Key takeaways
- Symbolic AI represents knowledge as named facts and rules and draws conclusions with logic or search.
- It is also called good old-fashioned AI, or GOFAI, after the decades when it led the field.
- Its strengths are traceable reasoning, exact rules and no need for training data.
- Its weaknesses are brittleness, hand-written rules and trouble with free text, images and other messy input.
- It is one of the two traditions neurosymbolic AI combines, alongside machine learning.
How symbolic AI works
Symbolic AI works by writing knowledge down as symbols a program can manipulate, then applying rules of logic or search to reach new conclusions.
A symbol stands for something in the world: a company, a country, a job title. Facts link symbols, such as "Acme is based in Germany", and rules say what follows, such as "if a company is based in Germany, it is in the DACH region". An inference engine applies the rules to the facts, and every conclusion comes with the chain of steps that produced it.
- Knowledge representation: how facts, categories and relations are written down.
- Rule-based and expert systems: if-then rules applied by an inference engine.
- Automated reasoning and search: working through possibilities to find a proof, a plan or a valid answer.
Why it is called good old-fashioned AI
The name good old-fashioned AI, or GOFAI, comes from the decades when symbolic methods were the mainstream of AI research, before machine learning took over.
From the 1950s on, researchers built theorem provers, planning systems and game programs that searched through possible moves. In the 1970s and 1980s, expert systems captured specialists' knowledge as if-then rules and were sold to businesses. They fell out of favour largely because writing and maintaining thousands of rules by hand proved slow and costly, a problem known as the knowledge acquisition bottleneck. Rule engines never left business software, though.
Symbolic AI vs machine learning
Symbolic AI follows rules people write, while machine learning finds its own patterns in examples. That one difference drives the rest.
A symbolic system can explain every decision, behaves the same way each time and needs no training data, but it only knows what someone wrote down. A machine learning model can cope with cases nobody described in advance and with messy input such as free text, but its reasoning is spread across learned numbers, and it can be wrong with no clear reason why.
Symbolic AI examples in sales and marketing
A lead-routing rule is a small, everyday example of symbolic AI. Take an illustrative rule: if the company is in the DACH region and has more than 200 staff, assign it to the enterprise team. It is clear, anyone can check it, and it gives the same answer every time.
It is also blind to what its author did not foresee: a 150-person company about to double in size, or a lead whose headquarters is recorded in the wrong country. Other examples are scoring sheets with fixed points per attribute, approval rules for discounts and the eligibility checks in a quoting tool.
Where symbolic reasoning helps and where it fails
Symbolic reasoning helps wherever a decision must follow a stated policy and be explained afterwards, and it fails wherever the input is too varied to describe in rules.
Compliance checks, eligibility, routing and calculations suit it well. Reading a buyer's email, judging tone or recognising a company from a misspelt name do not, because no one can write a rule for every variation. That gap is why research now pairs symbolic reasoning with neural models instead of choosing one.
Where Pebbles Ai fits
Common questions
What does GOFAI stand for?
GOFAI stands for good old-fashioned AI. The philosopher John Haugeland coined it in the 1980s for AI built on symbols and explicit rules.
Is symbolic AI still used today?
Yes. Rule engines in business software, route planners, scheduling tools and formal software verification all rest on symbolic methods, often beside machine learning.
Is an if-then rule in a CRM a form of AI?
In the broad sense, a single rule is the simplest form of symbolic reasoning, though few people would call it AI. What matters is that the rule is explicit and checkable.
Can symbolic AI learn?
Classic symbolic systems do not learn from data; people add and change their rules. Some research methods learn rules from examples, and neurosymbolic designs pair symbolic reasoning with a learning component.
About this page. Written by the Pebbles Ai team for people who run sales, marketing and growth. Last updated .
