Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reporting Risks in AI-based Assistive Technology Research: A Systematic Review

Zahra Ahmadi, Peter R. Lewis, Mahadeo A. Sukhai

TL;DR

This systematic literature review investigates how failures and risks are explored, assessed, and reported in AI-based assistive technologies for visually impaired users. By screening 648 papers and analyzing a representative subset of 100, the authors find that while most papers showcase working demos, human studies with sight-loss participants are relatively rare, and discussions of validity threats and failure consequences are largely absent. Although many studies acknowledge potential failures through metrics, explicit failure cases and scenario-based risk discussions are uncommon, highlighting a need for standardized reporting and more inclusive, ecologically valid evaluations. The work argues for guidelines to improve risk transparency and trust in AI-based assistive tools, with the broader aim of ensuring safer, more reliable technology for the vision-impaired community.

Abstract

Artificial Intelligence (AI) is increasingly employed to enhance assistive technologies, yet it can fail in various ways. We conducted a systematic literature review of research into AI-based assistive technology for persons with visual impairments. Our study shows that most proposed technologies with a testable prototype have not been evaluated in a human study with members of the sight-loss community. Furthermore, many studies did not consider or report failure cases or possible risks. These findings highlight the importance of inclusive system evaluations and the necessity of standardizing methods for presenting and analyzing failure cases and threats when developing AI-based assistive technologies.

Reporting Risks in AI-based Assistive Technology Research: A Systematic Review

TL;DR

This systematic literature review investigates how failures and risks are explored, assessed, and reported in AI-based assistive technologies for visually impaired users. By screening 648 papers and analyzing a representative subset of 100, the authors find that while most papers showcase working demos, human studies with sight-loss participants are relatively rare, and discussions of validity threats and failure consequences are largely absent. Although many studies acknowledge potential failures through metrics, explicit failure cases and scenario-based risk discussions are uncommon, highlighting a need for standardized reporting and more inclusive, ecologically valid evaluations. The work argues for guidelines to improve risk transparency and trust in AI-based assistive tools, with the broader aim of ensuring safer, more reliable technology for the vision-impaired community.

Abstract

Artificial Intelligence (AI) is increasingly employed to enhance assistive technologies, yet it can fail in various ways. We conducted a systematic literature review of research into AI-based assistive technology for persons with visual impairments. Our study shows that most proposed technologies with a testable prototype have not been evaluated in a human study with members of the sight-loss community. Furthermore, many studies did not consider or report failure cases or possible risks. These findings highlight the importance of inclusive system evaluations and the necessity of standardizing methods for presenting and analyzing failure cases and threats when developing AI-based assistive technologies.
Paper Structure (26 sections, 11 figures, 2 tables)

This paper contains 26 sections, 11 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Research Questions
  3. Questions
  4. Definitions and Scope
  5. Related Work
  6. Methodology
  7. Systematic Search and Data Extraction
  8. Inclusion/Exclusion and Sampling
  9. Sample papers sets
  10. Sample sets validity
  11. Questions validation
  12. Results and Analysis
  13. Systematic search results
  14. Systematic review results
  15. Question 1: Does the paper present a working demo or examples of how the system behaves?
  16. ...and 11 more sections

Figures (11)

  • Figure 1: Question flow chart - Questions 1, 2, and 5 are represented as conditional nodes. If Question 1 is answered 'yes,' the flow continues to Question 2; if 'no,' it jumps directly to Question 5. Similarly, if Question 2 is answered 'yes,' the flow continues to Questions 3 and 4 before moving to Question 5; if 'no,' it also jumps directly to Question 5. If Question 5 is answered 'yes,' the flow proceeds to Questions 6, 7, and 8. Each decision point is guided by a 'yes' or 'no' response, leading to a structured progression through the research questions.
  • Figure 2: Bar chart showing the yearly distribution of total papers and sample papers, spanning from 2017 to 2022. The x-axis represents the years in descending order from 2022 to 2017, while the y-axis indicates the percentage of papers. The bars compare the percentage of papers in the sample set to the total set for each year. The closeness of percentages between the sample and total sets for each year shows our sample papers are representative of the total number of papers. Also, it shows that the number of papers has increased over the years.
  • Figure 3: Final results for two sets of papers
  • Figure 4: The number of papers in the combined set according to their answer to Question 1: 85 Yes, and 15 No
  • Figure 5: The number of papers in the combined set according to their answer to Question 2: 27 Yes, and 73 No
  • ...and 6 more figures