AI Glossary
Cut through the hype. Plain English definitions of the AI terms you'll actually encounter as a business leader.
Advanced Concepts
An AI system that can choose steps and use available tools to work towards a goal, rather than only answer a question. Its actions depend on its permissions and design. Set limits, monitor results and require human approval for consequential actions.
Skills & Capability
The ability to work effectively and responsibly with AI: choose suitable tasks, give clear instructions, judge results and take responsibility for their use. It develops through practice and judgement, not simply knowing tool names.
Ethics & Governance
Responsibilities, rules and checks for how an organisation chooses, uses and monitors AI. It covers permitted data, approvals, output review, incident handling and accountability. It is an ongoing practice, not just a policy document.
Companies
The company that develops Claude AI models and assistant products and conducts AI safety research. Product features and data handling depend on the service, plan and settings.
Technical Concepts
Application Programming Interface: a defined way for software to request data or actions from another system. An API can connect business tools to an AI service. Permissions, security, usage limits and costs still need managing.
Core Concepts
Software that performs tasks such as recognising patterns, generating content or making predictions. AI can support human judgement or automate selected tasks, but its outputs still need appropriate checks.
Ethics & Governance
Systematic tendencies in an AI system's outputs that can lead to unfair treatment. They can arise from data, design choices or how the system is used, not just training data. Check effects on the people affected.
Tools & Platforms
OpenAI's conversational AI assistant. It can help with writing, questions, files and other tasks, depending on the available tools and account settings. ChatGPT is a product that can use different models, not another name for GPT-4.
Tools & Platforms
The name of Anthropic's AI assistant and model family. Claude can help with writing, analysis and coding. Available tools and data handling depend on the product, plan and settings; important outputs still need checking.
Technical Concepts
The amount of information a model can work with in one interaction, usually measured in tokens. It can include instructions, messages and supplied material. It is not the same as long-term memory or a promise to recall every detail.
Tools & Platforms
A name used for several AI assistants, not a single model. Microsoft 365 Copilot supports work in Microsoft 365, while GitHub Copilot supports software development. Features, licences and data protections differ, so check which product you are using.
Ethics & Governance
Handling information about people lawfully, fairly and securely. Before entering personal data into an AI tool, check the purpose, permissions, contract, settings and retention arrangements. A paid account does not automatically make every use appropriate.
Technical Concepts
Further training a pre-trained model using selected data to change how it performs or replies. It is different from adding a file to a chat or retrieving documents. Evaluate the adapted model before using it for a business task.
Tools & Platforms
Google's family of AI models and the name used for its AI assistant apps. Gemini features also appear in Google Workspace. Availability and data handling depend on the service, account and settings.
Core Concepts
AI that generates content such as text, images, audio, video or code using patterns learned during training. Generated content is not proof of accuracy, originality or permission to use it.
Models
An earlier OpenAI model introduced in March 2023 that accepts text and image inputs and generates text. GPT-4 is a model name, not a ChatGPT subscription. Check current product information rather than assuming it is the model available in ChatGPT.
Limitations & Risks
Plausible-sounding but false or invented AI output, such as a made-up quotation or source. It can occur even when an answer sounds confident. Verify important claims using dependable sources.
Core Concepts
An AI model trained on large amounts of data to work with language, including generating and summarising text. Apps such as ChatGPT use language models; the app and the model are not the same thing. Fluent output can still be wrong.
Core Concepts
A type of AI in which models learn patterns from examples rather than relying only on handwritten rules. For example, a model trained on labelled customer emails might suggest categories for new emails. Results depend on the data and testing.
Advanced Concepts
AI that can work with more than one kind of information, such as text and images. Depending on the model, it may also process or generate audio, video or other outputs. Not every multimodal system supports every format.
Core Concepts
A field of AI concerned with processing and generating human language. Examples include translation, sentiment analysis and finding information in text. NLP includes methods beyond large language models.
Companies
The organisation that develops ChatGPT and AI models including the GPT family. The ChatGPT app and OpenAI API are different services; features, charges and data handling depend on the service and plan.
Using AI
Instructions, questions or material you give an AI system for a task. Useful prompts state the goal, relevant context and required output. Share only information you are authorised to use.
Using AI
Designing and testing instructions and examples to guide an AI model towards a useful result. Iteration and checking matter: a better prompt does not guarantee a factual or safe answer.
Ethics & Governance
Designing, choosing and using AI with attention to safety, fairness, privacy and accountability. Assign human responsibility, check results and monitor how the system affects people.
Business Value
The financial gain from an investment after costs, expressed as a percentage of those costs. For AI, count setup, licences, training, checking and running costs. Time released is not automatically a cash saving.
Technical Concepts
A unit a model uses to process information. In text, a token may be a word, part of a word or punctuation. Token counts vary by language and model, and affect input limits and API costs. Tokens are not a fixed word count.
Technical Concepts
Examples or other information used to teach a model patterns and behaviour. Errors and gaps in that data can affect results. Uploading a document for a task is not itself model training; whether input is reused for training depends on the service, terms and settings.
Business Value
A specific task or business problem where you are considering using AI, such as summarising meeting notes. Define the intended outcome, necessary data, checks and responsible person before choosing a tool. Suitability and value need testing.
Business Value
Connecting steps in a process so software carries out agreed actions. It can use fixed rules, AI or both; not every automation needs AI. Uncertain or high-impact steps may still need human review.