Definitions

Artificial Intelligence (AI) is the ability of a computer system to perform tasks which usually require human intelligence. These tasks include analysing various types of data (e.g. text, numbers, images and audio), understanding human language and making informed decisions. AI improves its performance by learning from past experiences.

Generative AI (GenAI) is a type of AI that can create content, such as text and images, by analysing and learning from data. It's like asking the system “tell me something I don't know”.

Cognitive AI is a type of AI that focuses on extracting facts and insights from documents. It's akin to asking the system "what do we know for sure?".

Retrieval-Augmented Generation (RAG) AI combines information retrieval with content generation and provides answers only if relevant documents are found in the workspace. It's like asking the system "tell me what you found in the records”.

Electronic disclosure (eDisclosure) refers to the process of identifying, collecting and reviewing electronic data to produce evidence in legal disputes.

Technology Assisted Review (TAR) is a method often used in eDisclosure to expedite the document review process. It uses machine learning algorithms to classify documents based on input from human reviewers. Simply put, it's equivalent to asking the system “find me more like this”.

Continuous Active Learning (CAL) is an advanced form of TAR that involves ongoing learning from manual coding decisions made by reviewers. Unlike traditional TAR methods, CAL does not rely on a predefined seed set of documents and dynamically adjusts the document set presented to reviewers based on the evolving understanding of relevance.

Large Language Models (LLMs) are advanced AI systems that understand and generate text similar to humans. They learn from extensive amounts of text data and use this knowledge to respond to various inputs. Examples of LLMs can be found on our Blog page.

Machine Learning involves training algorithms to classify documents, predict relevance and identify patterns to improve document review efficiency in eDisclosure.

Natural Language Processing (NLP) is a branch of AI enabling computers to understand and process human language found in electronic documents and communications. For example, it can analyse the sentiment expressed in text, such as the tone of emails exchanged between parties in a legal dispute, to determine if there are any indications of hostility or agreement that may influence the case's outcome. Additionally, NLP can identify named entities - such as names of people, organisations, dates and locations in legal documents, facilitating quick identification and retrieval of relevant information during the eDisclosure process.

Algorithm Bias occurs when machine learning tools favour certain documents over others during review, often due to uneven representation of data sources in the training dataset. For example, if the dataset contains predominantly emails from the legal department, the algorithm may prioritise legal-related documents, potentially overlooking relevant information from HR (or other) departments.

Prompt Engineering (Prompting) is a technique used in AI, particularly with Large Language Models (LLMs), to guide the AI’s responses. It involves crafting specific inputs or ‘prompts’ that direct the AI to produce the desired output. This can be especially useful in tasks such as document tagging, where specific criteria are inputted by users and the AI tool returns documents which meet that criteria for second-level review. Effective prompting can greatly enhance the efficiency and accuracy of AI responses, particularly when there are large volumes of documents involved.

AI Sampling is an AI-powered method that enhances the accuracy and representativeness of sample sets by selecting unbiased data for review. By analysing data graphically, it provides a more comprehensive view than standard sampling to ensure all relevant aspects are covered. This leads to better model performance and more reliable results.