up with the UK-based President of the International Association for Artificial Intelligence and Law (IAAIL), Katie Atkinson, who is also Head of the Computer Science Department at the University of Liverpool.
Atkinson, along with several other speakers on legal AI, including yours truly, will be appearing at the now landmark European legal tech event, Lexpo in Amsterdam, 8th and 9th May.
Lexpo will be a great opportunity to hear Atkinson’s and others’ insights into many aspects of legal tech, ranging from how it connects to business development, using tech to improve operational efficiency, the need for design thinking in the law, and lastly, Artificial Lawyer’s favourite: the impact and uses of legal AI.
But, for those who do not want to wait until May, here is an interview with Atkinson which focuses on one of her main areas of research: computational models of argument.
First, please can you tell the readers a little about your academic focus in Liverpool (pictured above) in terms of AI and the law?
The University of Liverpool has been involved in the AI and Law community from its early days, including at the first International Conference on AI and Law in the late 1980s. Researchers at Liverpool have worked on a variety of topics covering, amongst others, case-based reasoning, data mining, natural language processing (NLP) and computational argumentation.
Our current focus is on collaborations with industrial partners to demonstrate how the theory can be put into practice and transformed into useable tools for legal professionals.
Also, congratulations on becoming President of the International Association for Artificial Intelligence and Law, which is, I understand, over 25 years old now. What is the IAAIL seeking to achieve now, especially given the rapid uptake of legal AI systems in the last two years?
It is a very exciting time to be involved in the AI and Law community. For nearly three decades the International Association for AI and Law has been promoting research and innovation on this topic and we have certainly seen increased interest in the community and its activities in the past couple of years. With legal applications of AI now becoming reality the community is keen to interact with practitioners to spread the word about research in AI and law, and transform research results into useful tools.
Computational Models of Argument (CMA) is one of your key areas of study. In this you are exploring how the defence and prosecution argue in court cases in order to see if you can model similar patterns of legal discourse and to see whether the software model will produce the same results. Could you tell us more about this process and what types of court case you have examined?
Computational Models of Argument is a research area in itself, but it has clear applications in the legal domain given the variety of legal tasks that involve argumentation.
To capture this in a model for use in automated reasoning, you first need to have a formal representation of the arguments and their interaction. Once you initialise the representation with data consisting of real arguments, you then need to be able to reason about which arguments attack and support each other and why. Following that, there are well defined algorithms you can run over the model to tell you which sets of arguments are the ‘winning’ ones and why.
In recent research at the University of Liverpool, we have taken three domains that are well known in the literature on AI and Law and have shown how we can represent the cases from these domains in a computational model of argument. These domains cover: possession of wild animals; trade secrets law; and the US exception to the Fourth Amendment. Now that we have shown we can model these cases, we are turning our attention to modelling more recent UK cases.
What has been the success rate in terms of the CMA system predicting/reaching the same result/decision as the actual court case?
In the wild animals domain we have 100% success; in the trade secrets domain we have 96% success and in the automobile exception domain we have 90% success. The total number of cases we have modelled so far is 47. We are very pleased with these initial results, but are working on expanding the number of cases fed into our model to show it can be scaled up.
What do you see as the main use of this software once it is fully developed? Would its main use be for litigation outcome prediction?
It could well be used for litigation outcome prediction to see whether there are certain arguments in a case that can easily be defeated or not. It could also be used as a tool to record and compare decisions in legal cases as part of training.
Bot-driven dispute resolution is already happening, with several legal tech companies exploring this area. Do you see your CMA work eventually producing a similar end result, i.e. a bot that could conduct actual legal mediation/dispute resolution?
Certainly in the short term I see the work on CMA being used in decision support systems in collaboration with a human. A particularly relevant aspect of the work on CMA is the ability to have explainable decisions.
Since modelling the reasoning through arguments makes explicit what the justification is for why an argument is or is not acceptable, this makes the work well suited to applications where explainability of outcomes is crucial.
On a technical note, are you using NLP and machine learning to analyse the court documents? And, what challenges does working with this type of unstructured data present (i.e. actual court dialogue between lawyers), given it can sometimes be quite unpredictable?
Indeed, it is a challenge to move from unstructured texts such as court documents to our structured formal models of argument. To do this, we use our own bespoke methods to analyse the documents and dialogues from which we extract the legal knowledge. This process still relies on human input and we use legal experts to validate the extraction of legal knowledge for use in the computational model.
An important focus in our work has been showing how we can perform automated reasoning over the legal cases in a variety of domains. Getting a higher level of automation for initialising our models is one of the next steps with the work.
Are you working with any law firms or corporates on CMA development?
Applying this research in any domain requires close engagement with the end users. The University of Liverpool currently has a Knowledge Transfer Partnership project, part-funded by Innovate UK, with Riverview Law, to assist the company in making use of AI research in its business.