AI Glossary

An AI system that can autonomously plan, make decisions, and take actions to achieve a goal with minimal human intervention. Agents can use tools, browse the web, write code, and chain multiple steps together. They represent the next evolution beyond simple chatbots.

A set of step-by-step instructions or rules that a computer follows to solve a problem or complete a task. In AI, algorithms are the mathematical recipes that allow systems to learn patterns from data. Different algorithms are suited to different types of problems.

The broad field of computer science focused on creating systems capable of performing tasks that typically require human intelligence. This includes learning from experience, understanding language, recognising patterns, and making decisions. Modern AI ranges from narrow task-specific tools to increasingly general-purpose systems.

Search technology that uses AI, particularly embeddings and language models, to understand the intent and meaning behind queries rather than relying solely on keyword matching. It can search across documents, databases, and websites to find the most relevant results. AI search understands synonyms, context, and even questions phrased in natural language.

A structured plan that defines how an organisation will adopt and leverage AI to achieve its business objectives. A good AI strategy identifies high-impact use cases, assesses data readiness, outlines required investments, and establishes governance frameworks. It aligns AI initiatives with broader business goals rather than adopting technology for its own sake.

Application Programming Interface. A set of rules and protocols that allows different software applications to communicate with each other. In the AI context, APIs are how businesses connect to AI services like ChatGPT, Claude, or custom models. You send data to an API endpoint and receive AI-generated results back.

The field of study and practice concerned with ensuring AI systems are developed and used in ways that are fair, transparent, safe, and beneficial to society. AI ethics covers issues like bias, privacy, accountability, and the societal impact of automation. It is increasingly important as AI becomes embedded in high-stakes decision-making.

The frameworks, policies, and processes that organisations put in place to manage the development, deployment, and monitoring of AI systems responsibly. AI governance covers risk assessment, compliance, model oversight, data management, and accountability structures. It ensures AI is used consistently with organisational values and regulatory requirements.

The use of AI-powered platforms to create automated multi-step workflows that connect different applications and services. Tools like Make, Zapier, and n8n allow users to build complex automations that include AI processing steps such as text analysis, document generation, and decision-making. These workflows run automatically based on triggers.

The use of technology, increasingly powered by AI, to automate repetitive business processes and workflows. Unlike simple automation that follows rigid rules, AI-powered automation can handle unstructured data, make judgement calls, and adapt to variations. It spans everything from invoice processing to employee onboarding.

Systematic errors in AI systems that produce unfair outcomes for certain groups of people. Bias can enter AI through skewed training data, flawed algorithm design, or biased human decisions embedded in historical data. It can result in discriminatory outcomes in hiring, lending, policing, and healthcare.

A branch of AI that enables computers to interpret and understand visual information from images, videos, and live camera feeds. It can identify objects, read text, detect defects, and track movement. Computer vision has become increasingly accurate thanks to deep learning advances.

A type of machine learning task where the model learns to assign data into predefined categories or classes. For example, an email is classified as spam or not spam. Classification models learn from labelled examples and then predict the category of new, unseen data.

A software application that simulates human conversation through text or voice. Modern AI chatbots powered by large language models can understand context, handle complex queries, and maintain natural conversations. They range from simple FAQ bots to sophisticated virtual assistants capable of completing transactions.

The practice of using AI and data analysis to divide a customer base into distinct groups based on shared characteristics, behaviours, or preferences. AI-powered segmentation goes beyond basic demographics to identify nuanced behavioural patterns. It enables targeted marketing, personalised experiences, and better resource allocation.

The delivery of computing resources including servers, storage, databases, and AI services over the internet on a pay-as-you-go basis. Cloud platforms like AWS, Google Cloud, and Microsoft Azure provide the infrastructure needed to train and deploy AI models without owning physical hardware. Most modern AI tools run in the cloud.

An AI chatbot developed by OpenAI that uses large language models (GPT-4 and beyond) to engage in conversational interactions. ChatGPT can answer questions, write content, analyse data, generate code, and assist with a wide range of tasks. It popularised generative AI for mainstream users when it launched in late 2022.

An AI assistant developed by Anthropic, designed with a focus on being helpful, harmless, and honest. Claude excels at long-form analysis, careful reasoning, coding, and working with large documents. It is available through a web interface, mobile apps, and an API, and is known for nuanced, thoughtful responses.

A subset of machine learning that uses artificial neural networks with many layers to learn complex patterns from large amounts of data. Deep learning powers most modern AI breakthroughs including image recognition, language translation, and generative AI. It requires significant computing power but can achieve remarkable accuracy.

The process of annotating data with meaningful tags or labels so that machine learning models can learn from it. For example, labelling images of products as "defective" or "acceptable", or tagging customer reviews with sentiment. Data labelling is often the most labour-intensive step in building supervised learning models.

An automated series of steps that collect, process, transform, and deliver data from source systems to destinations where it can be analysed or used by AI models. Data pipelines ensure that information flows reliably and consistently. They are essential infrastructure for any organisation that wants to use AI at scale.

The protection of personal and sensitive information used in AI systems, ensuring it is collected, stored, processed, and shared in compliance with privacy laws and individual consent. In New Zealand, the Privacy Act 2020 governs how organisations handle personal data. AI raises new privacy challenges around data used for training and inference.

A numerical representation of text, images, or other data in a format that captures semantic meaning. Words or phrases with similar meanings have similar embeddings, allowing AI to understand relationships between concepts. Embeddings are the foundation of modern search, recommendation, and retrieval systems.

Processing data and running AI models locally on devices or nearby servers rather than sending everything to the cloud. Edge computing reduces latency, improves privacy, and works in environments with limited internet connectivity. It is increasingly important for real-time AI applications in manufacturing, agriculture, and autonomous vehicles.

The ability to understand and explain how an AI system arrives at its decisions or predictions. Explainable AI (XAI) provides transparency into the reasoning process, making it possible for humans to verify, trust, and challenge AI outputs. It is particularly important in regulated industries like finance and healthcare.

A large AI model trained on broad, diverse data that can be adapted to a wide range of downstream tasks. Examples include GPT-4, Claude, and Gemini. These models learn general capabilities during pre-training and can then be fine-tuned or prompted for specific uses without building a model from scratch.

The process of taking a pre-trained AI model and further training it on a smaller, domain-specific dataset to improve its performance on a particular task. Fine-tuning preserves the general knowledge the model already has while adding specialised expertise. It is far cheaper and faster than training a model from scratch.

The process of selecting, transforming, and creating input variables (features) from raw data to improve the performance of machine learning models. Good feature engineering often has a bigger impact on model accuracy than the choice of algorithm. It requires both domain expertise and data science skills.

AI systems that can create new content such as text, images, audio, video, and code based on patterns learned from training data. Generative AI has surged in popularity since 2022 with tools like ChatGPT, DALL-E, and Midjourney. It works by predicting what should come next based on a given prompt.

A type of AI model trained on vast amounts of text data that can understand, generate, and reason about human language. LLMs like GPT-4, Claude, and Gemini contain billions of parameters and can perform a wide range of language tasks from writing to coding to analysis. They are the technology behind modern AI chatbots.

An open-source framework for building applications powered by large language models. LangChain provides tools and abstractions for connecting LLMs to external data sources, chaining multiple AI steps together, and building AI agents. It has become one of the most popular frameworks for developers building custom AI applications.

A subset of AI where systems learn and improve from experience without being explicitly programmed for every scenario. Instead of following hard-coded rules, machine learning algorithms find patterns in data and use those patterns to make predictions or decisions. It is the engine behind most practical AI applications today.

AI systems that can process and generate multiple types of data, such as text, images, audio, and video, within a single model. Rather than needing separate systems for each data type, multimodal AI understands context across formats. This makes interactions more natural and capable.

In machine learning, a model is the mathematical representation that an algorithm produces after being trained on data. It captures the patterns and relationships found in the training data and can then make predictions on new inputs. Think of it as the "trained brain" that results from the learning process.

An open protocol developed by Anthropic that standardises how AI models connect to external tools, data sources, and services. MCP provides a universal interface so that AI assistants can interact with databases, APIs, file systems, and other tools in a consistent way. It is designed to make AI integrations more interoperable and maintainable.

A computing system loosely inspired by the structure of the human brain, consisting of interconnected nodes (neurons) organised in layers. Data flows through these layers, with each layer extracting increasingly abstract features. Neural networks are the foundation of most modern AI, especially deep learning.

An NLP technique that automatically identifies and classifies named entities in text, such as people, organisations, locations, dates, and monetary values. It allows AI systems to extract structured information from unstructured text documents. NER is a foundational building block for many document processing workflows.

The branch of AI that deals with enabling computers to understand, interpret, and generate human language. NLP covers everything from basic text analysis to complex conversational AI. It bridges the gap between how humans communicate and how computers process information.

Platforms and tools that allow users to build, train, and deploy AI solutions without writing any code. No-code AI tools provide visual interfaces, drag-and-drop builders, and pre-built templates that make AI accessible to non-technical users. They are democratising AI adoption across organisations of all sizes.

A common problem where a machine learning model learns the training data too well, including its noise and outliers, and performs poorly on new, unseen data. An overfit model essentially memorises rather than generalises. It is a key challenge in building reliable AI systems.

The use of AI to tailor content, products, recommendations, and experiences to individual users based on their behaviour, preferences, and context. AI-driven personalisation operates in real time and can adapt across channels. It goes far beyond simple name insertion to dynamically adjust entire user experiences.

The use of statistical techniques and machine learning to analyse historical data and predict future outcomes. Predictive analytics identifies trends, forecasts demand, and estimates probabilities of events occurring. It transforms raw data into forward-looking business intelligence.

The practice of crafting effective instructions and inputs for AI models to get the best possible outputs. Good prompt engineering involves being specific, providing context, giving examples, and structuring requests in ways that guide the model towards the desired result. It is an essential skill for getting value from any AI tool.

A type of machine learning task where the model predicts a continuous numerical value rather than a category. For example, predicting a house price, future sales revenue, or temperature. Regression models learn the mathematical relationship between input variables and the target number.

A type of machine learning where an agent learns by interacting with an environment and receiving rewards or penalties for its actions. Over time, the agent discovers the best strategy to maximise its cumulative reward. It is the technique behind game-playing AI and increasingly used in robotics and optimisation.

An AI system that suggests relevant items, content, or actions to users based on their behaviour, preferences, and similarities to other users. Recommendation systems power the "you might also like" features on streaming platforms, online stores, and news sites. They use techniques like collaborative filtering and content-based filtering.

Software robots that mimic human interactions with digital systems to automate repetitive, rule-based tasks. RPA bots can click buttons, fill forms, copy data between systems, and follow predefined workflows. When combined with AI, RPA becomes "intelligent automation" capable of handling more complex, judgement-based tasks.

A technique that enhances AI language models by first retrieving relevant information from a knowledge base, then using that information to generate more accurate and grounded responses. RAG combines the strengths of search and generation to reduce hallucinations and keep answers factual. It is the most popular way to connect LLMs to private business data.

An approach to developing and deploying AI that prioritises safety, fairness, transparency, privacy, and human oversight throughout the entire AI lifecycle. Responsible AI goes beyond just ethics to include practical measures like testing, monitoring, incident response, and ongoing model evaluation. It is a holistic framework for trustworthy AI.

A machine learning approach where the model is trained on a labelled dataset, meaning each training example includes the correct answer. The model learns to map inputs to outputs by studying these examples. It is the most common type of machine learning used in business applications today.

An NLP technique that determines the emotional tone or opinion expressed in a piece of text, typically classifying it as positive, negative, or neutral. Advanced sentiment analysis can detect more nuanced emotions like frustration, excitement, or sarcasm. It is widely used for monitoring brand perception.

The AI capability of converting spoken language into written text. Modern speech recognition systems use deep learning to achieve near-human accuracy across multiple languages and accents. It enables voice-controlled interfaces, transcription services, and hands-free computing.

The dataset used to teach a machine learning model. The quality, size, and diversity of training data directly determine how well the model performs. Biased or incomplete training data leads to biased or inaccurate models. Curating good training data is often the most time-consuming part of an AI project.

A technique where a model trained on one task is reused as the starting point for a different but related task. Instead of learning from scratch, the model transfers its existing knowledge, requiring far less data and computing power for the new task. This is what makes fine-tuning possible.

The AI capability of producing human-like written text based on a given prompt or context. Modern text generation models can write articles, emails, code, poetry, and more in a variety of styles and tones. The quality has improved dramatically with large language models.

The process of breaking text into smaller units called tokens, which can be words, subwords, or characters. Tokenisation is the first step in how language models process text. The way text is tokenised affects model performance, cost (as pricing is often per-token), and the amount of text that can fit within a model's context window.

A machine learning approach where the model discovers patterns and structures in data without labelled examples or a predefined correct answer. Common techniques include clustering similar items together and reducing data complexity. It is useful for exploration and finding hidden groupings.

An AI-powered software agent that can understand natural language instructions and perform tasks on behalf of a user. Virtual assistants go beyond chatbots by taking actions such as scheduling meetings, sending emails, searching databases, and generating reports. They are becoming increasingly capable as AI agent technology matures.

A specialised database designed to store and efficiently search high-dimensional vector embeddings. When text, images, or other data are converted into embeddings, a vector database can quickly find the most similar items. It is the backbone of semantic search and RAG systems.