The Facets of Artificial Intelligence: A Framework to Track the Evolution of AI

Fernando Martínez-Plumed, Bao Sheng Loe, Peter A. Flach, Seán Ó hÉigeartaigh, Karina Vold, José Hernández-OralloIJCAI 2018 · DOI

Introduction: Mapping a Moving Target

Defining "Artificial Intelligence" (AI) has historically been a challenge. As the field evolves, what was once considered "intelligent" often becomes rebranded as standard computation—a phenomenon known as the "AI Effect." Early attempts to categorize the field often relied on simple binary distinctions, such as the classic 2x2 matrix proposed by Russell and Norvig, which contrasts thinking versus acting and human-like versus rational behavior.

The traditional 2x2 taxonomy of AI definitions focusing on human vs. rational and thinking vs. acting.

However, as AI permeates every aspect of modern life, these static definitions have become restrictive. This paper proposes a multi-dimensional "faceted" framework to provide a more nuanced, structured, and systematic way to track the "shape" of AI over time.

A Nine-Facet Framework

To move beyond binary definitions, the authors introduce nine facets that define the boundaries of AI research and application. This allows for a "polygonal" view where any given system exists on a gradient between multiple extremes:

  1. Functionality: Categorizing by techniques, applications, and specific tasks.
  2. Referent: Distinguishing between human-centric models and "universal" (theoretical) intelligence.
  3. Generality: The spectrum from narrow, task-specific AI to General AI (AGI).
  4. Location: Whether a system is centralized/integrated or distributed.
  5. Embodiment: Physical systems (robotics) versus virtual algorithms.
  6. Paradigm: The shift between discrete (logic-based), continuous (optimization), and probabilistic approaches.
  7. Actor: Identifying the drivers of research—Academia, Industry, or Government.
  8. Character: Balancing empirical experimentation with theoretical foundations.
  9. Nature: Classifying AI as Science, Engineering, Technology, or Philosophy.

Methodology and Datasets

The framework was validated through a quantitative bibliometric analysis of two major datasets:

  • AAAI/IJCAI Conference Data (1997–2017): Capturing the internal research perspective through thousands of peer-reviewed papers.
  • AI Topics Database (1905–2018): An archive of news, blogs, and documents reflecting the public and professional perception of the field.

Key Insights: The Evolution of the Field

The analysis reveals several profound shifts in the AI landscape over the last few decades.

The Copernican Shift and the Rise of Machine Learning

The authors advocate for a "Copernican view" of intelligence, where human-level ability is seen as just one point in a vast intelligence landscape. This is reflected in the data: there is a massive paradigm shift from "discrete" AI (logical and rule-based) to "continuous" AI (neural networks and optimization). Machine Learning has emerged as the dominant force, showing a steep, consistent upward trend that accelerated significantly after 2010.

Long-term trends in AI techniques, applications, and tasks showing the rise of Machine Learning and the decline of Expert Systems.

Industry Dominance and Virtualization

The "Actor" facet highlights a growing trend: industry is increasingly steering the discipline. This shift from academic to industrial dominance has significant implications for AI governance, as engineering concerns often outpace scientific and philosophical inquiry. Furthermore, despite the visibility of robotics, the "Embodiment" facet shows a clear trend toward "Virtual" systems, driven by the scalability of cloud-based AI.

  • Declining Paradigms: Traditional areas like "Heuristic search" and "Knowledge representation" have seen a steady decline in their share of research.
  • Cyclic Interests: Some topics, such as Cognitive Modeling and Planning, exhibit "bursty" or cyclic behavior rather than linear growth.
  • Engineering Focus: AI is increasingly viewed as a "Technology" or "Engineering" discipline rather than a purely scientific or philosophical pursuit.

Conclusion

The proposed faceted framework provides a robust tool for navigating the expanding boundaries of AI. By tracking these nine dimensions, stakeholders can better understand the transition from human-centric, academic roots to a universal, industry-driven, and continuous technological landscape. While current trends favor Deep Learning and virtual systems, the framework remains flexible enough to capture future shifts back toward symbolic logic or new, yet-to-be-discovered paradigms.

Summary updated

BibTeX
@inproceedings{martinezplumed2018facets,
  author = {Fernando Martínez-Plumed and Bao Sheng Loe and Peter A. Flach and Seán Ó hÉigeartaigh and Karina Vold and José Hernández-Orallo},
  title = {{The Facets of Artificial Intelligence: A Framework to Track the Evolution of AI}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2018},
  doi = {10.24963/ijcai.2018/718},
}
Figure