When people discuss 'types' of AI, they are usually referring to one of two different classification systems: one based on how capable the system is compared to human intelligence, and another based on how the system is built and what it can functionally do.
Classification by Capability
- Narrow AI (also called weak AI): systems built to perform one specific task very well, such as recommending products or recognising faces. Every AI system in use today falls into this category.
- General AI (AGI): a hypothetical system with human-level reasoning across virtually any task, able to transfer knowledge from one domain to another the way people do. This does not exist yet.
- Superintelligence: a theoretical future stage where a system would surpass human intelligence across all fields. This remains speculative and is debated among researchers.
Classification by Function
- Reactive machines: respond to current input with no memory of past interactions, like a chess engine evaluating the board in front of it.
- Limited memory systems: use recent past data to inform decisions, which describes most modern AI including self-driving car perception systems.
- Theory of mind and self-aware AI: forward-looking categories describing systems that could, in principle, understand beliefs and intentions or have their own internal state; neither exists in practice today.
Almost every AI tool you encounter in daily life, from a translation app to a chatbot, is narrow AI. Keeping this distinction in mind is a useful filter when reading dramatic headlines about AI, since most current systems are far more limited and specialised than science-fiction portrayals suggest.