AI is reshaping how legal professionals research, draft, review, and manage work. With new technology, comes new technical jargon. This glossary breaks down the key terms, so you can talk about AI with confidence.
A
- Agents – A software system that can perceive its environment, make decisions, and take actions to achieve specific goals—often autonomously. AI agents can use tools, respond to user input, and adapt based on context or feedback.
- AI Governance – The frameworks, policies, and controls that organizations use to manage AI responsibly.
- Alignment – Making sure AI systems behave in ways that match human goals and values. Since AI models don’t always follow instructions reliably, alignment is about making their responses safe, consistent, and predictable in real-world use.
- Algorithm – A set of rules or instructions that a computer follows to perform a task or solve a problem.
- Artificial Intelligence (AI) – Technology that simulates human intelligence to perform tasks like reasoning, learning, and decision-making.
B
- Bias – LLMs form biases from their training data, leading to potentially unfair outcomes.
C
- Classification – Teaching AI to group data into categories (e.g., “privileged vs. non-privileged” documents).
- Context Limits – The maximum amount of text that can be input into an AI model
D
- Deep Learning – A field of machine learning that uses multi-layered neural networks to learn complex representations of data, enabling the model to discover patterns. Deep learning is often used for tasks such as image recognition, and natural language processing/generation.
E
- Embedding – A way of turning words, sentences, or other data into numbers that capture meaning and similarity.
- Entity Recognition – The AI’s ability to identify key terms like names, dates, or clauses in a contract.
- Explainability – The ability to understand and justify how an AI reached its conclusion.
F
- Fine Tuning – Adjusting pre-trained AI models to a specific task or subject using task-specific data to improve performance and adapt to narrower subjects.
- Frontier AI – the most advanced artificial intelligence systems at the cutting edge of current technology. These are large-scale foundation models trained with massive computing resources on broad datasets, designed to perform a wide range of tasks rather than a single function. Because of their scale and adaptability, frontier AI models have the potential to deliver powerful new capabilities, but also introduce higher levels of risk
G
- Generative AI – A type of AI that can create new content—such as text, images, or audio—based on patterns it has learned from large amounts of data. For example, drafting a contract clause or generating a summary of case law.
- Grounding – Connecting an AI’s answers to trusted sources, such as a database or document, so that the response is more accurate and reduce hallucinations.
H
- Hallucination – When an AI system generates information that sounds convincing but is actually wrong or made up. For example, citing a legal case that doesn’t exist.
I
- Inference – The process of generating an output from an AI model after it’s been trained.
L
- Large Language Model (LLM) – An artificial intelligence model trained on extensive datasets, able to understand and generate human-like language for a range of natural language processing tasks.
M
- Machine Learning (ML) – Learns from data & improves without being programmed.
- Model – Algorithm designed to perform specific tasks, make predictions or solve problems by learning patterns from data
- Model Drift – Over time, AI performance can decline as data or conditions change. Regular monitoring is required.
N
- Natural Language Generation (NLG) – A field of natural language processing that involves the generation of human-like language by AI models. NLG can include text summarization and chatbot responses.
- Natural Language Processing (NLP) – A field of artificial intelligence that focuses on enabling models to understand, interpret, and generate human language. NLP can include translation, sentiment analysis, and speech recognition.
- Neural Network – A model inspired by the human brain, made up of interconnected nodes (“neurons”) that help AI process data and recognize patterns.
O
- Overfitting – When a model learns the training data too well, including noise, and performs poorly on new data.
P
- Parameters – The internal settings a model adjusts during training (often billions in modern LLMs).
- Predictive Analytics – Using AI to forecast outcomes (e.g., case results, contract risk) from historical data.
- Prompt – Providing specific input/queries to an AI model to shape the nature and content of the output
- Prompt Engineering – The art of crafting effective prompts to get better outputs.
R
- Reinforcement Learning – A training method where AI learns by trial and error, receiving rewards or penalties for its actions.
S
- Supervised Learning – Trains AI models using labeled data, where the AI learns to map inputs to corresponding outputs
- Semi-supervised Learning – Combining labeled and unlabeled data to guide training and improve understanding of patterns and relationships within the data
- System Prompt – The initial input or instruction given to a language model that dictates output and performance of tasks, guiding subsequent responses based on the provided context.
T
- Token – A chunk of text (often a few characters or words) that AI models process. Costs and limits are usually based on tokens.
- Training – Teaching AI models to perform specific tasks through exposure to relevant data
- Training Data – The data the AI is fed to learn language and patterns.
U
- Unsupervised Learning – Trains AI models on unlabeled data to uncover inherent patterns, often used for anomaly detection
V
- Validation – The process of assessing the performance, accuracy and reliability of an AI model
Z
- Zero-Shot / Few-Shot Learning – An AI’s ability to complete tasks without (zero) or with minimal (few) examples.
The Gist of It
This AI glossary for legal professionals is a living document. New terms will be added as they surface in community conversations and at events.
