Artificial intelligence can help people with law, democracy and policy making.

So how do the cold mathematical algorithms of AI translate into positive benefits for people? At Liverpool, it starts with an emerging field called computational models of argument (CMA), where debates are broken down argument by argument. “Humans use argumentation to debate and discuss in order to decide what to do,” says Professor Katie Atkinson, Dean of the School of Electrical Engineering, Electronics and Computer Science. “In CMA we are looking at formal models that can be turned into computer programs for automated argumentation.”

Building a successful model

To produce a successful model, a rigorous process of formalisation has been followed where all factors in a legal discussion of a case are taken into account in order to decide the case, either for the plaintiff or for the defendant.

For example, the objective facts of the case - was a person present at a scene, was an item in their possession, are collected for the case. Then the subjective issues are taken into account, which weigh the influence of societal values promoted by different decisions, like ensuring that the decision is fair to both parties yet does not reward undesirable behaviour. All these factors are weighted within complex algorithms that also take historical precedents into account and lead to a recommendation about which party the case should be decided for. These conclusions can then be evaluated against past cases with known outcomes, testing the processes’ conclusions against previous human judgements.

“You need to know which arguments attack each other, and which arguments beat each other, and why” Atkinson explains. “Once you have the factual matters established, you allow for the social values that you want to promote or demote by finding for plaintiff or defendant. We modelled legal domains in our formal model then turned this into a software program to see if it provided the same result as the original case -- and it did, with a high rate of success.”

In fact, the program, that forms part of Atkinson and colleagues’ ANGELIC methodology scored a 96%¹ success rate for getting the same original trial result for the domain of trade secrets law. Furthermore, investigating the cases it got wrong can pinpoint the decisive factors and even highlight any perceived inconsistencies with precedent cases and potential cases of miscarriages of justice, which in turn could assist law-makers when considering future cases.

Relevance of our research to industry

With highly effective working models, Atkinson is now working with top UK law firm Weightmans to see how the work can be implemented in an industrial setting. For such industry clients, the models could be used as business tools to advise on decisions, measure consistency in decision making and speed up case processing, as well as highlight problematic areas that require extra work, or probe whether to take on a case at all.

Atkinson makes clear that she sees these models as support tools rather than replacing humans, who can also make mistakes. "A particular feature of our work is that it explicitly promotes transparency of the reasoning through the AI tools explaining the recommendations in terms of the justifying arguments.”

Atkinson explains that the University of Liverpool has been heavily involved with the field of AI and law - going back to a 1987 conference in Boston, US - long before the recent boom in the field. As a result, Liverpool’s reputation in the area is extremely strong, and the University is known internationally for research in the area.

1. As reported in: 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), pp. 1-49.

Back to: Research