The broad field of simulating human intelligence in machines. Instead of just following a rigid script of code, AI enables machines to learn from previous experiences, adjust to new inputs, and perform tasks that normally require human reasoning.

A specific type of AI that generates new content (text, images, code) based on patterns it learned from massive amounts of training data. It doesn’t just copy/paste; it predicts what word should come next to form a coherent sentence.

The next evolution beyond GenAI. While GenAI waits for you to ask it a question, Agentic AI can act independently. It can plan, reason, and execute multi-step processes to achieve a goal you gave it.

AI built specifically for high-stakes, secure environments. Unlike public consumer tools, professional-grade AI is walled off, guarantees data privacy, and is rigorously tested by legal experts to ensure it doesn’t invent case law.

The foundational algorithm that allows computers to recognize patterns in data over time without being explicitly programmed to do so. For decades, machines had to be taught everything. With ML, they learn through exposure.

A massive AI model trained on billions of pages of human language. By reading practically the entire internet, an LLM learns the grammar, context, and nuance of how humans speak and write.

The technology that allows machines to read, hear, and interpret human language as it is naturally spoken, rather than requiring users to type in rigid computer commands or exact-match keywords.

A training method where the AI learns by trial and error. The AI takes an action, and a human gives it a “reward” (a thumbs up) or a “penalty” (a thumbs down), helping the AI refine its behavior over time.

Data that has been reviewed, organized, and stripped of errors, duplicates, or outdated information. Bad data corrupts AI systems. Maintaining clean data requires strict “data governance.”

Structured data is neat and organized (like an Excel spreadsheet of billable hours). Unstructured data is messy and text-heavy (like a massive folder of PDFs, emails, and Word docs). AI excels at making sense of the unstructured data.

The actual instructions or questions you type into an AI tool. The quality of your prompt directly dictates the usefulness of the AI’s response.

A technique where an AI model is forced to go “read” a specific, trusted database (like Westlaw or your firm’s own server) before it answers your question. It retrieves facts, then generates the answer based only on those facts.

When an AI confidently generates false information. Because LLMs predict text rather than look up facts, they can invent convincing but entirely fake concepts.

Technology Assisted Review. An older, but vital machine learning process used in discovery where an algorithm categorizes documents based on a human’s initial coding.

The skill of crafting highly specific, structured inputs to get the exact desired output from an AI model.

A system design requiring human interaction to approve, reject, or modify an AI’s output before it is finalized or sent.

An AI system whose internal decision-making process is so complex that even its creators cannot easily explain exactly *how* it arrived at a specific conclusion.

The opposite of a Black Box. AI systems designed to provide a clear, human-readable audit trail of how and why they made a specific decision.

How an AI reads text. It breaks words down into chunks called “tokens.” A token is roughly 3/4 of a word. AI systems charge money based on how many tokens you use.

The maximum amount of text an AI can “hold in its head” at one time during a single conversation.

Tying an AI’s responses to a specific set of verified facts or documents to prevent it from guessing or hallucinating.

Taking a general AI model (like GPT-4) and giving it extra training on a highly specialized dataset so it becomes an expert in that niche.

Systematic and repeatable errors in a computer system that create unfair outcomes, usually because the human data it was trained on contained historical prejudices.

Synthetic media where a person in an existing image, audio, or video is replaced with someone else’s likeness using AI.

A specialized database that stores information not as text, but as mathematical coordinates, allowing the AI to understand the *meaning* and relationship between concepts.

Searching by the *intent* and contextual meaning of a phrase, rather than just looking for exact keyword matches.

An AI technique that scans text to locate and classify key nouns—like people, organizations, dates, and money.

Technology that converts different types of documents (scanned paper, PDFs, images) into editable, searchable text data.