Abstract illustration of interconnected AI data streams and a stylized globe.

Decoding AI: How We Measure It and Why Governance Matters

Beau Callahan in Tech & Innovation December 2025 • 4 min read.

"A practical look at artificial intelligence measurements for better policy-making."

Artificial intelligence (AI) is rapidly transforming our world, influencing everything from political agendas to everyday life. As AI's capabilities grow, so does the urgent need to govern its development and deployment. However, governing AI effectively requires that we first understand how to empirically measure it. This challenge stems from the lack of a clear definition of what constitutes AI, creating ambiguities that hinder research and policy-making.

Quantifying AI is critical for assessing its true impact. For example, how does AI affect labor markets, technological leadership, or productivity? Clear metrics are essential for gauging the effectiveness of regulatory policies and fostering informed discussions about AI's societal implications. Patents offer a valuable lens through which we can examine AI inventions, providing a quantitative and qualitative record of technological advancements.

This article explores four distinct approaches to classifying AI patents, each reflecting different interpretations and definitions of AI. By comparing these methods, we aim to reveal how empirical conclusions about AI's characteristics and concentration vary depending on the chosen classification strategy. This deeper understanding is crucial for anyone seeking to navigate the complexities of AI governance and harness its potential for the benefit of all.

Four Ways to Classify AI Patents: How Definitions Shape the Narrative

Abstract illustration of interconnected AI data streams and a stylized globe.

To understand AI's multifaceted nature, let's examine four distinct methods of classifying AI patents. Each approach offers a unique perspective on identifying AI inventions, revealing how our understanding of AI can be influenced by the criteria we use:

Using US patent data from 1990 to 2019, the study analyzes the extent to which each approach qualifies AI as a general-purpose technology (GPT) and examines patterns of concentration.

Keyword Search: Focuses on recent trends in neural networks, robotics, and natural language processing (NLP). This approach captures the pulse of current AI development, highlighting areas of intense activity.
Scientific Citations: Identifies patents that cite academic research in AI. By tracing the lineage of AI technologies to their scientific origins, this method emphasizes the importance of foundational research.
WIPO Classification: Utilizes the World Intellectual Property Organization (WIPO) classification method, which accounts for both the hardware and software aspects of AI. This approach provides a comprehensive view of AI's technical underpinnings.
USPTO Approach: Employs the United States Patent and Trademark Office (USPTO) classification, capturing the widespread use of AI in various inventions. This broad perspective recognizes AI's integration into diverse technological domains.

In a sample of US patents, only 1.37% are identified by all four approaches. This disparity underscores the subjectivity inherent in defining AI. AI patent shares range from 3–17% of all US patents in 2019. This classification method captures significant variations in scale and scope.

The Path Forward: Informed Governance Through Better Measurement

Measuring artificial intelligence is no simple task. There are several differences and limitations to be considered, so there is no one-size-fits-all solution for identifying AI innovations. Policymakers and researchers need to carefully evaluate how AI is defined and measured. By embracing a diverse range of classification methods and continually refining our understanding of AI, we can pave the way for responsible and effective governance, maximizing AI's potential for the benefit of society.

About this Article -

This article was crafted using a human-AI hybrid and collaborative approach. AI assisted our team with initial drafting, research insights, identifying key questions, and image generation. Our human editors guided topic selection, defined the angle, structured the content, ensured factual accuracy and relevance, refined the tone, and conducted thorough editing to deliver helpful, high-quality information.See our About page for more information.

This article is based on research published under:

DOI-LINK: https://doi.org/10.48550/arXiv.2204.10304,

Title: Measuring Artificial Intelligence: A Systematic Assessment And Implications For Governance

Subject: econ.gn q-fin.ec

Authors: Kerstin Hötte, Taheya Tarannum, Vilhelm Verendel, Lauren Bennett

Published: 21-04-2022

Everything You Need To Know

Why is measuring artificial intelligence so crucial for effective governance?

Measuring artificial intelligence is essential because it allows us to assess its true impact on various aspects of society, such as labor markets, technological leadership, and productivity. Clear metrics are needed to gauge the effectiveness of regulatory policies. Without empirical measurement, governance becomes challenging, hindering informed discussions and potentially leading to unintended consequences as AI's capabilities grow. The ability to measure AI helps in making informed decisions and maximizes AI's potential for societal benefit.

What are the four distinct methods used to classify AI patents, and how do they differ?

The article highlights four methods for classifying AI patents: * **Keyword Search**: Focuses on current trends in neural networks, robotics, and natural language processing (NLP). * **Scientific Citations**: Identifies patents that cite academic research in AI. * **WIPO Classification**: Utilizes the World Intellectual Property Organization (WIPO) classification, accounting for both hardware and software aspects of AI. * **USPTO Approach**: Employs the United States Patent and Trademark Office (USPTO) classification, capturing the widespread use of AI in various inventions. These methods differ in their scope and focus, leading to varied interpretations of what constitutes AI. This also influences the assessment of its characteristics and concentration.

What is the significance of the disparity in AI patent identification across different classification methods?

The disparity in identifying AI patents across different methods, as shown by only 1.37% of US patents being identified by all four approaches, underscores the subjectivity inherent in defining AI. This variation, with AI patent shares ranging from 3–17% of all US patents in 2019, highlights that the chosen classification strategy significantly influences the empirical conclusions about AI's characteristics and concentration. This is a crucial point for policymakers and researchers because it emphasizes the need to carefully evaluate how AI is defined and measured when making decisions about governance and regulation.

How do the different approaches to classifying AI patents provide varied perspectives on AI development?

Each approach offers a unique lens through which to view AI development: * **Keyword Search** highlights areas of intense activity in recent trends. * **Scientific Citations** emphasizes the importance of foundational research by tracing technologies to their scientific origins. * **WIPO Classification** provides a comprehensive view of AI's technical underpinnings, accounting for both hardware and software. * **USPTO Approach** recognizes AI's broad integration into diverse technological domains. By comparing these methods, we can reveal how our understanding of AI can be influenced by the criteria we use, leading to a deeper understanding of AI's multifaceted nature.

What are the key takeaways for policymakers and researchers regarding AI measurement and governance?

Policymakers and researchers should carefully evaluate how artificial intelligence is defined and measured. Because there is no one-size-fits-all solution, embracing a diverse range of classification methods and continually refining our understanding of AI are essential. This approach allows for responsible and effective governance, ultimately maximizing AI's potential for societal benefit. The choice of classification method significantly impacts the analysis of AI characteristics and concentration, so understanding the implications of each method is critical for informed decision-making.