Prolog: The Overlooked Pioneer of Modern Artificial Intelligence
PROLOG: AI’s overlooked pioneer
From the logic programming of the 1980s to neuro-symbolic AI: why this remarkable language still deserves our attention.
I remember the 1980s, when choosing to specialise in Artificial Intelligence during my Computer Science degree at the Autonomous University of Barcelona (UAB) seemed almost crazy.
People would ask me, “What exactly is that? And what is it useful for?”
Neural networks, expert systems, computer vision, bioinformatics… I look back on those years with enormous affection!
From the UAB to Germany: my early years in AI
I eventually moved to Germany and joined Siemens Business Services in its banking division. There, I worked on computer security and improving cryptographic protocols for home banking.
I discovered something that genuinely surprised me: it felt as though Germany was fifty years ahead of us.
Today, in the age of neural networks, large language models (LLMs) and generative AI, I often think back to one of those early pioneers that, in my opinion, has never received the recognition it deserves: Prolog.
What is Prolog, and why does it matter in Artificial Intelligence?
Prolog is a logic programming language designed to represent knowledge, establish relationships and solve problems through inference. Its name comes from the French PROgrammation en LOGique.
Unlike many imperative programming languages, where we specify a sequence of instructions, in Prolog we generally describe what we know and what we want to find out. Its inference engine searches for answers consistent with the facts and rules provided.
This way of modelling problems is closely linked to an important tradition in AI: expert systems and symbolic reasoning.
How Prolog reasons: facts, rules and queries
One of the easiest ways to understand Prolog is through a simple example. We first declare facts, then define a rule and finally ask whether a particular relationship holds.
father(john, anna).
father(john, peter).
siblings(X, Y) :-
father(P, X),
father(P, Y),
dif(X, Y).
?- siblings(anna, peter).
true.This small program infers that Anna and Peter share a father. It is a simplified teaching example: a real application would define more carefully what we mean by siblings, half-siblings and other family relationships.
Here is the interesting part: we did not explicitly code every possible search path. We described relationships and allowed the logic engine to find solutions that satisfy the rules.
Where does Prolog really excel?
Although most AI conversations today revolve around LLMs, Prolog remains particularly interesting for problems centred on relationships, formal rules and constraints.
Represent specialised knowledge and support decisions governed by explicit rules.
Derive conclusions from formally defined facts, rules and relationships.
Express relationships between entities without manually coding every possible path.
Solve scheduling, resource allocation and other constrained problems with tools such as CLP(FD).
Work with formal grammars, parsing and linguistic relationships.
Make certain conditions and constraints explicit, which can help when building auditable systems.
The SWI-Prolog documentation offers further resources on constraint programming.
Why is Prolog less prominent in today’s AI landscape?
The rise of deep learning, the availability of large datasets and the enormous Python ecosystem have shifted attention towards statistical and neural approaches.
But learning patterns from data and reasoning explicitly through rules are different capabilities. They can also be highly complementary.
A neural network can be exceptionally effective at recognising patterns in images or text. A logic engine can represent constraints, formal relationships and conditions we want a system to enforce.
Neural AI
Learns patterns from examples and can generalise to new cases. Its behaviour is not always easy to interpret.
Symbolic AI
Works with explicit representations of knowledge, relationships and rules. Its usefulness depends on the quality of those representations.
That is why I resist the idea that AI must be either exclusively neural or exclusively symbolic.
Can Prolog complement LLMs? The promise of neuro-symbolic AI
Perhaps one of the most exciting directions for AI involves bringing together the best of both worlds: neural models to interpret, recognise and learn, and symbolic engines to represent knowledge, apply constraints and check specific conditions.
Imagine a system that uses a language model to interpret a request written in natural language, then relies on explicit rules to verify whether a proposed operation meets certain requirements.
Of course, a logic engine does not automatically guarantee that every conclusion is true. Its outputs depend on the accuracy of the facts, the rules and the technical integration. Nevertheless, it can provide a valuable additional layer of structure and control.
It really is fascinating! And for me, that is one of the keys: we must not forget the foundations while exploring new technologies.
Are we underestimating logic programming?
Prolog is not the answer to every AI problem. But ignoring it means overlooking one of the most elegant tools for building logic-based systems.
What role should it play in modern AI? What applications have we yet to explore fully?
Understanding the history of Artificial Intelligence can also help us imagine its future.
If I can help or advise you, I would be delighted. You can find me at my own venture, Aicertia — European AI Governance, Compliance, Training & Certification.
From technical understanding to AI governance
At Aicertia, we help organisations identify their AI systems, understand their risks and turn regulatory and governance requirements into practical measures.
Discover AicertiaContinuous AI GovernanceFrequently asked questions about Prolog
What is Prolog in Artificial Intelligence?
It is a logic programming language that represents facts, rules and relationships and answers queries through inference.
Is Prolog still used today?
Yes. It is still used in education, research and specialised applications involving logic, constraints, knowledge representation and reasoning.
Can Prolog replace Python or LLMs?
Not as a general replacement. These tools have different strengths and approaches and, in some applications, can complement one another.
What is neuro-symbolic AI?
It is an area of research that seeks to combine neural learning methods with symbolic representations or reasoning mechanisms.


