This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Research Protocols, is properly cited. The complete bibliographic information, a link to the original publication on https://www.researchprotocols.org, as well as this copyright and license information must be included.
Artificial intelligence (AI) has emerged as a major driver of technological development in the 21st century, yet little attention has been paid to algorithmic biases toward older adults.
This paper documents the search strategy and process for a scoping review exploring how age-related bias is encoded or amplified in AI systems as well as the corresponding legal and ethical implications.
The scoping review follows a 6-stage methodology framework developed by Arksey and O’Malley. The search strategy has been established in 6 databases. We will investigate the legal implications of ageism in AI by searching grey literature databases, targeted websites, and popular search engines and using an iterative search strategy. Studies meet the inclusion criteria if they are in English, peer-reviewed, available electronically in full text, and meet one of the following two additional criteria: (1) include “bias” related to AI in any application (eg, facial recognition) and (2) discuss bias related to the concept of old age or ageism. At least two reviewers will independently conduct the title, abstract, and full-text screening. Search results will be reported using the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) reporting guideline. We will chart data on a structured form and conduct a thematic analysis to highlight the societal, legal, and ethical implications reported in the literature.
The database searches resulted in 7595 records when the searches were piloted in November 2021. The scoping review will be completed by December 2022.
The findings will provide interdisciplinary insights into the extent of age-related bias in AI systems. The results will contribute foundational knowledge that can encourage multisectoral cooperation to ensure that AI is developed and deployed in a manner consistent with ethical values and human rights legislation as it relates to an older and aging population. We will publish the review findings in peer-reviewed journals and disseminate the key results with stakeholders via workshops and webinars.
OSF Registries AMG5P; https://osf.io/amg5p
DERR1-10.2196/33211
Artificial intelligence (AI)—defined as “[the] designing and building of intelligent agents that receive percepts from the environment and take actions that affect that environment” [
An aging global population [
There is an increasing presence of technology and AI in our daily lives, with substantial applications in health care [
Technologies that are created on the basis of inaccurate assumptions about older people can cause users (ie, older people) to internalize negative stereotypes, reducing their self-efficacy and willingness to engage with technologies in general [
The goals of this study are interdisciplinary in nature and aim to explore how age-related bias is encoded and amplified in AI systems and understand any corresponding societal, legal, and ethical implications. This review will address the following research questions:
What is known about age-related bias in AI technology?
How do AI systems encode, produce, or reinforce age-related bias?
What literature exists on the extent of age-related bias in AI systems?
What is the state of knowledge on older people’s experiences of age-related bias in AI systems?
What are the social, legal, and ethical implications of age-related bias in AI systems?
This study will contribute to the global conversation about bias in AI systems and the associated concerns of fairness [
This scoping protocol was developed using guidance from the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) reporting guideline [
As scoping review questions are recommended to be broad [
Under an information specialist’s guidance, the research team developed a search strategy consistent with scoping review methodology [
Main concepts included in the search strategy. AI: artificial intelligence.
This section will describe the completed search strategy for the scoping review. The search strategy was informed by test searches in Scopus, Medline, IEEE Xplore, ACM Digital Library and Google Scholar with the key search terms “artificial intelligence” and “ageism.” The first 200 results in each database were screened by checking their title and abstract for relevant records. There were no relevant search results that explicitly discussed AI and ageism, so the concept “ageism” was expanded and changed to “age” to capture more records discussing aging as suggested by the information specialist. Next, individual key terms were searched to gather synonyms, and a synonym list was generated (see
We conducted test searches by combining synonyms of our key concepts in Scopus, a multidisciplinary database that matched the nature of our study, to examine which synonym combinations could generate relevant records. After searching for all the synonyms proposed, we found 5 key papers that discussed age-related algorithmic bias, 53 relevant articles, and 29 additional synonyms occurring in the titles, abstracts, or keywords of these records (
Printed in English
Peer-reviewed publications and conference papers
Available electronically in full text
Meet one of the two criteria below:
Report “artificial intelligence” (algorithms that predict or classify data), “bias,” and terms related to “age” (aging, older, demographic)
Report facial recognition and age or demographics
Theses and dissertations
Conference abstracts and proceedings
Perspectives and editorials
Books and book chapters
Letters to editors
Manuscripts using nonhuman samples
Manuscripts that do not use human data
Children as the target population
Theoretical analysis
Mathematical formulations
Nonhuman studies
The final search strategy was developed in Scopus and then translated to the other 5 databases (Web of Science, CINAHL, EMBASE, IEEE Xplore, and ACM digital library). As IEEE Xplore had limitations on the number of terms and wildcards used for the search, we iteratively tested one theme or combinations of themes using different subsets of the proposed synonyms. A synonym was deleted if its addition to the search did not produce relevant results. We screened the first 200 records produced in each testing and eliminated the corresponding synonyms if none of the results were relevant.
The search parameters included peer-reviewed publications and conference papers published in English and available electronically in full text. Due to the study’s interdisciplinary nature, we did not limit the study design for inclusion. The search strategy was also not restricted by publication date since the term “artificial intelligence” has existed for over 50 years [
Given the anticipated paucity of academic research studies directly focused on ageism in AI, grey literature will increase the breadth and relevance of our findings. With the search strategy established, an iterative grey literature search strategy will be used to retrieve documents in the public domain that are relevant to any of our research questions to ensure that all relevant information about age-related bias in AI is captured. Grey literature will be retrieved by searching grey literature databases (OpenGrey and Grey Literature Report). Targeted searches of websites identified by the research team (eg, Algorithm Watch, Healthcare Information and Management Systems Society, The Centre for Data Ethics and Innovation) will also be conducted to retrieve documents such as white papers, policy papers, technical papers, and government reports. These documents will be downloaded in PDF form and added to a separate Microsoft Excel table to record the website source. After a thorough full-text review of each source, a rating scale of 0 to 4 (0=no reference to AI and ageism, 1=mentioned “age” in a list of types of biases, 2=one sentence related to the age-related bias, 3=two or three sentences related to age-related bias, 4=more than three sentences relevant to AI and ageism) representing the relevancy of the document was used to identify which sources were most relevant to the study. The included sources (anything with a rating above 0) had the relevant portions of text with the corresponding page numbers highlighted and documented, which will be themed by the research team according to each research question. To date, we have completed a preliminary manual Google search using the terms “artificial intelligence” and “ageism,” which identified 213 results in November 2021. A reviewer (JS) from the research team opened each web page to screen the content on the page for relevance. We found additional pages from law-related blogs that referenced employment discrimination related to age-related algorithmic bias.
Given the anticipated legal and ethical implications of ageism in AI, a review of relevant legislation, regulations, and jurisprudence (court cases) will be used to augment our academic and grey literature searches. These data sources will address research question 5 (What are the social, legal, and ethical implications of age-related bias in AI systems?). This process will be led by the team’s legal scholars and focused on understanding the legal and regulatory framework to protect and prevent age-related bias and unjust discrimination in AI. The iterative legal search strategy will begin with a review of relevant secondary sources including legal dictionaries and encyclopedias, followed by a review of legal treatises, law reviews and journals, statutes, and administrative regulations, and finally an analysis of the relevant case law. The legal databases WestlawNext Canada and CanLII will be canvassed in this legal review. Given the relative novelty of AI in the legal realm, a broad keyword search will be used to capture the relevant material. The keywords include “artificial intelligence;” “A.I.”; “machine learning”; “ageism”; and “discrimination.” The keyword search will be periodically refined to limit search results to various legal domains, including employment law, human rights law, and health law.
The search results will be exported into Covidence, a commonly used web-based literature review tool. The eligibility of the publications was determined based on a screening guideline established by 2 reviewers (JS and CHC;
We will chart the data based on the primary research questions using tools such as Google sheets or Covidence.
Article title
Data charted by (initials)
Author(s)
Year
Country
Aim or purpose
Study design
Branch of AI
Algorithms as described
Type and source of data
Does the article report age as demographic information of the study population?
Does the article report on the experience of older people with age-related bias?
Data set: yes or no
AI algorithm: yes or no
Methods proposed to mitigate bias, if any
Legal implications
Societal implications
Ethical implications
Data charting will serve as the first step to summarizing the results. We will record each study based on fundamental information including article title, author(s), publication year, country, and study aims. Based on what is commonly reported in other AI reviews, we will potentially include technology-related information such as the aim of the technology, stage of the technology development, data used, and validation methods. To synthesize the findings, we will conduct a thematic analysis and use a narrative description to describe the work according to study design (quantitative or qualitative), any emerging patterns identified, ethical implications, as well as legal considerations. Collation of the findings will inform gaps for future studies in the field of AI and ageism.
To allow for stakeholder involvement and additional insights beyond the literature, the preliminary summary document will be circulated to stakeholders, including our national and international research collaborators with expertise or interests in aging, subject experts from the Temerty Centre for Artificial Intelligence Research and Education in Medicine at the University of Toronto, a senior’s advocate, and older adults. These stakeholders have been involved from the early stages of the research conceptualization as knowledge users on our grant application.
The database searches resulted in 7595 records when the searches were piloted in November 2021. Data will be abstracted in a tabular format to support drafting of a narrative summary. A scoping review publication will serve as the main presentation of the findings. The remaining stages of the search is proposed to reach completion by December 2022.
The findings of this review will provide foundational information to advance our understanding of the concept and extent of digital ageism, which occurs when technologies deliberately or inadvertently exclude older adults, prioritize younger adults, or fail to recognize the diverse needs of the older adult demographic through various means [
A strength of our review is that our study is interdisciplinary and will shed light on AI and age-related bias regarding older adults from societal, legal, ethical, and technical perspectives. We have a rigorous methodology based on a scoping review framework and a comprehensive search strategy that includes interdisciplinary and discipline-specific databases. A team of researchers from different fields will interpret and generate findings that will foster further discussions and provide a direction for future work related to AI and older adults. One of the potential limitations of this study is the exclusion of publications in non-English languages as well as studies that do not discuss bias or age-related bias explicitly, potentially excluding research that unknowingly uses skewed data due to age-related bias embedded in specific AI algorithms. The inclusion of literature that explicitly discusses or recognizes the potential for age-related bias allows us to answer our current research questions. Our future work will explore the presence of implicit age-related bias in AI, as well as how ageism is reflected in a subset of AI algorithms.
To our best knowledge, this is the first scoping review to explore how age-related bias is encoded or amplified in AI systems and consider the societal, legal, and ethical implications. This scoping review protocol documents the search strategy and outlines the in-depth process for our rigorous synthesis of the literature on AI and ageism. Once the review is complete, we will connect with organizations at provincial, national, and international levels to discuss the findings and build the corresponding interview guides for in-depth semistructured interviews. Our review has the potential to establish the intersection of AI and ageism, advance knowledge about digital ageism, and inform future regulation and policy in this currently uncharted territory.
Search strategies and synonyms.
artificial intelligence
Open Science Framework
Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews
This work is funded through CHC’s grant as lead principal investigator from the Social Sciences and Humanities Research Council (00360-2020). RN was supported by the Wellcome Trust (213660/Z/18/Z) and the Leverhulme Trust (RC-2015-067) through the Leverhulme Centre for the Future of Intelligence.
All authors have made substantial intellectual contribution to conceptualize the protocol development. CHC, KL, JS, and AL developed the manuscript. All authors have edited, reviewed, and approved the manuscript for submission.
None declared.