We need to think about new approaches when modeling knowledge for AI

June 28, 2022 by Tassilo Pellegrini
David Chaves Fraga - Poster & Demo Co-Chair of Semantics 2022

In this interview David Chaves-Fraga, Poster & Demo co-chair of SEMANTiCS 2022, talks about challenges in the convergence of AI and knowledge graphs when it comes to robustness and maturity.

Declarative approaches to knowledge representation have had a long tradition in AI research and are gaining popularity again as the big data hype is fading. How can we make sense of this development?

David Chaves-Fraga: My feeling is that declarative approaches are usually a win-win situation for most of the processes where data is involved. They have a set of features (e.g., explainability) that have been demonstrated to be really relevant for a trustable AI. However, declarative solutions are usually costly, so we need to think about new engineering methodologies, workflows, and standard procedures to create and maintain these approaches in a sustainable way.

Knowledge graph construction is usually associated with a lot of manual and costly efforts, especially when maintenance is concerned. Are there viable approaches to automate KG construction and thus reduce efforts?

David Chaves-Fraga: Definitely, it is a topic that catches our attention, and we should put more effort in the near future. We have already seen that semantic tabular annotators (e.g., from the SemTab challenge) are a good starting point for automating the construction of knowledge graphs, although there are still many open research topics that need to be addressed for having robust and mature solutions. We should also not forget the relevance of incorporating users in the loop.

Q3: What are hot topics we should pay attention to when it comes to the intersection of AI and knowledge graphs in the near future?

David Chaves-Fraga: In terms of fundamental research, I think we need to pay more attention to other fields such as databases, which have been using AI techniques for data management tasks (e.g., data cleansing) for a long time. On the other hand, I see a clear application opportunity for the exploitation of knowledge graphs with AI techniques with the emergence of decentralized data ecosystems.