Article type
Abstract
Background:
Adoption of the use of artificial intelligence (AI) to facilitate decision-making in policy is growing rapidly and has raised many ethical, social, and economic questions. Deployment of AI tools is reported in many areas of public policy, including education, healthcare, policing welfare, and immigration. Issues surrounding its transparency and the risks to human rights, equity, potential for bias, and furthering marginalization of certain populations make AI use in decision-making for policy a potentially problematic issue.
Objectives:
To create a map of the landscape of equity studies in AI and provide insight into the nature of the disciplines exploring AI and the types of tools being used or developed
Methods:
A systematic search was conducted across 21 scholarly databases covering a broad spectrum of disciplines: science, social science, health, business, politics, education, justice, and computer science. Included articles identified in the search were piloted with the inclusion/exclusion criteria, exported into EPPI-Reviewer, de-duplicated, and screened for relevance for inclusion in the map. Studies were coded for the area/discipline of the implementation of AI and types of equity issue addressed. An AI tool was developed to assist in the screening and coding of records.
Results:
The search identified 26,166 records, and of these, 5914 references met the inclusion criteria for the map. General bias and inequalities in AI in the area of computer science (n = 3395) and health (n = 1582) accounted for the majority of studies, with significant gaps identified for many underrepresented and marginalized populations. There were fewer than 50 systematic or scoping reviews, and these mainly focused on gender or race.
Conclusions:
The map has identified a rich, complex, diverse, and rapidly expanding landscape across many areas of public policy. Due to the size and extent of the literature, AI itself is needed to help facilitate reviewers complete work in this area at the pace in line with the expansion of the topic.
Relevance:
To improve understanding of inequity in AI, studies—particularly further evidence syntheses—are needed that focus on the impact of equity in the use of AI tools across the full range of underrepresented or marginalized groups.
Adoption of the use of artificial intelligence (AI) to facilitate decision-making in policy is growing rapidly and has raised many ethical, social, and economic questions. Deployment of AI tools is reported in many areas of public policy, including education, healthcare, policing welfare, and immigration. Issues surrounding its transparency and the risks to human rights, equity, potential for bias, and furthering marginalization of certain populations make AI use in decision-making for policy a potentially problematic issue.
Objectives:
To create a map of the landscape of equity studies in AI and provide insight into the nature of the disciplines exploring AI and the types of tools being used or developed
Methods:
A systematic search was conducted across 21 scholarly databases covering a broad spectrum of disciplines: science, social science, health, business, politics, education, justice, and computer science. Included articles identified in the search were piloted with the inclusion/exclusion criteria, exported into EPPI-Reviewer, de-duplicated, and screened for relevance for inclusion in the map. Studies were coded for the area/discipline of the implementation of AI and types of equity issue addressed. An AI tool was developed to assist in the screening and coding of records.
Results:
The search identified 26,166 records, and of these, 5914 references met the inclusion criteria for the map. General bias and inequalities in AI in the area of computer science (n = 3395) and health (n = 1582) accounted for the majority of studies, with significant gaps identified for many underrepresented and marginalized populations. There were fewer than 50 systematic or scoping reviews, and these mainly focused on gender or race.
Conclusions:
The map has identified a rich, complex, diverse, and rapidly expanding landscape across many areas of public policy. Due to the size and extent of the literature, AI itself is needed to help facilitate reviewers complete work in this area at the pace in line with the expansion of the topic.
Relevance:
To improve understanding of inequity in AI, studies—particularly further evidence syntheses—are needed that focus on the impact of equity in the use of AI tools across the full range of underrepresented or marginalized groups.