A Simple Guide to Legal AI - Part 2
Published on Thu, October 06 byUwais Iqbal

The Different Types of AI

A common categorization of the different kinds of AI is based on what the AI system is capable of doing. Researchers usually break AI down into two kinds: Strong AI and Weak AI.

Strong AI refers to an AI system which has reached a level of human intelligence. The system can learn, perceive, understand, and function completely like a human being.

At the moment, we haven’t developed such systems so Strong AI is still something purely theoretical. Many researchers are sceptical as to whether Strong AI systems are even possible. Strong AI is also known as general AI or artificial general intelligence (AGI).

Weak AI refers to an AI system that can perform a specific and precise task like extracting a field from a document or clustering a set of documents. The AI systems we have today are all examples of Weak AI.

Weak AI is also known as Narrow AI.

The claim that AI will replace lawyers certainly grabs attention and is great for headlines. However, it's a claim about Strong AI. Strong AI systems are still a theoretical estimation - we don’t know if we’ll ever get to that point. It’s fair to say that lawyers are still around today and will be around for a lot longer than we think.

Where AI can be useful in the legal profession is in the form of Weak AI. With a focus on specific tasks, AI systems can be developed to help scale expertise and save associates from hours of grueling grunt work.

Natural Language Processing and its Branches

Natural Language Processing (NLP), not to be confused with Neuro-Linguistic Programming (also NLP!), is a branch of AI that enables computers to understand human language in both written and verbal forms. NLP sits at the intersection of AI and computational linguistics and draws on research from both fields.

Natural Language Understanding (NLU) is a sub-branch of NLP which uses syntactic (grammar) and semantic (meaning) analysis of text and speech to determine the meaning of a sentence.

Natural Language Generation (NLG) is another sub-branch of NLP. While NLU focuses on how machines can understand text, NLG focuses on how machines can generate text in a way that is indistinguishable from human composition.

NLP evolved from computational linguistics where the methods traditionally used were based on a statistical analysis of words in a text. More recently, with the development of Machine Learning and Deep Learning algorithms, NLP has been supercharged with models like BERT and GPT-3 that can process, understand and generate text in a way that hasn’t been possible previously.

Some common applications of NLP in legal include:

  • Extracting key fields from a collection of contracts as part of a due-diligence use case
  • Automatically organizing and clustering documents as part of an eDiscovery use case
  • Classifying documents against a pre-defined legal taxonomy
  • Comparing clauses to identify deviations in wordings and surface risky positions during contract review

NLP should be of particular interest to legal professionals and the legal sector more broadly. The medium of any legal work is the written word. With the broad adoption of technology, this written word is created, processed and stored in digital formats.

Law firms and legal departments don’t realize this, but they are sitting on mountains of well-drafted, high quality and high-fidelity textual data. NLP can be leveraged to put this data to work in applications that could reduce manual grunt work and enhance and improve existing workflows.

The Different Types of Machine Learning

The different algorithms and approaches in Machine Learning can be summarized in three broad types:

  1. Supervised Learning
  2. Unsupervised Learning
  3. Reinforcement Learning

Supervised Learning is a type of ML where a model is trained on a specific task. A typical example of a supervised learning task is classifying spam emails. The model is supervised meaning that it is given lots of examples of emails along with information about whether the email is spam or not. From the labelled data, the model can learn what spam emails look like so it can then successfully detect and classify spam emails. There is a barrier to entry in order to use supervised learning since it requires a training dataset that is labelled by humans. In the legal industry, supervised learning underpins many contract review and contract analysis tools.

Unsupervised Learning is a type of ML where the model is completely data-driven. A typical example of an unsupervised learning task is automatically clustering documents, such as in eDiscovery applications. The model is unsupervised so it learns how to cluster based on patterns that exist within the data itself. Unsupervised learning is attractive for many applications because of the low barrier to entry. All that is needed is the raw dataset. It doesn’t require data to be labelled by humans.

Reinforcement Learning is a type of ML where the model is trained to learn from mistakes in a simulated environment. Reinforcement learning is typically used in robotics and manufacturing applications. The model learns to perform a task by reinforcement; positive outcomes are rewarded and negative outcomes are punished. It is quite similar to how pets are trained to perform particular actions by reinforcing correct behaviors with rewards.

Supervised learning is particularly important since it is a way of capturing expertise that can then be scaled using a model. Within the legal space, there are many specific tasks that require particular expertise. For example, extracting fields from a set of contracts is a specific task that can be taught to a model using supervised learning. Supervised learning could be used to train models to learn how to perform these specific tasks and legal expertise could be scaled so the burden of manual work is considerably reduced.

A00F7639-6405-48EB-AB30-612C9D3779FA.pngPutting it all together

In the context of law firms and legal teams, there are particular manual tasks that can be taught to machines. In particular, NLP with supervised learning can be used to train weak AI systems to perform specific tasks like comparing two clauses to identify deviations in wording or extracting dates and party names from a set of contracts. These kinds of weak AI systems can help scale legal expertise and reduce the need for manual grunt work.

Uwais is the Founder of simplexico - The Legal AI Consultancy. He has over 5 years of experience designing, building and delivering Machine Learning and NLP solutions across leading legal tech start-ups and a corporate innovation lab. He has held roles as a Machine Learning Engineer at Eigen Technologies, and as a Senior NLP Data Scientist at Thomson Reuters Labs and ThoughtRiver. Uwais also holds a BSc and MSc in Physics from Imperial College London. Connect with Uwais here.

We use cookies to monitor the performance of our website, improve user experience, and assist in our marketing efforts. By continuing to browse our site, you agree to our use of cookies.