IB Digital Society 3.6 Artificial Intelligence Key Definitions

The laws, rules, standards, and oversight that guide how AI is developed and used. Because it varies widely between countries and is often slower than the technology, governance of AI is uneven and incomplete.

A period when interest and funding in artificial intelligence collapse after inflated promises fail to materialise. AI has gone through several such winters followed by renewed booms.

The principle that the people and organisations who build and deploy an algorithm are answerable for its effects.

A systematic error in an algorithm that produces unfair outcomes, often by favouring one group over another.

A hypothetical AI that could understand, learn, and apply knowledge across any task as flexibly as a human. Also called strong or full AI, it does not yet exist.

A field of computer science dedicated to building systems that can replicate human intelligence and answer complex queries.

The basic unit of an artificial neural network, loosely modelled on a brain cell. It takes in numbers, combines them using weights, and passes on a signal if the combined value is strong enough.

The use of technology to perform tasks without human intervention.

A system whose internal workings cannot be easily inspected or explained. Many neural networks are black boxes: they can be accurate while no one can fully explain how they reached a particular output.

The field of AI that enables machines to interpret images and video, such as recognising objects, faces, or medical features, often using deep neural networks.

The mid-twentieth-century study of control and communication in animals and machines, especially feedback loops. It helped lay the groundwork for artificial intelligence by treating thinking as information processing.

A systematic skew in a dataset, from how it was collected or what it left out, that makes any system built on it treat some groups unfairly.

A form of machine learning that uses artificial neural networks with many layers to learn complex patterns directly from large amounts of raw data, powering tasks such as image and speech recognition.

A machine learning application that identifies or verifies a person from an image or video by comparing patterns in their facial features against stored data.

AI that produces new text, images, audio, or code by learning patterns from vast training data. It is a powerful form of narrow AI, not a general intelligence, despite how capable it can seem.

A subset of artificial intelligence that enables systems to learn from data and improve their performance over time without explicit programming.

The observation, made by Gordon Moore in 1965, that the number of transistors on a chip roughly doubles about every two years, driving rapid growth in computing power.

AI built to perform one specific task or a narrow set of tasks, such as recommending videos or recognising speech; it does not understand or think beyond what it was trained for. All AI in use today is narrow AI.

Computational models inspired by the human brain that use neurons to learn from data to perform tasks like classification, prediction, and pattern recognition.

A type of machine learning where an agent learns by trial and error in an environment, taking actions and adjusting its behaviour to maximise a reward signal over time.

A speculative AI that would surpass human intelligence across virtually every field. It remains hypothetical and is central to debates about the long-term risks and future of AI.

A type of machine learning where a model is trained on labelled data, meaning each example is tagged with the correct answer, so it learns to predict the label for new, unseen inputs.

A hypothetical future point at which AI becomes able to improve itself so rapidly that technological change accelerates beyond human control or understanding. It is speculative and heavily debated.

The set of examples used to teach a machine learning model. The quality, quantity, and representativeness of the training data shape what the model learns and any bias it inherits.