Glossary

A helpful glossary of terms and guides to help you understand commonly used terminology and acronyms around AI technology

AI Agents – a system or program capable of autonomously performing tasks on behalf of a user or another system. Designing workflow and utilizing available tools, these can encompass multiple functionalities including decision-making, problem-solving, interacting with external environments and executing actions.^xiv

AIaMD (Artificial Intelligence as a Medical Device) – Artificial Intelligence as a Medical Device refers to AI-based software intended for medical purposes - such as diagnosis, treatment, monitoring, or prevention of disease - that is regulated as a medical device. In the UK, the product will only be classed as AIaMD when it meets the formal definition of a medical device based on its intended use and risk level.

Algorithm – A set of complex instructions and parameters used to perform tasks (such as calculations and data analysis) usually using a computer or another smart device.

Algorithmic bias – AI systems can have embedded bias which can manifest through various pathways including narrow training datasets or biased decisions made by humans in the design process.

Artificial intelligence (AI) – Basic definition from Oxford English Dictionary: “the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages”. To look at how this is translating in capitalist society, the UK Government’s 2023 policy paper on ‘A pro-innovation approach to AI regulation’^xv defined AI as “products and services that are ‘adaptable’ and ‘autonomous’.”

Artificial general intelligence – Sometimes known as general AI, strong AI or broad AI, this often refers to a theoretical form of AI that can achieve human-level or higher performance^xvi across most cognitive tasks. (Also termed as Superintelligence).

Artificial neural network – A computer structure inspired by the biological brain, consisting of numerous interconnected computational units (‘neurons’) connected in layers. Data flows between these units as between neurons in a human brain.^xvii Outputs of a previous layer are direct inputs for the next, and there can be hundreds of layers of units. An artificial neural network with more than three layers is considered a deep learning algorithm.^xviii

Automated decision-making – a process where a computer system makes decisions without human intervention, based on algorithms that analyse input data and apply predefined rules or learned patterns.

Deep learning – A subset of machine learning that uses artificial neural networks to recognise patterns in data and learn from them. The idea being to emulate the decision-making process of the human brain, fundamentally learning by example. Deep learning is increasingly used in complex learning tasks such as voice and image recognition, object detection and autonomous driving.^xix

Deepfakes – Images and video that are deliberately altered to appear genuine, often used to generate misinformation and disinformation. Advances in generative AI have made the production of deepfakes much easier.^xx

Foundation models – A machine learning model trained on a vast amount of data so that it can easily be adapted for a wide range of general tasks, including being able to generate outputs (generative AI).^xxi (See also large language models).

Frontier AI – Defined by the Government Office for Science as ‘highly capable general-purpose AI models that can perform a wide variety of tasks and match or exceed the capabilities present in today’s most advanced models’.^xxiiThe most widespread examples are the large language models including: ChatGPT (OpenAI); Claude (Anthropic); and Bard (Google).^xxiii

Generative AI – An AI model that generates text, images, audio, video or other media in response to user prompts.^xxivUsing machine learning techniques these models create new data synthesised from characteristics to the data it was trained on. Generative AI applications include chatbots, photo and video filters, and virtual assistants.

Hallucinations – When an AI model generates factually incorrect or nonsensical information that sounds plausible. To illustrate, Large Language models (LLMs), such as ChatGPT, often generate phrases that do not make sense as they are unable to contextualise or verify accuracy.

Interpretability – Some models are so complex that it may be impossible to know how the model produced the output.^xxvInterpretability is the level to which a machine learning system’s decision-making process can be explained in terms that can be understood by humans.^xxvi(Interpretability is also referred to as transparency or explainability).

Large language models (LLMs) – Machine learning models trained on large datasets that can recognise, understand, and generate text and other content. ChatGPT being the most obvious example.

Machine learning – A type of AI that works by finding patterns in existing data (known as ‘training data’) and using these patterns to inform the processing of new data to make predictions or generate other outputs. Training machine learning systems for specific applications can involve different forms of learning, such as supervised, unsupervised, semi-supervised and reinforcement learning.

Model – a program trained on large amounts of data that processes inputs and produces outputs based on learned patterns. LLMs are one type of AI model, but there are also models for image generation, speech recognition, and many other tasks. (see also, algorithm)

Narrow AI – Sometimes known as weak AI, these AI models are designed to perform a specific task (such as speech recognition) and cannot be adapted to other tasks.^xxvii

Natural language processing – Focuses on programming computer systems to understand and generate human speech and text. Algorithms are trained to find linguistic patterns in sentence and paragraph structures, words used and context to recreate and scribe speech. Applications include speech-to-text converters, online tools that summarise text, chatbots, speech recognition and translations.^xxviii

Open-source – Open-source often means the fundamental code used to run AI models is freely available for testing, scrutiny and further development.^xxix

Responsible AI – Refers to the practice of designing, developing, and deploying AI with certain values, such as being trustworthy, ethical, transparent, explainable, fair, robust and upholding privacy rights.^xxx

Robotics – Machines that are capable of automatically carrying out a series of actions and moving in the physical world.^xxxi Modern robots contain algorithms that typically, but do not always, have some form of artificial intelligence. Applications include medical robots for performing surgery.^xxxii

Superintelligence – A theoretical form of AI that has intelligence greater than humans and exceeds their cognitive performance in most domains.

Supervised learning – Supervised learning is a category of machine learning that uses labelled datasets to train algorithms to predict outcomes and recognize patterns. Unlike unsupervised learning, supervised learning algorithms are given labelled training to learn the relationship between the input and the outputs.^xxxiii

Training datasets – The set of data used to train an AI system. Can be labelled (for example, pictures of optic nerves in patients over 60 labelled accordingly) or unlabelled. Data used to train is crucial to a product being “fit for purpose” and scalable.

Unsupervised learning – an approach to training AI models where unlabelled data has been used to find hidden structures or patterns. Often utilised for things like customer clustering in marketing sectors.