AI can broadly be understood through two different traditions: symbolic AI and statistical AI.
Symbolic AI is based on explicit rules, logic, and structured knowledge. It works by applying human-defined rules to facts or inputs. For example, a business rules engine might say: “If the order value is over £500 and the customer is new, require manual approval.” Symbolic AI is usually deterministic, meaning the same input and rules produce the same output. Its strengths are transparency, consistency, and explainability. Its weakness is that it struggles with messy, ambiguous real-world situations where the rules are hard to define fully.
Statistical AI learns patterns from data. Instead of being given fixed rules, it is trained on examples and learns what is likely. This includes systems such as spam filters, recommendation engines, image recognition models, and modern generative AI such as ChatGPT. Statistical AI is powerful because it can handle complex and messy data, including language, images, and behaviour patterns. Its weakness is that it can be less explainable and may produce confident but incorrect outputs.
In simple terms:
| Type | Core idea | Example |
|---|---|---|
| Symbolic AI | Follows rules and logic | Expert systems, rules engines, knowledge graphs |
| Statistical AI | Learns patterns from data | ChatGPT, Netflix recommendations, spam filters |
Many modern AI systems are hybrid. They may use statistical AI to recognise patterns and symbolic AI to apply rules, constraints, or checks.