• Question: How do you convert legal facts and arguments into mathematical equations? Does this algorithm produce a variety of weighted outcomes, or are they binary? How can this save time in a courtroom?

    Asked by Cameron to Lewis, Katie, Christina on 8 Jan 2020.
    • Photo: Katie Atkinson

      Katie Atkinson answered on 8 Jan 2020:


      We represent legal cases using graphs, whereby the nodes in the graphs represent arguments and edges in the graphs represent atttacks between arguments. There are well defined algorithms that are then run on the graphs to determine, based on the attacks between arguments, which are the winning ones – those that are not attacked or are able to defeat any attackers. That’s a simplified summary but broadly describes how we represent the legal cases and reasoning so that the algorithms can be turned into a computer programme.

      Our models produce a binary outcome (i.e. decide for the plaintiff or decide for the defendant), as is done in court but we do use weightings to handle degrees of strength of arguments.

      Our experiments have used real world court cases to see if our AI programme can replicate the past outcomes (which it can with a very high level of accuracy), but we envisage the research being used by law firms in the first instance. We have worked with a major law firm to apply this approach to sets of their cases; the benefit for them is that when a new case comes in, if it was to be fed into the AI tool, they could quickly identify strong/weak arguments that could help get an early indication of the potential for success. Another benefit is that important factors from past cases are captured in our model so time could be saved searching for these and checking consistency.

Comments