Definition
Symbolic AI represents knowledge as explicit symbols and rules, and reasons with logic. Connectionist AI represents knowledge in the weights of a network of simple units, and learns it from data. Neuro-symbolic AI combines the two.
In simple words
One way to teach a child about birds is to give rules: birds have feathers, birds lay eggs. That's symbolic. The other way is to show thousands of bird photos until they just know. That's connectionist. People use both, and AI researchers are learning to as well.
Why it matters
This is the great divide in AI's history, with each side winning at different times. It also explains today's trade-off: neural networks are powerful but hard to explain, and rules are clear but brittle.
How it works
| Symbolic | Connectionist | |
|---|---|---|
| Knowledge is in | Rules a person can read | Millions of weights |
| Comes from | Experts | Data |
| Strength | Clear, exact, explainable | Handles messy input, like images |
| Weakness | Brittle, hard to scale | Hard to explain, needs a lot of data |
| Examples | Expert systems, Prolog, planners | CNNs, Transformers, LLMs |
Common mistakes
- Thinking symbolic AI is dead. Planners, solvers and knowledge graphs are used widely today.
- Assuming it must be one or the other. Many of the best systems combine them.
Check yourself
Summary
- Symbolic: explicit rules and logic.
- Connectionist: weights learned from data.
- Each covers the other's weakness, so they're increasingly combined.
Finished reading?
Sources: Russell & Norvig, Artificial Intelligence: A Modern Approach