A glossary of useful terms
WHEN YOU READ ABOUT AI, a lot of complicated terms are thrown around with the assumption everyone knows what’s being described. The reality is, unless you’re involved in the industry, most people could do with a refresh
of key terms:
Algorithm: A sequence of instructions or rules which tells a computer what to do, whether that’s completing a task or solving a problem. A number of logical operations can be chained together to achieve complicated tasks,
calculations, processing or reasoning.
Decision tree: A branching method for deriving strategies and reaching goals. It will illustrate all possible outcomes and assign them values so that decisions can become automated.
Deep learning: Often described as the next frontier of machine learning, deep learning involves researching and designing algorithms which can learn with multiple levels of abstraction. With this, neural networks will
be composed of cascading layers of information.
Machine learning: Not to be used interchangeably with AI, machine learning refers to the process by which an AI performs its function. It’s the science behind getting computers to do things they’re not explicitly trained
to do, and how the experts provide knowledge to computers with data, observations and real-world interaction. In other words, it doesn’t rely on typical rules-based programming.
Natural language processing: When an AI is trained to interpret, understand and respond to human communication.
Neural network: A computer system or program that’s modelled on the operations of neurons in the human brain. The large number of processors involved will operate in tiers – the first receiving raw input, and each successive
tier receiving the processed version. Each node will have its own knowledge, including rules its either been programmed with or developed itself, and only the last tier produces the output.
Reinforcement learning: This refers to how software agents can determine the behaviour of an AI by providing reward feedback to improve performance and support learning.
Superintelligence: A hypothetical AI that boasts intelligence far superior to humans.
Supervised learning: When systems are provided with both the input and desired outcome. Programmed with algorithms with known quantities, it’s a basis from future processing and judgements.
Unsupervised learning: AI training that doesn’t classify or label information. The algorithm must act without guidance or prior training to work with unsorted data. It’s how AIs are regularly tested to see the potential.
