In plain words
For its first few decades, most AI research worked like this: write down what you know as facts and rules, then let the computer combine them logically. If every bird can fly, and a sparrow is a bird, the computer concludes that a sparrow can fly.
This is symbolic AI. The symbols are words or labels the computer shuffles, like sparrow, bird and can-fly. It is also called good old-fashioned AI, or GOFAI.
Its strengths are clarity and precision. You can read every rule and trace exactly why it reached a conclusion. Its weakness is that the world is messy. Penguins are birds that cannot fly. Every rule needs exceptions, and the exceptions need exceptions.
Machine learning took over because it learns patterns from data instead. But symbolic ideas live on in tax software, route planners and research that combines the two.
Everyday example
Working out who's who at a wedding
At a wedding someone says: "Ramesh is Priya's father, and Priya is Arjun's mother." You apply a rule you already know: your parent's parent is your grandparent. So Ramesh is Arjun's nana. No guessing, just facts plus a rule, and you can explain every step. Symbolic AI reasons exactly like this, with thousands of hand-written facts and rules.
Where the example breaks: You also know countless unwritten things about families, like that a nana is usually older. Symbolic AI only knows what someone wrote down, and that unwritten common sense is the part nobody managed to write in full.
How it actually works
Symbolic AI rests on one bold claim: that thinking is the manipulation of symbols by rules.
Symbols and rules
Knowledge is stored as symbols (objects, properties, relations) and rules that link them, often in formal logic: if X is a parent of Y and Y is a parent of Z, then X is a grandparent of Z.
Reasoning as search
The program searches through chains of rules to reach a goal, such as proving a statement or finding a sequence of moves. Clever search strategies stop this from taking forever.
The big hypothesis
In 1976 Allen Newell and Herbert Simon stated that "a physical symbol system has the necessary and sufficient means for general intelligent action": symbol manipulation alone would be enough for intelligence.
Where it struggled
Writing rules for seeing, language and common sense proved endless. Systems were brittle: one unexpected case and they failed. Learning-based methods handled this messiness better.
Where it lives on
Planning and scheduling, checking that chips and software are correct, knowledge graphs, and neuro-symbolic systems that pair a neural network with a logic engine.
- Also called
- GOFAI, good old-fashioned AI, a name coined by philosopher John Haugeland in 1985
- Early landmark
- Newell and Simon's Logic Theorist, presented in 1956
- Best-known products
- Expert systems of the 1970s and 1980s
- The rival approach
- Learning from data with neural networks
Questions people ask
What is the difference between symbolic AI and machine learning?
Symbolic AI uses rules that people write by hand. Machine learning learns its own patterns from examples. Symbolic systems are transparent but brittle; learned systems handle messy data well but are harder to explain.
What does GOFAI stand for?
Good old-fashioned artificial intelligence. Philosopher John Haugeland coined the name in his 1985 book Artificial Intelligence: The Very Idea, to describe AI built on symbols and explicit reasoning.
Is symbolic AI still used?
Yes, though rarely under that name. Rule engines run tax and insurance software, planners schedule flights and factories, and formal logic checks chip designs. Researchers also combine symbolic reasoning with neural networks, an area called neuro-symbolic AI.
Why did symbolic AI fall out of favour?
Writing rules for everything proved impossible for messy tasks like recognising faces or understanding speech. Systems were expensive to maintain and broke on unexpected inputs. Learning from data scaled better as data and computing power grew.