Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Why Isn't Relational Learning Taking Over the World?

David Poole

TL;DR

The paper argues that real-world AI should shift from purely perceptual modeling to relational learning that focuses on entities and their relations. It surveys relational data, knowledge graphs, and standard benchmarks, highlighting challenges in dataset construction, negative information, and evaluation that hinder generalization. It emphasizes the need for public, domain-relevant datasets, provenance, and cross-domain modeling to realize practical relational learning. The work also discusses training, aggregation, and evaluation limitations, outlining a forward path toward more scientifically grounded, decision-oriented relational AI across domains.

Abstract

Artificial intelligence seems to be taking over the world with systems that model pixels, words, and phonemes. The world is arguably made up, not of pixels, words, and phonemes but of entities (objects, things, including events) with properties and relations among them. Surely we should model these, not the perception or description of them. You might suspect that concentrating on modeling words and pixels is because all of the (valuable) data in the world is in terms of text and images. If you look into almost any company you will find their most valuable data is in spreadsheets, databases and other relational formats. These are not the form that are studied in introductory machine learning, but are full of product numbers, student numbers, transaction numbers and other identifiers that can't be interpreted naively as numbers. The field that studies this sort of data has various names including relational learning, statistical relational AI, and many others. This paper explains why relational learning is not taking over the world -- except in a few cases with restricted relations -- and what needs to be done to bring it to it's rightful prominence.

Why Isn't Relational Learning Taking Over the World?

TL;DR

The paper argues that real-world AI should shift from purely perceptual modeling to relational learning that focuses on entities and their relations. It surveys relational data, knowledge graphs, and standard benchmarks, highlighting challenges in dataset construction, negative information, and evaluation that hinder generalization. It emphasizes the need for public, domain-relevant datasets, provenance, and cross-domain modeling to realize practical relational learning. The work also discusses training, aggregation, and evaluation limitations, outlining a forward path toward more scientifically grounded, decision-oriented relational AI across domains.

Abstract

Artificial intelligence seems to be taking over the world with systems that model pixels, words, and phonemes. The world is arguably made up, not of pixels, words, and phonemes but of entities (objects, things, including events) with properties and relations among them. Surely we should model these, not the perception or description of them. You might suspect that concentrating on modeling words and pixels is because all of the (valuable) data in the world is in terms of text and images. If you look into almost any company you will find their most valuable data is in spreadsheets, databases and other relational formats. These are not the form that are studied in introductory machine learning, but are full of product numbers, student numbers, transaction numbers and other identifiers that can't be interpreted naively as numbers. The field that studies this sort of data has various names including relational learning, statistical relational AI, and many others. This paper explains why relational learning is not taking over the world -- except in a few cases with restricted relations -- and what needs to be done to bring it to it's rightful prominence.

Paper Structure

This paper contains 10 sections, 2 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Knowledge Graphs
  3. Datasets
  4. Relational Predictions
  5. Training
  6. Evaluation
  7. Looking Forward
  8. Acknowledgements
  9. Appendix A: Proportion of Wikidata entities with few triples
  10. Appendix B: An MRR agent

Figures (4)

  • Figure 1: A clicker question
  • Figure 2: Part of the Wikidata knowledge graph about Christine Sinclair; from Figure 16.1 of Poole:2023aa
  • Figure 3: Some of the triples in FB15k. The subject and object entities have their identifier replaced with the name. The verb (property) names have been abbreviated.
  • Figure 4: For each top-level entity: the Wikidata ID, the number of triples with the entity as subject, the number of local entities that are the subject of fewer than 10 triples, the number with 10 or more triples, and the (truncated) English name of the entity. Sorted by number of triples. These numbers were from 2025-08-19; they change over time as the knowledge base evolves. The Python code to generate the contents of this table is available from the author.