AI Is Not What the Movies Told You

Artificial intelligence, as it exists and is being deployed today, is not a sentient robot with goals and desires. It does not "think" in the way humans do. Understanding what it actually is — and isn't — is essential for making sense of both its genuine capabilities and the real concerns it raises.

Modern AI is, at its core, a set of mathematical techniques that allow computers to find patterns in data and use those patterns to make predictions or generate outputs. That sounds simple, but when applied at enormous scale to vast datasets, the results can be remarkably powerful — and sometimes surprising even to the people who built the systems.

Machine Learning: The Foundation

Traditional software is programmed with explicit rules: "if X happens, do Y." Machine learning flips this. Instead of writing rules, you feed a system large amounts of example data and let it figure out the rules itself by adjusting its internal parameters until it gets good at the task.

For example, to build an image classifier that distinguishes cats from dogs, you don't write rules describing what a cat looks like. You feed the system millions of labelled images and let it find the statistical patterns that distinguish the two — shapes, textures, proportions — through a process of trial and error.

Neural Networks and Deep Learning

The dominant approach in modern AI is the artificial neural network, loosely inspired by how neurons in a brain connect to one another. A neural network consists of layers of mathematical nodes. Data passes through these layers, getting transformed at each step, until an output emerges. The network learns by adjusting the strength of connections between nodes based on how wrong its outputs are.

Deep learning refers to neural networks with many layers (hence "deep") — sometimes hundreds. These deep networks are remarkably good at handling complex, unstructured data: images, audio, video, and text.

Large Language Models (LLMs): How ChatGPT and Its Peers Work

The AI tools generating the most public attention right now — ChatGPT, Gemini, Claude, and others — are called Large Language Models. They are trained on enormous quantities of text from the internet, books, and other sources. Their core task during training is deceptively simple: predict the next word in a sequence.

By doing this billions of times across vast datasets, these models develop internal representations of language that capture an enormous amount of information about how words, concepts, and ideas relate to one another. The result is a system that can generate coherent, contextually appropriate text — answer questions, summarise documents, write code, translate languages — without being explicitly programmed for any of these tasks.

What AI Cannot Do

Despite impressive capabilities, today's AI systems have real and important limitations:

  • They do not "understand" in a meaningful sense. They manipulate patterns in data without genuine comprehension.
  • They hallucinate. LLMs will confidently produce false information when they don't "know" the answer, because they are optimised to produce plausible-sounding text, not truthful text.
  • They are brittle outside their training distribution. AI systems can fail dramatically when faced with situations that differ significantly from their training data.
  • They reflect the biases in their training data. If the data they learned from contains biases, those biases will be reflected in their outputs.

Why It Matters

AI is being deployed in consequential domains — medical diagnosis, legal analysis, hiring decisions, financial risk assessment, weapons systems — where errors carry serious consequences. Understanding the basics of how these systems work, and what their fundamental limitations are, is no longer a niche technical concern. It is increasingly a matter of public literacy. The more clearly citizens, policymakers, and journalists understand what AI is actually doing, the better equipped they are to evaluate its risks and benefits honestly.