Artificial intelligence is often described in four types, based on how much an AI system can remember and understand. This framework helps separate the AI we use every day from the AI that appears in films. Here are the four types, with examples of each.
1. Reactive machines
Reactive machines are the simplest AI. They respond to the current input using fixed rules and have no memory of the past. The classic example is IBM's Deep Blue, the chess computer that beat Garry Kasparov in 1997: it evaluated the position on the board and chose a move, without learning from earlier games. A basic spam rule or a simple recommendation rule also behaves reactively.
Strengths: predictable and fast. Limits: cannot learn or improve. Read more: What is a reactive machine?
2. Limited memory AI
Limited memory AI can use recent data to make better decisions. Most AI in use today belongs here. Self-driving cars watch the speed and position of nearby vehicles over the last few seconds, chatbots use the recent conversation, and recommendation systems learn from what you watched. The model is trained on large past datasets, then uses short-term memory while it works.
Strengths: learns from data, powers most real-world AI. Limits: the memory is short and the model does not truly understand. Read more: Limited memory AI.
3. Theory of mind AI
Theory of mind AI would understand that people have beliefs, emotions and intentions that affect their behaviour, and respond accordingly. It would, for example, notice that a customer is frustrated and change how it helps. Today's systems can detect emotions in text, voice or faces to a limited degree, but nothing yet has a real model of a human mind. This is an active research area. Read more: Theory of mind AI.
4. Self-aware AI
Self-aware AI would have consciousness and a sense of its own existence. It does not exist and there is no agreed route to build it. It appears in science fiction, and in debates about the long-term future of AI. Read more: Self-aware AI.
Comparison table
| Type | Memory | Learns? | Exists today? | Example |
|---|---|---|---|---|
| Reactive machine | None | No | Yes | Deep Blue chess computer |
| Limited memory | Recent data | Yes (trained) | Yes | Self-driving car, chatbot |
| Theory of mind | Models of people | Research stage | Not yet | Emotion-aware assistant |
| Self-aware | Self-model | Unknown | No | Science fiction |
A second way to group AI: narrow, general and super
AI is also grouped by capability. Weak or narrow AI does one task well and is everything we have today. Artificial general intelligence would match human ability across tasks, and artificial superintelligence would exceed it. Our overview What is artificial intelligence? ties these together.
What this means for business
Almost every business AI project today, such as chatbots, forecasting and document search, uses limited memory AI. Netstech builds these for companies; see our AI development services.
Frequently asked questions
Which type of AI is ChatGPT?
Large language models are generally described as limited memory AI: they are trained on past data and use the current conversation as short-term context.
Does theory of mind AI exist?
Not in a full sense. Researchers test whether language models can pass theory of mind tasks, but there is no system with a genuine model of human minds.
Is self-aware AI possible?
Nobody knows. There is no accepted scientific definition of machine consciousness, and no system today is self-aware.
