Foundations
Artificial intelligence (AI)
In Arabic: الذكاء الاصطناعي
The whole field, and the widest of these words. It covers any attempt to get a machine to do something we would call intelligent if a person did it: recognising a face, translating a sentence, planning a route, deciding a claim. That includes systems built from hand-written rules, which is how most of the field worked for its first forty years, as well as everything that learns from data. The term was coined in 1956. It is worth keeping the word this broad, because the moment you use it to mean only chatbots you lose the language for everything else a business already runs.
In practice the word covers four layers that are often confused with one another, and knowing which one a vendor means is most of what a first meeting is for. A rule-based system is still AI and still deployed in banks and hospitals; a machine learning model is AI that learned its rules from examples; a deep learning model is machine learning built from many-layered networks; and generative AI is a deep learning model that writes rather than sorts. Each layer sits inside the one above it, so every generative system is also a machine learning system, and none of the older layers stopped working when the newest one arrived.
- Rule-based systems
- Behaviour written out by a person as explicit conditions. Predictable, auditable, and still the right answer wherever the rules are known and must not change on their own.
- Machine learning
- Behaviour learned from examples instead of written down. Handles problems nobody can state as rules, at the cost of being only as good as the data it was shown.
- Deep learning
- Machine learning with many-layered networks, which learn their own features rather than being handed them. This is the layer that made vision and language work.
- Generative AI
- Deep learning models that produce new content rather than labelling existing content. The newest layer, and the one people now mean when they say AI without qualifying it.