TLDR: This research paper explores a hybrid approach to knowledge base management, combining symbolic AI (ontologies, expert systems, Semantic Web) with neural AI (generative models). It details the LaCAS platform for area studies as a case study, demonstrating how structured conceptual models guide generative AI agents through sophisticated prompt engineering to animate and enrich knowledge bases, offering a vision for autonomous, evolving knowledge ecosystems.
In an era where artificial intelligence is rapidly evolving, a new research paper explores the fascinating intersection of traditional symbolic AI and modern generative AI. Titled “Animating a Knowledge Base: From Ontologies to Generative AI Models. From Expert Systems and the Semantic Web to Generative AI: Model-Driven and Data-Driven Approaches in Area Studies,” the paper by Peter Stockinger delves into how these two distinct AI paradigms can be harmoniously combined to create more dynamic and intelligent knowledge bases, particularly within the field of area studies.
Understanding the Foundations: Symbolic AI and Expert Systems
The journey begins with symbolic AI, an approach that has been around for decades. Symbolic AI relies on explicit representations of knowledge, using formal logic and conceptual models (ontologies) to solve problems. Think of it as teaching a computer rules and facts, much like a human expert would reason. A prime example of symbolic AI in action is expert systems, which were designed to mimic human expertise in specific domains, such as medical diagnosis or archaeological classification. These systems consist of a knowledge base (a collection of facts and rules) and an inference engine (which applies these rules to solve problems).
While powerful, symbolic AI systems faced challenges, including the labor-intensive process of building and maintaining their knowledge bases, and ensuring the quality and coverage of the data. However, their strength lies in their explainability and the structured nature of their knowledge.
The Evolution: From Expert Systems to the Semantic Web
The principles of symbolic AI found a new home in the concept of the Semantic Web. Envisioned by Tim Berners-Lee, the Semantic Web aims to transform the internet into a vast, interconnected knowledge base where data is not just displayed but also understood by machines. Technologies like RDF (Resource Description Framework) and OWL (Web Ontology Language) are at its core, allowing for the creation of knowledge graphs that describe relationships between entities. Wikidata, a massive collaborative knowledge base, serves as an excellent example, turning Wikipedia entries into structured, interlinked data points, acting as a “knowledge hub” for various entities like languages or historical events.
This approach emphasizes the distinction between data (the objects being described) and metadata (the elements used to describe them), providing a structured way to manage and publish information.
LaCAS: A Real-World Semantic Platform for Area Studies
The paper introduces a concrete application of these concepts: the LaCAS (Open Archive in Language and Cultural Area Studies) platform, deployed at Inalco (National Institute for Oriental Languages and Civilizations) in Paris. LaCAS is an ecosystem designed to collect, process, and exploit research data in area studies – multidisciplinary research focused on specific geopolitical, linguistic, cultural, or historical regions.
The platform features a specialized thesaurus and a vast documentary fund, harvesting resources from various online repositories. It organizes a diverse range of knowledge objects into ten macro-domains, each further specialized. For instance, the “Linguistic Diversity” macro-domain includes “Languages of the world.” Each knowledge object is meticulously documented using a conceptual schema, aligning with major Linked Open Data (LOD) repositories like Wikidata and BnF Rameau. The platform, powered by the Okapi software environment from Ina, offers services for querying the knowledge base, consulting documentation, and exploring collections.
Documenting Languages: The Quechua Example
To illustrate the practical application, the paper uses the example of documenting the Quechua language. The LaCAS platform provides a structured documentation for languages, based on a conceptual model that defines properties like “Linguistic Typology,” “Geographical Environment,” and “Sociolinguistic Status.” For Quechua, this means detailing characteristics such as being an agglutinative language, having a subject-object-verb (SOV) sentence structure, and its status as a vulnerable minority language according to UNESCO. This structured approach ensures a comprehensive and consistent expertise of each language, forming a rich and evolving knowledge graph.
The Future is Hybrid: Integrating Generative AI
The most forward-looking aspect of the paper is its exploration of a hybrid approach, merging symbolic AI with neural (or sub-symbolic) AI, particularly generative AI. This hybridization aims to address two key challenges in knowledge base development: selecting relevant properties for conceptual models and efficiently localizing, qualifying, and analyzing large volumes of data.
Generative AI tools, such as conversational agents (chatbots) powered by large language models (LLMs) like ChatGPT or Mistral, can significantly assist in this process. By using specialized prompts aligned with the LaCAS ontology, these agents can locate and extract relevant information from vast corpora, compare and evaluate it against the conceptual model, and generate structured presentations. This “AI-guided expertise” can accelerate the reasoned updating of the knowledge graph without compromising critical rigor.
Also Read:
- AI-Powered Scenario Generation for Complex Service Environments
- Beyond the Black Box: LLMs and the Evolution of AI Agents
Prompt Engineering and the Ecosystem of Agents
The success of this hybrid approach hinges on “prompt engineering” – the art of designing precise and contextualized instructions for generative AI models. The paper outlines various types of prompts, from generic ones used repeatedly for specific knowledge domains to contextually anchored prompts that work within defined semantic spaces. It also discusses prompts that simulate reasoning, adjust models to new data, or manage sequences of tasks. This sophisticated prompt design, combined with a robust conceptual model and data quality criteria, ensures that AI agents contribute effectively and reliably to the knowledge base.
Ultimately, the vision is an “animated” knowledge base, transformed into a kind of artificial organism. This ecosystem would comprise a community of specialized artificial agents capable of autonomously enriching, adapting, and differentiating the knowledge base, making decisions for expertise projects, and interacting with other knowledge ecosystems. While still speculative, this perspective suggests a profound evolution in how we manage and interact with knowledge, pushing the boundaries of human-made intelligence. For more details, you can refer to the full paper available at https://arxiv.org/pdf/2509.01304.
