What is an Artificial Intelligence?

AI spans a huge range of techniques — from rule-based expert systems and classical machine learning to today's large language models trained on vast text corpora using deep neural networks. The current wave of practical AI is dominated by transformer-based models that learn statistical patterns across enormous datasets, giving them broad, general capability rather than the narrow, single-task focus of older systems. A useful distinction is between narrow AI (built to do one thing very well — recognize faces, translate text, play chess) and the more general-purpose capability modern LLMs display, where a single model can reason across wildly different domains and tasks without being purpose-built for each one. Agentic AI takes this further, pairing a general-purpose model with tools it can call, turning raw language capability into the ability to actually take action in software systems. Whatever the technique, the throughline of practical AI work is the same: a model's outputs are ultimately a statistical prediction, not certainty, which is why real systems layer verification, human oversight, and guardrails around the model rather than trusting its output blindly.