Part 1 of the Alphabet Soup of Responsible AI Adjectives: Context
In the last decade, artificial intelligence (AI) has experienced rapid growth and adoption, along with accompanying calls to ensure that AI is beneficial to society. Governments, organizations, and individual contributors have committed attention and funding for investigating, informing, and improving AI models to ensure “good” AI. In doing so, these institutions have helped contribute to a burgeoning sector of research.
But what to call this research? Lots of options exist in the AI alphabet soup: Explainable AI, Interpretable AI, Transparent AI, Responsible AI, Trustworthy AI, Principled AI, Ethical AI, Antiracist AI, AI for Good. Although these adjectives may appear interchangeable, there are notable differences in usage between general internet users and researchers. In this blog post, I compare usage of these terms as measured through Google Trends, a reflection of Google searches, with the Dimensions database, a collection of research publications. In Part 2 of this blog post, I dive deeper into the definitions of these terms.
Google Trends Analysis of AI Adjectives
Google Trends is a tool for evaluating trends in Google searches and can be useful for understanding broader interest among internet users in particular topics. Figure 1 highlights Google Trends search interest for AI adjectives by Google users in the US in the last decade (ie., 2011–2021).
Interest was generally low in all terms prior to 2016. Search interest rose fast and continues to be high in AI for Good. Both Explainable AI and Ethical AI have seen a spike in interest since 2018 (note that Explainable AI is often abbreviated to XAI, but this abbreviation was not used for this analysis as XAI has other unrelated search interpretations). There is also growing search interest in Responsible AI. On the other hand, Transparent AI saw growth between 2012 and 2019 but appears to have decreased in searches between 2019 and 2021. Trustworthy AI and Interpretable AI have low rates of search interest, and Principled AI and Antiracist AI have no detectable search interest relative to the other terms.
Dimensions Analysis of AI Adjectives
Although Google Trends is a good resource for understanding searches by the general public, another tool — Dimensions — is helpful to understand trends in research. Dimensions is a free resource with over 100 million research publications, ranging from articles in scholarly journals to preprints, books, and conference proceedings. Its robust search features allow for analysis of key terms. Figure 2 illustrates the number of publications in the Dimensions database with a key term in the title or abstract since 2011.
Explainable AI is the most used term with a dramatic rise in usage in titles and abstracts since 2017. Trustworthy AI, Ethical AI, and Responsible AI have also shown moderate increases in usage after 2019. Interpretable AI has experienced a small uptick in usage. Transparent AI, AI for Good, and Principled AI occur infrequently, and Antiracist AI does not appear.
Commentary
Both the Google Trends and the Dimensions analyses demonstrate the novelty of this field, with most interest in and usage of terms occurring from 2016 onward. The generally increasing patterns in both sources suggest that developing AI that is beneficial to society will likely continue to be an increasing area of research and public interest.
Figure 3 illustrates the overlapping interests of terms as reflected in these two datasets. Explainable AI is a major topic of interest for both general internet searches and for researchers specifically. This likely reflects both general and researcher-specific interest in improving the explainability of AI models.
Researchers have greater usage of Interpretable AI and Trustworthy AI. Perhaps this is because internet users generally search for Explainable AI which may closely overlap with Interpretable AI. Researchers who use Interpretable may want to consider Explainable to reach a broader audience, or consider a more detailed explanation of why Interpretable is preferable or distinct from Explainable. Trustworthy AI may be a new term for general audiences and likely worth defining.
Google searchers have much more interest in AI for Good, Ethical AI, and Responsible AI, as well as moderately more interest in Transparent AI, that is currently not reflected in the research space. Searches for these terms may not return peer-reviewed research. Users can continue to advocate for more research related to these ideas, and researchers can respond by proposing and collaborating with users to make these ideas a reality.
Principled AI and Antiracist AI both rarely appear in Google searches or research publications. Researchers using these terms should be aware that they are less frequent and may not be familiar to readers. Greater awareness may need to develop among the public.
Thanks for taking the time to read this post. Feel free to comment with thoughts. Check out Part 2 where we take a deeper dive into the nuanced and overlapping meanings of these terms.
This blog post is part of a Deep Dive into Responsible Data Science and AI series.
Disclaimer: Support for this blog series was provided by RTI International. The opinions expressed by the author are their own and do not represent the position or belief of RTI International. Material in this blog post series may be used for educational purposes. All other uses including reprinting, modifying, and publishing must obtain written consent.
