Simon Vandevelde is a post-doctoral researcher at the DTAI research group at KU Leuven University, Belgium.
His main research focus is on user-friendly KR, through which he aims to make KR approaches more accessible for industry.
In applied Knowledge Representation, we wish to build systems that “mimic” human reasoning by formally representing knowledge and letting reasoning engines reason over them. Adopting and implementing such knowledge-based systems can have many benefits for companies, such as recording processes formally, speeding up tasks, or supporting employees using faithful reasoning. Yet, implementing it typically requires an external AI expert or knowledge engineer, as companies lack expertise on modelling. This knowledge engineer then works together with the company’s domain expert(s) to build the knowledge base through a process known as knowledge acquisition.
It is well-known that building an accurate knowledge base is a difficult, costly, and error-prone process, among others due to communication errors between both parties. For this reason, this difficulty is sometimes called the knowledge acquisition bottleneck, and is cited as one of the main challenges for broader KR adoption.
However, what if domain experts could somehow be capable of modeling their knowledge by themselves? The bottleneck would effectively be removed, and KR would become much more accessible to industry. In essence, this is the “holy grail” of user-friendly KR: non-experts that formally express their knowledge without having to spend too much time learning the inns and outs of complex modeling languages such as First Order Logic. While achieving this goal is very far off, the field of KR has already seen some small steps towards this goal, which we aim to cover in this tutorial.
The tutorial consists of three main sub-topics, which will be organized according to the following schedule:
Formalisms In this section, we introduce the idea of using alternative formalisms that trade expressiveness for user-friendliness. Some of these formalisms repurpose popular standards intended for other purposes, such as UML, DMN, or feature models, while others introduce new languages entirely, such as ACE, COOM, cDMN, or PROLEG. Using alternative languages has benefits but also costs; through exercises, we aim to shed some light on these trade-offs.
Validation The next segment of the tutorial covers interacting with the knowledge as an important form of validation. When non-experts model knowledge, it is doubly important that they can effortlessly check if their formalizations behave like they expect them to. Learning to inspect the output of a powerful solver (e.g., clingo, IDP-Z3, or SAT solvers) is quite a skill in itself, and presents a significant barrier for non-experts. We will show how interactive interfaces aim to bridge this gap, and make validation more accessible to non-experts.
Role of LLMs To finish the tutorial, we will briefly touch on the possible roles of an LLM in the knowledge acquisition process, a research angle that we find missing in state-of-the-art research so far.
The tutorial is open to any interested researcher or practitioner who wishes to learn more about the topic. As our focus is user-friendliness, the tutorial should be suitable for everyone.
During the tutorial, we will have hands-on exercises, so participants are encouraged to bring their laptops. It is not required to install additional software; all exercises are purely browser-based.
Slides will be available soon.
Simon Vandevelde is a post-doctoral researcher at the DTAI research group at KU Leuven University, Belgium.
His main research focus is on user-friendly KR, through which he aims to make KR approaches more accessible for industry.