Artificial intelligence (AI) is the field of computer science concerned with building systems that perform tasks normally requiring human intelligence — recognising speech, understanding language, making decisions or identifying patterns in data.
How modern AI works
Most systems in use today are built on machine learning: instead of being programmed with explicit rules, a model is trained on large amounts of data and learns statistical patterns from it. The dominant approach is deep learning, which uses neural networks with many layers.
A model does not "know" facts the way a database does. It predicts what output is most likely given an input, which explains both its flexibility and its failures.
Main branches
- Natural language processing — translation, summarisation, question answering.
- Computer vision — image classification, object detection, medical imaging.
- Speech technology — recognition and synthesis.
- Robotics and control — planning and movement in the physical world.
- Recommendation systems — ranking content for a specific user.
Common limitations
- Hallucination — a language model can produce fluent text that is factually wrong.
- Bias — a model reflects patterns in its training data, including unfair ones.
- Opacity — it is often hard to explain why a particular output was produced.
- Data dependence — quality collapses when input differs from training data.
For this reason, high-stakes uses — medicine, law, finance, hiring — require human review rather than automatic acceptance.
History in brief
The term was coined in 1956 at the Dartmouth workshop. Progress came in waves: early symbolic systems, expert systems in the 1980s, statistical methods in the 1990s, and the deep learning era that began around 2012. The introduction of the transformer architecture in 2017 led directly to today's large language models.
Practical view
AI is best understood as a tool that shifts the cost of certain tasks, not as a replacement for judgement. Its greatest value appears where the task is repetitive, tolerant of small errors, and easy to verify.