Glossary Terms

Uncertainty or potential harm that could materially affect an organisation’s objectives or direction. Consider how a proposed AI initiative changes dependencies, accountability and the consequences of failure.

Continuing an initiative because resources have already been spent, rather than judging its future costs and benefits. Pilot decisions should use current evidence, including the option to stop.

Machine learning using examples paired with target labels or values. Performance depends on label quality, representative examples and evaluation on data that has not informed model training.

A machine learning approach that fits a decision boundary, or a related regression function, using selected training examples. Evaluate feature preparation, configuration and performance for the intended application.

Problem-solving approaches inspired by the collective behaviour of groups such as ants or birds. They can support optimisation, but candidate solutions still need assessment against real business constraints.

A framework for examining strengths, weaknesses, opportunities and threats. Use it to organise relevant evidence and priorities, then translate findings into decisions rather than leaving them as a descriptive list.

A numerical data structure that generalises scalars, vectors and matrices to multiple dimensions. Tensors represent model inputs and calculations; their shape and meaning must align with the intended computation.

Data reserved for assessing a model after development decisions have been made. Keep it independent of training and tuning, and check that it represents the conditions the model will encounter.

Estimating future values from data ordered over time. Useful forecasts should account for uncertainty, changing conditions and a suitable comparison baseline before informing business decisions.

The process of dividing input, often text, into units that a model or system can process. Tokens may be words, word fragments or characters; token counts are not the same as word counts.