Beyond the Buzzword

Artificial intelligence has become one of the most discussed — and most misunderstood — topics in modern life. It appears in political speeches, earnings calls, classroom debates, and science fiction in equal measure. But what is AI, really? And how does the technology behind headline-grabbing tools like large language models and image generators actually work?

A Working Definition

At its broadest, artificial intelligence refers to computer systems designed to perform tasks that typically require human intelligence — things like recognising speech, identifying images, making decisions, or generating text. The field encompasses a wide range of techniques, from simple rule-based programs to complex neural networks trained on vast amounts of data.

The Key Branches of AI

Machine Learning (ML)

Rather than being explicitly programmed with rules, machine learning systems learn from data. Feed a model thousands of labelled photos of cats and dogs, and it will learn to distinguish between them. This approach powers spam filters, recommendation engines, and fraud detection systems.

Deep Learning

A subset of machine learning, deep learning uses artificial neural networks — loosely inspired by the structure of the human brain — with many layers (hence "deep"). Deep learning excels at processing unstructured data like images, audio, and text, and is the engine behind most modern AI breakthroughs.

Large Language Models (LLMs)

LLMs like those powering popular AI chatbots are deep learning models trained on enormous amounts of text. They predict what word or phrase should come next in a sequence — a deceptively simple objective that, when scaled massively, produces systems capable of writing essays, answering questions, and coding software.

Computer Vision

This branch enables machines to interpret visual information — identifying objects in photos, reading handwriting, or analysing medical scans. It underpins self-driving vehicle technology, facial recognition, and quality control in manufacturing.

What AI Cannot Do

Despite impressive capabilities, current AI systems have significant limitations that are important to understand:

  • They don't "understand" — they pattern-match. An LLM generating a convincing medical explanation has no comprehension of medicine; it has learned what plausible medical text looks like.
  • They can hallucinate. AI systems can confidently produce false information because they optimise for plausibility, not truth.
  • They reflect their training data. Biases present in training data are absorbed and can be reproduced or amplified.
  • They lack common sense and real-world grounding in the way humans naturally reason about physical situations.

AI in Everyday Life

You likely interact with AI systems multiple times a day without thinking about it:

  • Email spam filters
  • Streaming recommendation algorithms (what to watch or listen to next)
  • Navigation apps predicting traffic
  • Voice assistants
  • Autocorrect and predictive text
  • Photo sorting and face recognition on smartphones

The Policy Questions Ahead

Governments around the world are grappling with how to regulate AI — balancing innovation benefits against risks around privacy, misinformation, labour displacement, and safety in critical systems. The EU's AI Act, enacted in 2024, is the world's first comprehensive AI regulation framework, classifying AI applications by risk level and imposing corresponding obligations. Other jurisdictions are watching closely and developing their own approaches.

Understanding what AI actually is — rather than the science-fiction version — is essential for anyone following these policy debates.