“No serious activity, and no legal practice, takes place without some guiding theoretical framework, be that framework conscious or unconscious.”
In this paper we look at legal practice from a technological perspective. Preeminent feature of this work is to extend on research paper by Schultz (2015) on arbitral decision making and postulate that the core driving theoretical framework for technological advances in legal profession is grounded in legal realism [1]. Firstly, we describe this theoretical framework, legal realism, in more detail to gain deeper insight into the decision-making processes relevant to legal profession. Secondly, we dive deeper into the nature and the basis of artificial intelligence (AI), as linked to legal profession. This is based on our technologically driven point of view while looking back into the legal practice. Lastly, we aim to propose a resolution in terms of delivering on an artificial intelligence-based decision-making platform, dispute-resolver operating in the adjudication of complex disputes. Can we design and build trusted AI arbitrageur?
Legal Realism
Based on Shaffer (2021), there are three core components of legal jurisprudence namely i) moral theorizing (natural law), ii) analytic theorizing (positivism) and socio-legal theorizing (legal realism) [2]. It is the third component we are going to focus our attention on. Legal realism focuses on the interaction of the ‘legal’ and external ‘extra-legal’ space in law’s development and implementation and application. Specifically in international context, this theoretical framework, firstly, focuses on importance of empiricism. It is this bridge to social sciences that we explore further. Secondly, this theoretical framework is grounded in philosophical pragmatism encapsulating legal institutions, norms and practices in forming and designing and driving social expectations and behaviours. This second component is critical when dealing with technological advances such as AI algorithmic solutions, as we shall explore further. This approach differs from normative and doctrinal approaches to international law in a sense of expanding on the predefined ‘rule of law’ and ‘rule of state’.
Furthermore, the author breaks down legal realism into three distinct interconnected dimensions namely i) behavioural, ii) critical and iii) pragmatic. For our purposes, we expand on new legal realism which defines law in terms of external and internal factors – such as reason versus power and must be viewed as a combination of these factors, not purely as a list of rules, regulations or pre-defined doctrines. As per Shaffer (2018), “Empirics inform pragmatic decision-making; pragmatic demands for decision-making inform the empirical questions asked. Empiricism needs concepts, and concepts must be updated pragmatically in response to empirical changes in the world so as to pursue human goals.”
In the international context, new legal realism forms deconstructive function by highlighting objective data-driven and empirical variations among different legal jurisdictions or international bodies. On the other side, new legal realism forms a constructive, pragmatic function by gluing new objective data-driven solutions and forming potentially new more robust legal bonds and solutions. This is our theoretical framework – data-driven deconstruction and subsequent data-driven construction approach to building complex decision-making AI systems. It is in this context that we explore a specific legal practice – arbitral decision making.
Schultz (2015) argues that arbitrators, like every other dispute-resolvers, are likely to rely on both ‘legal’ as well as ‘extra-legal’ factors when deciding the outcome and delivering final decision. Generally, on one side, arbitrators decide cases purely and entirely based on law within predefined book of rules, norms and principles dictating the likely outcome of the case. On the other hand, and other extreme, arbitrators decide the case based on external factors disregarding the rule of law, norms, or doctrines. The decision process is likely going to be driven by legal doctrines however also by external factors as it is not likely that arbitrageur, as the decision maker, operates in total isolation from current events, human conditions, or other daily influencing events. In short, application of law is one of the factors affecting arbitrageur in making final decision.
The author delineates these potential other external factors such as psychology of the judge, her personality, or subconscious biases. In addition, the author explores law & economics, empirical approach to legal realism further. This is the connection to our solution. The starting point in any economic analysis is the notion of maximisation of utility functions. Following on this reasoning the author spits the external factors into financial and non-financial. This makes sense, given the process of making “great” decisions will lead to more demand for arbitrageurs’ decision-making capabilities, as such leading to more – business. In short, the decision maker is incentivised to apply legal doctrines in the best interest of the parties involved. At the same time, any “outlandishly bad decisions” are likely going to be penalized in a form of – no business. The second factors, non-financial, play role in the adjudication of complex decisions. In this case, reputation has financial value, as well as non-financial value in terms of “fame”. The reputation here can be viewed for arbitrageurs’ survival ability and for the ability to be reappointed. This is driven by delivering optimal decision and resolving potentially complex cases to the satisfaction of all parties involved. In order to extend on Schultz (2015) paper, we expand on the notion of empiricism and pragmatism by providing our ability to track and monitor and measure decisions based on the officially established legal doctrines, norms and rules as well as the inherent and proposed external financial and non-financial factors as exposed by author.
Artificial Intelligence in Law
Expanding on previous section and across this paper, we position our perspective from outside of an arbitral decision-making legal practice as being technologically driven decision maker and technologist, AI platform builder, and pose and attempt to answer this simple question: Can we design and build trusted AI arbitrageur? In brief, can we design a complex AI system capable of making judicial decisions based on officially established legal doctrines, norms, and rules as well as the inherent and proposed external financial and non-financial factors as exposed by Schultz (2015)?
Generally, it is becoming a common perspective that AI is transforming the field of Law.[1] Faggella D., (2021) outlines number of viable business solutions to current AI application in legal practice and current applications in Law.[2] Based on the conclusion, legal practice is forced to adopt these new technological advances starting with added and enhanced efficiency in operational readiness. Legg and Bell (2020) argue that AI can reduce costs in some elements of legal practice and as a result lawyers can focus more on added-value proposition to be delivered directly to the clients [9]. The author’s conclusion is that AI as a mechanism is unlikely to replace the legal profession. There are various levels to be crossed for this be an eventuality. One of the AI techniques that was mentioned in the article relates to Natural Language Processing (NLP). This is machine’s ability to read and write text. In legal context we argue this is one of the core algorithms to be implemented at the early stages of the AI implementation. However, also we see this space as potentially capable to predict the outcome of the case based on predefined legal doctrines, norms, and legal rules. When expanded further to include financial and non-financial external factors of the arbitragers, potentially being the final solution to a fully trained AI complex system.
AI in Law in not a panacea, yet. Surden (2019) provides an overview of current AI algorithms as applied to legal professions. He concludes that todays’ form of AI is capable to deliver results by harnessing patters, rules and heuristic proxies that allow for decision making in a narrow context. The AI algorithms are still far from dealing with abstractions, understanding meaning and reasoning and handling completely unstructured tasks [7]. In the same vein, McCarl (2021) continues to highlight the limitation of current AI algorithmic solutions starting with Susskind AI vision:
“an online service that contains vast stores of structured and unstructured legal materials (primary and secondary sources), that can understand legal problems spoken to it in natural language, that can analyse and classify the fact pattern inherent in these problems, that can draw conclusions and offer legal advice, and that can even express this guidance in some computer-simulated voice.”
The author describes numerous obstacles AI is facing when dealing with legal profession one of which is the fact that “legal rules tend to be incomplete, logically and semantically ambiguous, sometimes inconsistent, and frequently hard to find or even to identify. [Additionally,] the law is adversarial; legal problems frequently have no one right answer.” [4]. Hence the dilemma of an arbitrager and the topic of our discussion. To continue in this argument, Eliot (2021) in his short note, provides interesting view that yes, it is going be rather challenging to build high sophisticated AI judge in a form of a black box [6]. However, on the other hand, AI can add value to the legal profession in simplification and ability to align across multiple levels and to “complete” legal rules (if this is ever possible) [8]. Similarly, Sanders (2021) outlines that AI is already used to review legal documents (via NLP mechanisms) as well as assisting in research prior to or during a case and potentially providing prediction as to the ultimate case decision or outcome and notes “…AI can analyze contracts both in bulk and individually much more efficiently than a human could. This makes it much easier to make comments, allows firms to quickly move through contracts, and reduces the number of mistakes and overlooked details. AI is incredibly efficient, which means it can perform legal research much faster, so lawyers are able to build better cases…”
The connection between AI and legal realism is the data – the empirical part of the legal realism. The data and empirical approach to decision making. Utilization of legal rules and regulations and norms in making arbitrager decision if taken with data related to that specific arbitrageur – in terms of number of cases decided and participated in, decision processing and reasoning or measure of reputational exposure across the arbitral decision-making space. Then, it is the ability of the AI mechanism to connect these data points that we wish to explore further. Finally, it is the pragmatical part of legal realism that can formulate potential AI algorithms delivering objective judicial decisions – inclusive of ‘legal’ as well as ‘non-legal’ factors as stated previously.
AI Arbitrageur
In this paper we argue that it is the theoretical framework of legal realism that is the core backbone of potential AI solutions in legal practice. In conclusion, we have assumed the legal practice is focused on arbitral decision making in line with Schultz (2015). In legal profession, NLP, natural language processing is a branch of AI and deep learning space that is aimed to deliver on text recognition and potential text generation. Current use cases are i) language translation applications such as Google translate, ii) word processors such as Microsoft Word and Grammarly that employ NLP to check grammatical accuracy of texts, iii) Interactive Voice Response (IVR) applications used in call centers to respond to certain users’ requests, iv) personal assistant applications such as OK Google, Siri, Cortana, and Alexa, v) text generation and creation of Machine Generated News Streams or legal documents and vi) sentiment analytics among other uses.
There are several techniques currently used in NLP whereby syntactic analysis and semantic analysis are the two main modeling architectures. NLP algorithm is used to apply grammatical rules to a group of words and derive meaning from this selection, this is syntactic analysis approach. On the other hand, semantic analysis requires the algorithms to understand the meaning and interpretation of words and how sentences are structured. This is more challenging approach. However, as we have seen in previous sessions, we can also utilize prediction modeling in a form of LSTM RNN [11]. This deep-learning approach to text reading and generation utilizing neural networks is of our interest here.
Practically to start building AI based decision-making platform, dispute-resolver operating in the adjudication of complex disputes in the legal practice, would require full comprehension of legal realism as the core theoretical framework. Starting point being the rule of law, legal norms, and black letter of law, however the ‘non-legal’ factors need to be included and considered to be incorporated in the overall algorithmic solution. As outlined by Schultz (2015) data points involving financial and non-financial incentive schemes and norms as used by individual decision makers. In short, we are at the inception of being able to design, build and implement rudimentary AI Arbitrageur. However, more inter-disciplinary work is still needed in this space.
References
[1] Schultz T., (2015), “Arbitral Decision Making: Legal Realism and Law & Economics”, King’s College Dickson Poon School of Law, Legal Studies Research Paper Series, No.2015-28
[2] Shaffer G., (2021), “Legal Realism and International Law”, UC Irvine School of Law Research Paper Series, No.2018-55. < https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3230401 >
[3] Sanders K., (2021), “The Future of AI in Law: Changing the Legal Landscape”, National Law Review, Vol XII, No. 19, April 2022. < https://www.natlawreview.com/article/future-ai-law-changing-legal-landscape >
[4] McCarl R., (2021), “The Limits of Law and AI”, University of Cincinnati Law Review, Vol. 90, No. 3, 2022. < https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3805453 >
[5] Zuckerman A.S., (2020), “Artificial Intelligence – Implications for the Legal Profession, Adversarial Process and Rule of Law”, Oxford Legal Studies Research Paper No. 9, 2020. < https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3552131 >
[6] Eliot L.B., (2021), “Legal Simplification and AI”, Stanford University Centre for Legal Informatics < https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3955411 >
[7] Surden H., (2019), “Artificial Intelligence and Law: An Overview”, University of Colorado Legal Studies Research Paper No. 19-22, < https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3411869 >
[8] Eliot L.B., (2021), “Unintended Consequences in the Law and AI”, Stanford University Centre for Legal Informatics < https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3995606 >
[9] Legg M., Bell F., (2020), “Artificial Intelligence and the Legal Profession: Becoming the AI-Enhanced Lawyer”, UNSW Law Research Series, Vol. 63
[10] Maas A.l., Daly R.E., Pham P.T., Huang D., Ng A.Y., (2004), “Learning Word Vectors for Sentiment Analysis”, Stanford University
[11] Hochreiter S., (1997), “Long-Short Term Memory”, Neural Computation, December
[1] Toews R.,(2019),”AI Will Transform the Field of Law”, FORBES, https://www.forbes.com/sites/robtoews/2019/12/19/ai-will-transform-the-field-of-law/?sh=2a7dbf897f01 “The law touches every corner of the business world. Virtually everything that companies do—sales, purchases, partnerships, mergers, reorganizations—they do via legally enforceable contracts. Innovation would grind to a halt without a well-developed body of intellectual property law. Day to day, whether we recognize it or not, each of us operates against the backdrop of our legal regime and the implicit possibility of litigation.” [2] Faggella D., (2021),”AI in Law and Legal Practice – A Comprehensive View of 35 Current Applications”, Blog https://emerj.com/ai-sector-overviews/ai-in-law-legal-practice-current-applications/
Comments