AI Glossary & Resource Hub

A pattern in which AI systems autonomously plan, execute, and adapt multi-step tasks with minimal human intervention. Rather than responding to a single prompt, an AI agent breaks a goal into sub-tasks, calls external tools or APIs, evaluates intermediate results, and iterates until the objective is met.

Software systems that use large language models or other AI techniques to perceive their environment, make decisions, and take actions toward a defined goal. Unlike simple chatbots, agents can use tools, access databases, call APIs, and chain multiple reasoning steps together autonomously.

Australia's voluntary AI Ethics Framework, published by the Department of Industry, Science and Resources, outlines eight principles for responsible AI: human, societal and environmental wellbeing; human-centred values; fairness; privacy protection and security; reliability and safety; transparency and explainability; contestability; and accountability.

A measure of how prepared an organisation is to successfully adopt and scale artificial intelligence. AI readiness spans data quality and accessibility, technology infrastructure, talent and skills, governance frameworks, leadership commitment, and organisational culture.

A structured plan that defines how an organisation will adopt AI to achieve its business objectives. A sound AI strategy covers use case prioritisation, data foundations, technology choices, talent development, governance, and a phased roadmap with measurable outcomes.

Prudential Standard CPS 234 (Information Security) is issued by the Australian Prudential Regulation Authority. It requires APRA-regulated entities (banks, insurers, superannuation funds) to maintain information security capabilities commensurate with the threats they face, including those arising from AI and data platforms.

The broad field of computer science concerned with building systems capable of performing tasks that typically require human intelligence. This includes learning from data, recognising patterns, understanding language, making decisions, and generating content. Modern AI spans narrow applications (e.g. fraud detection) through to general-purpose systems such as large language models.

The thirteen Australian Privacy Principles (APPs) form the cornerstone of the Privacy Act 1988. They govern how organisations collect, use, store, disclose, and provide access to personal information. The APPs apply to any AI system that processes personal data.

Automated Machine Learning (AutoML) refers to tools and techniques that automate the process of building machine learning models. This includes automated feature engineering, model selection, hyperparameter tuning, and evaluation. AutoML lowers the barrier to entry for ML by reducing the need for deep data science expertise.

A branch of AI that enables machines to interpret and understand visual information from images and video. Applications include object detection, image classification, facial recognition, optical character recognition (OCR), and quality inspection in manufacturing.

The set of policies, processes, standards, and metrics that ensure data is managed as a strategic asset. Data governance covers data quality, access control, privacy, lineage, classification, and lifecycle management. Without governance, AI projects cannot move safely from prototype to production.

A centralised repository that stores large volumes of raw data in its native format, whether structured, semi-structured, or unstructured. Data lakes offer low-cost, scalable storage and are well suited for exploratory analytics and machine learning workloads.

A modern data architecture that combines the low-cost, flexible storage of a data lake with the performance, reliability, and governance features of a data warehouse. The lakehouse supports both business intelligence and AI/ML workloads on a single platform, eliminating the need for separate systems.

An organisational and architectural approach that treats data as a product, owned and managed by domain teams rather than a centralised data team. Data mesh principles include domain ownership, data as a product, self-serve data infrastructure, and federated computational governance.

An automated series of processes that move and transform data from source systems to a destination such as a data lake, warehouse, or lakehouse. Pipelines handle extraction, validation, transformation, enrichment, and loading of data, ensuring it arrives in the right format for analytics and AI.

A centralised repository of structured, cleaned, and organised data optimised for fast analytical queries. Data warehouses use schema-on-write, meaning data must be structured before it is stored. They excel at business intelligence but are typically expensive to scale and poorly suited to unstructured data or ML workloads.

A unified data and AI platform built on Apache Spark that provides a lakehouse architecture for analytics, data engineering, data science, and machine learning. Databricks combines a collaborative workspace, managed infrastructure, and governance tools (Unity Catalog) to help enterprises consolidate their data stack and accelerate AI adoption.

A subset of machine learning that uses multi-layered neural networks to learn complex patterns from large volumes of data. Deep learning powers many modern AI breakthroughs including image recognition, speech understanding, natural language processing, and generative AI.

The process of fundamentally changing how an organisation operates and delivers value by adopting digital technologies. In the AI era, digital transformation increasingly centres on data modernisation, automation, and embedding intelligence into core business processes.

Numerical vector representations of data (text, images, or other content) that capture semantic meaning in a format AI models can process. Similar items produce similar embeddings, enabling tasks such as semantic search, recommendations, and clustering without explicit rules.

ETL (Extract, Transform, Load) and ELT (Extract, Load, Transform) are two approaches to moving data into analytical systems. ETL transforms data before loading it into the destination; ELT loads raw data first and transforms it in place. ELT has become the dominant pattern in cloud and lakehouse architectures because it preserves raw data and leverages the processing power of the destination platform.

A centralised repository for storing, managing, and serving the engineered features (input variables) used by machine learning models. Feature stores ensure consistency between training and production environments, reduce duplicated effort, and improve model reliability.

The process of taking a pre-trained AI model and further training it on a smaller, domain-specific dataset to improve its performance on particular tasks. Fine-tuning allows organisations to adapt general-purpose models to their industry terminology, internal processes, and specific use cases without training a model from scratch.

A phenomenon where an AI model generates output that sounds plausible but is factually incorrect, fabricated, or inconsistent with its source material. Hallucinations are a well-known limitation of large language models and represent a significant risk in enterprise applications where accuracy is critical.

The Information Security Registered Assessors Program (IRAP) is an Australian Signals Directorate (ASD) initiative that provides a framework for assessing the security of systems against the Australian Government Information Security Manual (ISM). IRAP assessments are required for cloud services used by Australian government agencies.

A type of AI model trained on vast quantities of text data that can understand, generate, summarise, and reason about natural language. LLMs such as GPT-4, Claude, and Llama power applications ranging from conversational assistants and content generation through to code writing and complex analysis.

A subset of artificial intelligence in which systems learn patterns and make predictions or decisions from data without being explicitly programmed for each scenario. Machine learning encompasses supervised learning (from labelled examples), unsupervised learning (finding hidden patterns), and reinforcement learning (learning through feedback).

The simplest version of an AI solution that can be deployed to validate a use case and deliver initial business value. An AI MVP focuses on a narrow, well-defined problem, uses available data, and prioritises learning and iteration over feature completeness.

Machine Learning Operations (MLOps) is the discipline of deploying, monitoring, and managing machine learning models in production. MLOps borrows principles from DevOps and applies them to the ML lifecycle, covering experiment tracking, model versioning, automated testing, deployment pipelines, and performance monitoring.

A broader discipline than MLOps that encompasses the operationalisation of all types of AI and analytical models, including ML models, rules-based models, and decision models. ModelOps focuses on governance, lifecycle management, and performance monitoring across the full model portfolio.

The My Health Records Act 2012 governs the Australian digital health record system and sets strict rules around the collection, use, and disclosure of health information. Any AI system that accesses or processes My Health Record data must comply with the Act's specific privacy and security requirements.

A branch of AI focused on enabling computers to understand, interpret, and generate human language. NLP underpins applications such as chatbots, sentiment analysis, document summarisation, translation, and search. Modern NLP is largely powered by transformer-based large language models.

A computing architecture inspired by the structure of the human brain, consisting of layers of interconnected nodes (neurons) that process information. Neural networks learn by adjusting the strength of connections between nodes based on training data. They are the foundation of deep learning and modern AI.

The Notifiable Data Breaches (NDB) scheme, part of the Privacy Act 1988, requires organisations to notify affected individuals and the Office of the Australian Information Commissioner (OAIC) when a data breach is likely to result in serious harm. The scheme applies to any breach involving personal information, including breaches originating from AI systems.

Australia's primary federal privacy legislation, governing how organisations and government agencies handle personal information. The Act includes the thirteen Australian Privacy Principles (APPs) and applies to any AI system that collects, stores, uses, or discloses personal data.

The practice of designing and refining the instructions (prompts) given to large language models to elicit accurate, useful, and consistent outputs. Effective prompt engineering involves clear task specification, providing context and examples, setting constraints, and iterative testing.

A small-scale, time-boxed project designed to demonstrate the feasibility and potential value of an AI use case before committing to full implementation. A well-structured PoC defines clear success criteria, uses representative data, and produces measurable results that inform the business case for scaling.

An AI architecture that enhances large language model responses by first retrieving relevant information from an external knowledge base, then using that information to generate grounded, accurate answers. RAG reduces hallucinations and enables AI systems to work with current, organisation-specific data without retraining the underlying model.

A type of machine learning where an agent learns to make decisions by taking actions in an environment and receiving rewards or penalties based on outcomes. The agent improves its strategy over time by maximising cumulative reward. Reinforcement learning powers applications such as robotics, game playing, and recommendation optimisation.

An approach to developing and deploying AI systems that prioritises fairness, transparency, accountability, privacy, and safety. Responsible AI practices include bias testing, explainability, human oversight, impact assessments, and ongoing monitoring of AI system behaviour.

Return on Investment for AI initiatives, measured across direct cost savings, revenue uplift, risk reduction, efficiency gains, and strategic value creation. AI ROI often follows a J-curve pattern: upfront investment in data foundations and infrastructure precedes accelerating returns as AI capabilities scale across the organisation.

The basic units of text that large language models process. A token can be a word, part of a word, or a punctuation mark. Tokenisation varies by model, but as a rough guide, one token is approximately three-quarters of a word in English. Token counts determine model context windows (how much text can be processed at once) and directly affect compute costs.

The complete cost of implementing and operating an AI or data platform over its lifecycle, including licensing, infrastructure, implementation services, training, ongoing operations, and opportunity costs. TCO analysis helps organisations make informed platform decisions and avoid hidden costs that emerge after deployment.

A machine learning technique where a model trained on one task is reused as the starting point for a different but related task. Transfer learning dramatically reduces the data and compute required to build effective models, because the pre-trained model has already learned general patterns that transfer to new domains.

A neural network architecture that uses a mechanism called self-attention to process input data in parallel rather than sequentially. Transformers are the foundation of modern large language models and have revolutionised natural language processing, computer vision, and generative AI since their introduction in 2017.

Databricks' unified governance solution for data and AI assets. Unity Catalog provides centralised access control, audit logging, data lineage, data discovery, and fine-grained security (row-level, column-level, and dynamic data masking) across all Databricks workspaces and cloud environments.

A specialised database designed to store, index, and query high-dimensional vector embeddings efficiently. Vector databases enable similarity search at scale, powering applications such as semantic search, recommendation engines, and RAG-based AI systems.