Top-down and Bottom-up Legal System as a Gödel Machine with Object-Oriented framework
Cun Heng – The Chinese University of Hong Kong
-- The past decade has witnessed the exponential growth of data-driven industrial revolutions, from e-commerce to the internet of things (IoT), and from social media to its notorious abuses. A similar albeit milder trend is sweeping across the landscape of the legal profession, which has experienced radical changes brought by advancements in Artificial Intelligence (“AI”) in recent years. There were two dominant approaches to digitally represent the legal reasoning: the top-down approach where legal experts were engaged to analyze and design the legal system by its operational parts, and the contrasting bottom-up approach that uses machine learning to build up models for legal reasoning from the decided cases.
The top-down approach focuses on representing the domain knowledge into Ontological Structures (essentially typologies) that can be better communicated to the computer scientist in a CRISP-DM cycle. This top-down approach has long been employed in the legal profession to communicate legal concepts and logic relationships concisely and precisely. Vigorous attempts have been made to translate natural-language-based legal texts into machine-readable (executable) versions through tools such as Ontology Web Language (OWL). The top-down approach, despite its conceptual simplicity and clarity, suffers from a significant drawback: there is a knowledge acquisition bottleneck because such typological classifications are expensive to create, maintain and update such system.
On the other hand, the bottom-up approach has been grabbing the headlines in recent years, with vendors of legal analytic products that allegedly beat the best lawyers in legal analysis for the specialized domain of the law while taking a fraction of the time. However, when digesting such results, one must stop to ponder what technology stands behind the veil of the miraculous black box. There is an irresistible temptation to improve the predictive accuracy, and when the algorithm runs out of the permissible, admissible, and legal material of a case, it pulls in “extraneous” factors: parties, their lawyers, judges, and even the content of a judge’s breakfast, which are the prejudicial and “prohibited zone of reasoning” that the law consistently seeks to exclude from the calculation of justice.
In light of the ambitious progress to digitize the law, some normative and doctrinal controls must be in place to ensure that the learning programs do not merely learn from whatever “trivia” the past offers, but focus on the desirable normative values we aspire to enshrine in our legal systems. My recent paper, “Top-down and Bottom-up Legal System as a Gödel Machine with Object-Oriented framework,” therefore seeks to establish the possible fabrics of a framework representing the legal system using autopoiesis. An autopoietic system conception operates top-down while building up the ontological infrastructure with Object-Method-Property model bottom-up, which may together create a useful bridge between the two dominate approaches while maintaining the value-based legal system as we know it.
My paper starts top-down by discussing the system dynamics of the law as a self-reproducing, self-referential closed autopoiesis social system and then advances to analyze the Gödel Machine concept (“GM”) in computer science, which may be employed to digitally re-produce the organics of law’s operation as a social system. Then, the paper presents some fundamentals of the Object-oriented programming (“OOP”) paradigm and how its inherent similarity to legal reasoning forms the foundation which allows the building blocks of existing legal materials to fit bottom-up into the scaffold of the system of law designed top-down. A line must be drawn between the part of the law that should be built top-down and the elements that must be taken in bottom-up. The importance of maintaining a human-values-centered top layer of the law and benefiting from automating the lower layers through machine learning must be balanced.
The OOP-GM model could have multiple benefits. First, it can maintain the adaptivity of the legal system to social changes while consistently enforcing normative expectations. Second, the OOP-GM offers a comprehensive model of typological construction of legal concepts while maintaining compatibility to integrate with most objected-oriented programming languages. Thirdly, OOP-GM offers greater analytical precision for the communication of law, which provides for greater general access to the law as well as structural clarity in legal drafting, practices, and education. Lastly, machine learning functions employing specific control on the input parameters is unlikely to be influenced by extraneous factors, which provides for greater fairness and confidence in the legal system. My paper is on file in draft with Machine Lawyering. Please contact the editor if you would like to review a copy.