On Friday 23rd of July 2021, Legislate’s CEO Charles Brecque participated in the legal AI panel at ICML as part of the workshop on deploying and monitoring machine learning solutions. The panel included Jessica Montgomery, Executive Director of the Accelerate Programme for Scientific Discovery at the University of Cambridge and Teresa Scantamburlo, Post-doctoral researcher from the European Centre for Living Technology. The panel was chaired by John Armour, Professor of Law and Finance at Oxford University. This article provides answers to some of the questions discussed on the panel.
What is Machine Learning?
Machine Learning is a field which focuses on the development of models (and in practice algorithms) which are trained to identify patterns in data. Machine learning algorithms can find patterns between input data and a labelled target (supervised learning) or can identify clusters in a data set’s structure (unsupervised learning). Finally, because machine learning algorithms are learning the patterns themselves, the relationships they learn will be accompanied by a degree of uncertainty which is influenced by the data they have access to during the training and whether there is an actual pattern or not. This means that a machine learning model’s forecast will never be 100% certain and might be irrelevant if the out-of-training data being analysed by the model is too different from the data used in the training.
Which examples of Machine Learning exist today?
Many examples of Machine Learning exist in our daily lives ranging from spam mail predictors to image recognition in smart phone devices. These applications work well because there are many examples of “cat pictures” and the features which define a cat are fixed. However, deploying machine learning algorithms can be challenging when there are not many examples of the outcome you are training the model to predict or when the outcome can change which will in turn “surprise” the trained model. As a result, not all use cases have been able to capture the benefits of machine learning, in particular the legal system.
What are the challenges today with applying Machine Learning to the legal system?
The legal system is particularly challenging for machine learning algorithms due to the fact that the majority of data is text based (i.e. highly variable) and the domain typically requires legal expertise to be understood correctly. Moreover, acquiring legal data to train machine learning models can be challenging due to its sensitive nature around privacy and data processing. As a result, public legal databases are rare and have limited occurrences of outcomes you might wish to predict such as in case law. These types of challenges have heavily limited the scope of machine learning in the legal system to applications which automate simple tasks or require a human in the loop to interpret the result and manually finish the task.
What solutions did Legislate suggest to overcome these challenges?
Whilst many use cases in the legal system would require sensitive data to be collected for training purposes, Legislate’s knowledge graph circumvents this challenge. Whereas a machine learning model will induce patterns from the data to make decisions on out of sample data, a knowledge graph will deduce patterns in the data based on the rules it has been given. This means that the only data which needs to be collected by the Legislate is the data contained in a contract which is then assembled by the knowledge graph applying legal expertise as rules.
Can Machine Learning make the legal system fairer?
Machine learning can make the legal system fairer by making services more accessible either by reducing the cost of the service through automation, or by making legal data more accessible through aggregate insights. For example, Legislate provides aggregate statistics of the main terms used in a contract which gives a useful starting point to someone who is not familiar with contracts. Reducing the cost of legal services through automation will require the automation to provide results of equal quality as humans which might be challenging due to the complexity of the legal system. However, knowledge graphs combined with machine learning can potentially overcome these challenges in due course.
Legislate is an early-stage legal technology start-up which allows large landlords and small businesses to easily create, sign and manage contracts on their own terms. Legislate’s knowledge graph approach (United States Patent No. 11,087,219) is unlocking the full potential of contract data. Legislate’s team marries technical and legal expertise to create a painless, and unique, contracting experience for its users. Legislate is backed by Parkwalk Advisors and Perivoli Innovations.