Decoding Legal Complexity: How Ontologies and Taxonomies help to make Legal Applications explainable

As we follow our quest on legal informatics we need to turn to the  important topic of explainability and transparency of applications in the legal domain. While this issue is widely discussed in the context of Generative AI its general importance for any legal application needs to be stressed.

Introduction

In recent years, the integration of software into the legal domain has marked a significant transformation in how legal professionals approach research, case management, and even decision-making processes. This digital evolution, driven by advancements in artificial intelligence (AI) and machine learning (ML), promises to enhance efficiency, reduce manual labor, and improve accuracy in legal proceedings. However, this integration also raises critical concerns regarding the transparency and explainability of the outcomes generated by these software solutions.

The importance of transparency and explainability in the legal context cannot be overstated. Legal decisions impact human lives, rights, and societal norms; thus, the processes leading to these decisions must be scrutinised with utmost diligence. The application of software in legal analyses, predictions, and recommendations introduces a layer of complexity, where the rationale behind a decision or advice can be obscured by the "black box" nature of certain AI algorithms. This opacity challenges the fundamental principles of fairness, accountability, and justice that underpin the legal system.

Ensuring that the outcomes of software computations in the legal domain are transparent and explainable is paramount for several reasons. Firstly, it upholds the principle of due process, ensuring that the parties subjected to legal computation understand the basis of the computational results relating to them. Secondly, it fosters trust among legal professionals and the public in the technology that increasingly influences legal outcomes. Lastly, it contributes to the iterative improvement of legal software by allowing for the critique and refinement of algorithms based on clear insights into their functioning.

The more we are introducing automation in the legal domain, it becomes imperative to address these challenges head-on, advocating for the development and implementation of technologies that are not only powerful and efficient but also transparent and accountable. This article aims to discuss some tools and technologies that help to support transparency and explainability in legal computations.

Complexity of Legal Information

Legal information is inherently complex, necessitating structured representation for effective understanding and application. Taxonomies and ontologies play a crucial role in this context by organising legal concepts and their interrelations in a systematic manner. Ontologies, in particular, offer a formal and explicit specification of conceptualisations within the legal domain, enabling software systems to reason with legal concepts effectively. They support various legal informatics applications, including semantic annotation and legal argumentation, by providing a structured framework for the analysis and interpretation of legal text.

Now what are Taxonomies and Ontologies?

Taxonomies and ontologies are essential tools in organising knowledge across various domains, including the legal field. Taxonomies provide a hierarchical classification system, arranging legal concepts into categories and subcategories based on their characteristics and relationships. This structure simplifies the organization of complex legal information, making it more accessible.

Ontologies take this organization a step further by not only categorising concepts but also defining the properties of these concepts and the relationships between them. In the legal context, ontologies allow for a more nuanced understanding of legal terminology, principles, and their applications. They facilitate the identification of connections between different legal concepts, supporting more sophisticated information retrieval and data analysis.

In practice, taxonomies and ontologies are used in various legal technologies, such as legal information retrieval systems, document management systems, and legal decision support systems. They enable these technologies to process legal information more efficiently and with greater accuracy. For example, a legal research tool powered by a well-designed ontology can provide more relevant search results by understanding the context and relationships between search terms.

Moreover, the use of taxonomies and ontologies in the legal field facilitates better communication and understanding between legal professionals and enhances the interoperability of legal information systems. By providing a common framework for organizing legal knowledge, these tools can help bridge the gap between different legal jurisdictions and languages, contributing to the globalization of legal information and practices.

For example imagine a legal ontology designed for contract law. This ontology includes concepts such as "Contract," "Offer," "Acceptance," and "Consideration," each with defined attributes and relationships. For example, an "Offer" must be made by an "Offeror" to an "Offeree" and may include conditions that need to be met for acceptance. The ontology maps out how these concepts relate, such as an "Acceptance" following an "Offer" leading to a "Contract." This structure allows legal software to understand and analyse contracts, identifying whether all legal criteria are met for a valid agreement.

Explainability and Transparency

Now why are Taxonomies and Ontologies important in the context of explaining the out of a legal computation?

Ontologies and taxonomies provide a structured framework that helps in interpreting and explaining the outcomes of computations, especially in complex domains like law. They define and categorise concepts and their relationships, making it possible for computational systems to "understand" and process legal information in a human-like manner..

By defining and organising concepts and their relationships, ontologies and taxonomies enable computational systems to "understand" complex data structures and reason about them logically. This structured knowledge allows these systems to process information, draw conclusions, and explain their reasoning processes transparently. For instance, in a legal decision-support system, ontologies help in mapping the legal reasoning behind a case outcome by linking facts, evidence, and legal principles, thereby providing a clear, explainable path from input data to conclusion.So systems based on such concepts are able to “reason” about the data, make inferences, and provide explanations for their outcomes, making the decision-making process more transparent and understandable to users

However when systems are based on statistical self learning concepts they follow (are least currently) a different approach. Generative AI, particularly models like GPT (Generative Pre-trained Transformer), is therefore not inherently transparent. While the outcomes it generates can be impressive, understanding the specific reasoning behind each decision or prediction is challenging if not impossible. The complexity of these models, which can contain billions of parameters, makes it difficult to trace how they arrive at a particular output. Efforts to improve the explainability of AI are ongoing, but transparency remains a significant challenge in the field of generative AI.

Another important aspect is the accessibility of code to verify the computational processes. Open source software is generally considered more transparent than proprietary software because its source code is accessible to the public. This openness allows users and developers to examine the code to understand how the software works, identify potential flaws, and verify security. In contrast, proprietary software keeps its source code private, limiting the ability to scrutinise or understand the software's inner workings. However, transparency also depends on the community's engagement and the documentation provided, not just the openness of the source code.Proprietary code my be audited and reviewed by trusted organisations to certify its compliance with  principles and thus secure the necessary levels of reliability.

Conclusion

To make a computational outcome transparent, it involves designing systems that can explain their processes and decisions in human-understandable terms. This includes e.g. incorporating explainable AI (XAI) techniques that detail how input data lead to specific conclusions, ensuring the data and algorithms used are accessible and understandable, and implementing interfaces that allow users to query and understand the reasoning behind outcomes. Additionally, documenting the decision-making process and providing clear, accessible explanations of the algorithms and data used are essential steps towards transparency.

As the legal domain is obviously and understandably quite sensible any application of software will be viewed critically. Therefore the further use of computational automation in more complex areas of the legal domain will require us to invest and develop systems that meet highest standards of explainability and transparency.