AI in legal decision-making - the human-machine Interface

PhD roundtable with Niels Chr. Ellegaard.

iCourts invites to PhD roundtable with Niels Ellegaard on his ph.d. research project “AI in legal decision-making  - the human-machine Interface”. 

The purpose of the research project is to:

  • explore and establish the limits, benefits and risks of using AI in the field of legal decision-making by courts and arbitration courts, through confrontation with an empirically tested Legal Decision Model,
  • On the basis of the result of this analysis on limits, benefits and risks of AI in legal decision-making; resolve the task of implementing proper human oversight and creating the right Human-Machine Interface when using AI in such legal decision-making processes.

To limit the scope of the Project it will be based on legal decision-making conducted by courts and arbitration courts operating under Danish law. Technology wise, the Project will deal only with Generative AI, and the scope of legal decision-making process will be adjudication in the form of subsumption (i.e. the process of the courts/arbitration courts' application of the law to the facts), which essentially mirrors the use case designated under the EU AI Act as described above.

The Project will address the legal requirements of the EU AI Act in respect of legal decision-making processes with the courts and arbitration tribunals, and to do so, first address the relevant theoretical and empirical aspects.

The overarching hypothesis governing the Project is that: 

  •  AI will in most aspects be able to simulate human legal decision making, but
  •  certain boundaries will exist in terms of AI's capabilities in legal decision-making.

 In this regard it is the assumption that the existence of such boundaries will not be overcome by further evolution of AI. Rather, the boundaries are due to inherent properties of legal decision-making, which cannot be simulated by AI, since legal decision-making is dependent upon agency, dynamics and fluidity that will require human participation.

 If it is possible to ascertain these inherent properties of legal decision-making it will also be possible to design the human-machine interface properly and hence assess the requirements under the EU AI Act.