AI Topics: the terms behind AI products, in plain English

The principles for building AI that is fair, transparent, safe and respectful of people's rights.

The policies, roles and controls that decide how AI is built, approved, monitored and held accountable.

Using AI to manage network traffic, energy and faults in 5G infrastructure.

Using models to detect threats, flag unusual behaviour and respond faster than rule-based tools.

Fraud detection, credit scoring, trading and customer service powered by models.

How clinics, payers and medtech use AI for diagnosis support, documentation, triage and operations.

Adding models to connected devices so they can detect, predict and act on their own data.

Predictive maintenance, quality inspection and production planning with models and sensors.

Personalisation, demand forecasting, pricing and store operations driven by AI.

Forecasting demand, routing, inventory and risk across suppliers and logistics.

A structured review that scores where a company sits on the maturity model and where to invest next.

A staged scale that describes how advanced an organisation's AI capability is and what the next stage needs.

The slow loss of accuracy when the real world changes but the model does not.

Watching a live model's accuracy, speed, cost and behaviour so problems are caught before users notice.

The founders, investors, accelerators and platforms that fund and grow AI companies.

A plan that links AI projects to business goals, data, people and risk, with a sequence and a budget.

Using AI to find, screen and schedule candidates, and the skills a company needs to hire for AI work.

Letting models recommend or make routine decisions from data, with people handling the exceptions.

Tailoring content, offers and experiences to each user in real time from their behaviour.

Reporting tools that use AI to surface trends, answer questions in plain language and forecast.

Designing products and operations around AI from the start rather than adding it later.

Spotting data points that do not fit the usual pattern, used for fraud, faults and security alerts.

A hypothetical AI that could learn and reason across any task at a human level, as opposed to today's narrow systems.

Artificial intelligence (AI) is software that performs tasks which normally need human judgement, such as recognising images or speech, understanding languag...

Software that holds a text conversation with users, from scripted menus to language-model assistants.

Managed services from AWS, Azure and Google Cloud for training, hosting and running AI models.

Updating a model from new data after launch so it keeps improving instead of drifting.

The technology that lets software understand and respond in natural dialogue, by text or voice.

Building the pipelines and storage that collect, clean and deliver data to analytics and AI systems.

Collecting and using data in ways that are lawful, fair and honest with the people it describes.

Tagging examples (this is a cat, this invoice is overdue) so a model can learn from them.

Storage that keeps raw data in any format until it is needed, cheaper and more flexible than a warehouse.

Turning model outputs and metrics into charts people can act on.

A central, structured store of business data organised for reporting and analysis.

Choosing a course of action from evidence and measurement rather than instinct alone.

Machine learning with many-layered neural networks, the technique behind image recognition, speech and large language models.

An open-weight family of language models from China known for low training cost and strong reasoning benchmarks.

Rebuilding processes and products around data and AI, not just moving them online.

A live virtual copy of a machine, building or process, fed by sensor data and used to simulate and predict.

Running models on the device or near it (a phone, a camera, a factory controller) instead of in a distant data centre.

Moving raw data into storage first and reshaping it there, the modern alternative to transforming before loading.

Methods that show why a model produced a result, needed wherever a decision has to be justified to a person or a regulator.

AI that creates new text, images, code or audio from a prompt rather than classifying existing data.

Offshore centres a multinational owns and runs itself for engineering, data and AI work.

Generative Pre-trained Transformer, the family of large language models that made chat-style AI mainstream.

The difference between general connected devices and the regulated medical devices that connect to clinical systems.

A model trained on vast amounts of text that can read, summarise, answer and write; the core of most generative AI products.

A way of building AI where a model learns patterns from data instead of following rules written by hand.

The branch of AI that reads, understands and writes human language, from search to chat.

NVIDIA's packaged microservices for running AI models on its hardware with a standard API.

A formal map of the concepts in a domain and how they relate, which helps AI reason and search consistently.

Writing and structuring instructions so a language model gives reliable, useful output.

Research into running machine learning on quantum computers for problems classical hardware struggles with.

Models that predict what a user will want next from past behaviour, the engine behind product and content feeds.

Giving a language model your own documents at answer time so it replies from your data instead of guessing.

How to measure the return on an AI investment: cost saved, revenue added and risk reduced against what it cost to build and run.

Working out what text means, not just which words it contains, so systems can match intent rather than keywords.

Reading text to judge whether its tone is positive, negative or neutral, at scale.

Models that turn speech into text and text into speech, and understand what was said.

Creating artificial data that behaves like the real thing, used when real data is scarce, private or sensitive.

Splitting text into the small units a language model actually reads, which is also how usage and cost are measured.

AI helpers that complete tasks on request, such as booking, answering account questions or drafting messages.

A vision of the internet built on blockchains, tokens and user-owned data.