Loading field map
The subfield of knowledge representation and reasoning (KR&R) emerged from early artificial intelligence as the systematic study of how to encode human knowledge in a form usable by computational agents. Its foundational paradigm was Symbolic AI, which treated knowledge as explicit, declarative structures composed of symbols that could be manipulated by logical inference engines. This established a durable agenda centered on formal logic, giving rise to families like first-order logic and production rule systems, which aimed to model intelligent reasoning through theorem proving and heuristic search.
By the 1970s and 1980s, the limitations of pure logic for capturing everyday conceptual knowledge spurred rival structural paradigms. Semantic Networks proposed knowledge as graphs of interconnected concepts, emphasizing inheritance and relational reasoning. The Frame paradigm, reacting to the rigidity of logic, organized knowledge into prototypical units with slots and defaults, modeling expectation-driven reasoning. These schools shared the symbolic commitment but introduced structured, often procedural, representations that dominated applied expert systems.
A major concurrent school was Probabilistic AI, which challenged the deterministic foundations of symbolic approaches. It introduced frameworks for representing uncertain knowledge and performing plausible reasoning, most canonically through Probabilistic Graphical Models like Bayesian networks. This paradigm framed reasoning as belief updating under uncertainty, establishing a lasting technical agenda that treated uncertainty not as a flaw but as a core feature of knowledge.
The rise of Connectionism and later Deep Learning presented a profound alternative, subsuming representation within learned, distributed numerical parameters. While initially seen as oppositional to explicit KR, this led to the modern integrative paradigm of Neuro-Symbolic AI, which seeks hybrid architectures combining the learning power of neural networks with the compositional generalization and explicit reasoning of symbolic systems. This represents the current frontier in reconciling the field's historic agendas.
Throughout, the KR&R subfield has been defined by the tension and synthesis between these durable schools: the precision of logic, the structured organization of frames and networks, the calculus of probability, and the subsymbolic vectors of connectionism. The core pursuit remains the creation of formalisms that are both expressively adequate for real-world knowledge and computationally tractable for automated reasoning.