symbolic AI
/ sim-BOL-ik AY-EYE /
Symbolic AI is the original, top-down approach to making machines intelligent: spell out human knowledge as explicit symbols and rules, then let the computer manipulate them with logic. If you wanted a system to reason about family trees, you would write rules like 'if X is a parent of Y, and Y is a parent of Z, then X is a grandparent of Z.' Intelligence, on this view, is the careful shuffling of meaningful symbols according to clear rules — much like solving a logic puzzle or doing algebra.
This was the dominant paradigm from the 1950s into the 1980s, and it gave us logic theorem-provers, planners, and expert systems. Its great strength is transparency: because every step follows a stated rule, you can read the system's reasoning and see exactly why it concluded what it did. It also handles crisp, well-defined domains — chess, formal logic, scheduling — elegantly, without needing mountains of training data.
Its great weakness is the messy real world. Hand-writing rules for everything a human knows turns out to be impossibly hard: common sense alone has billions of unstated facts, and rules are brittle when reality doesn't fit the neat categories. Symbolic AI struggles with perception (is that a cat or a fox?) and with knowledge too fuzzy to write down. Modern AI leaned the opposite way, toward learning patterns from data (connectionism). The two are not enemies, though — much current research tries to combine the readability of symbols with the flexibility of learning, an approach called neuro-symbolic AI.
A symbolic medical helper might hold the rule: 'IF the patient has a fever AND a stiff neck AND a rash, THEN suspect meningitis and recommend urgent tests.' You can read it, check it, and argue with a doctor about it — but someone had to write every such rule by hand, and the system knows nothing the rules don't state.
Symbolic AI reasons with human-readable rules — powerful, transparent, and brittle outside its hand-coded knowledge.
Symbolic AI is often nicknamed GOFAI ('good old-fashioned AI'). It did not 'fail' — it excels at logic and structured reasoning — but it could not scale to the messy, unstated knowledge of everyday perception, which is where learning-from-data took over.