How Knowledge Graphs Relate to AI
Stanford CS520 course concluding chapter (2020).
Summary
- Symbiotic relationship: KGs power AI apps (personal assistants, recommenders, search engines); AI builds KGs via extraction (entities/relations), alignment (schema mapping, entity linking), cleaning, inference, QA.
- Graph Data Science: Emerging field leveraging graph structure + ML (feature engineering) for insights, predictions (e.g., finance risk/opportunities).
- Long-term AI: Explicit, expressive KGs (beyond simple graphs; semantic networks → description logics) needed for reasoning, explainability, challenges like commonsense/self-awareness.
Key Concepts
- Knowledge Graphs
- Graph Data Science
- Semantic Networks
- Description Logics
- Knowledge Representation
- Entity Extraction
- Relation Extraction
- Schema Mapping
- Entity Linking
- Feature Engineering
- Knowledge Engineering
- Commonsense Reasoning
Key Entities
Key Claims
- KGs essential for scalable, understandable AI vs. black-box LMs/NLP.
- Exercises highlight KG indispensability (e.g., not unnecessary for AI success).
Raw Excerpts
1. Introduction
In this concluding chapter, we will discuss different ways in which the knowledge graphs intersect with Artificial Intelligence (AI). … three themes: knowledge graphs as a test bed for AI algorithms, emerging new specialty area of graph data science, and knowledge graphs in the broader context of achieving the ultimate vision of AI.
2. Knowledge Graphs as a Test-Bed for Current Generation AI Algorithms
Knowledge graphs have a two way relationship with AI algorithms. … Hence, knowledge graphs enable AI systems, which provide motivation and a set of requirements for them. AI techniques are also fueling our ability to create the knowledge graph economically and at scale.
3. Knowledge Graphs and Graph Data Science
Graph data science is an emerging discipline … Because of the high impact use cases possible through graph data science, it is becoming an increasingly sought after software skill in the industry today.
4. Knowledge Graphs and Longer-Term Objectives of AI
Early work in AI focused on explicit representation of knowledge … Creating AI programs that can master a domain, formulate a hypothesis, design an experiment, and analyze its results is a challenge that is out of reach of any of the current generation systems.
5. Summary
We considered three different ways the work on knowledge graphs intersect with AI: as a test-bed for evaluating machine learning and NLP algorithms, as an enabler of the emerging discipline of graph data science, and as a core ingredient to realizing the long-term vision of AI. …