This fourth conference examines the growing impact of digitization and artificial intelligence (AI) on the law.
It puts forward the hypothesis that there are limits to the computability of legal reasoning and hence to the use of machine learning and related functionalities of AI systems to reproduce the central processes of the legal system. Machine learning uses mathematical functions to perform analysis and classification tasks. In "deep learning" approaches, data is organized into concepts expressed hierarchically, with more complex concepts being built up from simpler ones. The classification of data into concepts is achieved through the use of weights whose values are adjusted over time, as the system receives new inputs. Thanks to its recursive operations, the system "learns" or self-adjusts, so that, in principle, errors are progressively purged from the system. There is more than a superficial resemblance here to the workings of the legal system, but there are also crucial differences and, above all, limits to the extent to which the linguistic categories of legal thought can be expressed in mathematical form.
The project to apply machine learning to law has the goal of a perfectly complete legal system, i.e. one in which the content and application of rules can be fully specified, regardless of the diversity and variability of the social circumstances in which they take place. In this world of "legal singularity", law operates in a perpetual balance between facts and norms. Put forward as a response to the incompleteness and contingency of law, the project of legal singularity is also a proposal to eliminate legal reasoning as the basis for resolving disputes and distributing powers, rights and responsibilities. As such, it risks undermining one of the key institutions of a liberal-democratic order. To avoid such an outcome, thought will need to be given to defining limits to the use of machine learning and other data-driven approaches in the administration of law. In the final analysis, the success of such measures will depend on the law's ability to maintain the autonomy of its operations in the face of global technological change, an outcome that is far from guaranteed.
