Loading the context…
Loading the context…
Computer systems performing tasks such as pattern recognition, prediction or generating content that otherwise involve human judgment.
Many current systems use machine learning: they learn statistical patterns from data rather than relying solely on manually specified rules. Training develops a model, and using that model on new inputs produces predictions or generated outputs. Performance depends on the task, data and evaluation conditions. An output can sound plausible while being wrong, so the intended use determines which checks and human decisions are needed.
Imagine a hypothetical fraud model screening 1,000 transactions, of which 100 are actually fraudulent. It flags 80 of those and also flags 100 legitimate transactions. It has detected 80% of the fraud cases, but only 80 of its 180 alerts—about 44.4%—are true fraud. Both figures matter: improving detection while overwhelming investigators with false alarms can create a different operational problem. The example is illustrative, not a commercial model's result.
WTO lifts goods-trade forecast as AI investment offsets disruption
Jamaica Customs sets out push for faster compliance and digital controls