Accessible Explainable Artificial Intelligence in Decision Making 2022 - 2025
Consent and Inclusion, Diversity, Equity and Accessibility (2022-2023)
Guide Dog Access Barriers in Canada (2022-2023)
Accessible and Inclusive Procurement and Supply Chain Model (2021-2024)
International levels of employment study (2018)
Educational Outcomes Survey (2019)
Guide dog advocacy priorities study (2019)
Cost of Vision Loss in Canada Update (2019)
Money habits survey (2018) (Funded by Bank of Canada)
Accessible Explainable Artificial Intelligence in Decision Making (2022 - 2025)
Summary
Artificial intelligence (AI) is becoming a major tool in decision-making across various sectors such as healthcare, finance, and public services. AI systems can analyze large amounts of data quickly and efficiently, helping organizations make more informed choices. However, while AI has many benefits, it also presents challenges, especially regarding fairness, transparency, and accessibility. Many AI systems are complex and difficult to understand, which makes it harder for people to trust them. This report focuses on the importance of making AI both accessible and explainable, ensuring that it benefits everyone, including persons with disabilities.
One of the key findings of this report is that AI systems can sometimes make decisions that unintentionally discriminate against certain groups. AI models learn from large datasets, but if these datasets contain biases, the AI system can replicate and even reinforce them. This is especially concerning for individuals with disabilities, who may be underrepresented in the data used to train AI systems. If AI tools do not consider their needs, these individuals may face barriers in accessing services, employment, and healthcare. Additionally, many AI applications lack accessibility features, making them difficult to use for people with visual, auditory, or learning disabilities. This highlights the need for AI developers to prioritize accessibility from the beginning of the design process. AI should not only be designed for general usability but should also proactively remove barriers that may exist for individuals with varying abilities. Implementing inclusive design principles from the outset can ensure that AI systems are beneficial for all users.
Another critical issue is the lack of transparency in AI decision-making. AI models are often described as “black boxes” because their decision-making processes are difficult to interpret. This lack of explainability can lead to mistrust, particularly when AI is used in high-stakes areas like loan approvals, hiring, and medical diagnoses. People affected by AI decisions have the right to understand how these decisions are made. Therefore, clear and interpretable AI models are necessary to ensure fairness and accountability. Developers should aim to create AI systems that provide clear explanations of their decisions in a way that is understandable to the general public. This is particularly important for persons with disabilities, as they may need information presented in accessible formats, such as screen-reader-friendly text or simplified explanations. Further, AI explanations should be adaptable to different audiences, providing more detailed technical insights for experts while maintaining clarity and simplicity for non-specialist users. Transparency also includes allowing individuals to challenge or appeal AI decisions, ensuring accountability within AI-driven processes.
To address these challenges, this report provides several key recommendations. First, AI developers should improve transparency by designing systems that clearly explain how decisions are made. This includes using techniques like explainable AI (XAI), which allows users to see and understand the reasoning behind AI-driven outcomes. XAI methods, such as visual representations of decision pathways or step-by-step explanations, can help users grasp AI logic more easily. Second, fairness must be a top priority. Developers should conduct regular audits to identify and eliminate biases in AI models, ensuring that they do not discriminate against marginalized groups. This is especially crucial for people with disabilities, as biased AI models can lead to unfair hiring practices, denial of services, or misrepresentation in decision-making processes. It is also recommended that AI systems undergo third-party evaluations to ensure fairness and accountability, as independent reviews can help identify and mitigate unintended biases more effectively.
Another important recommendation is to make AI systems more accessible. AI tools should follow established accessibility guidelines, such as those outlined in the Web Content Accessibility Guidelines (WCAG), to ensure that people with disabilities can use them easily. This includes incorporating features like voice commands, screen reader compatibility, and adjustable text sizes. By designing AI with accessibility in mind, developers can create more inclusive systems that serve a broader population. Additionally, involving diverse perspectives in AI development is crucial. Individuals with disabilities, as well as experts in accessibility, should be included in AI design and testing processes to ensure that these systems meet the needs of all users. Their input can help identify potential barriers and suggest improvements that may not be immediately obvious to developers without lived experience. In addition to accessibility testing, AI systems should offer customization options, allowing users to adjust settings according to their unique needs. Providing adaptive interfaces can further enhance usability for a diverse range of users.
Finally, governments and organizations must establish regulations to ensure that AI is fair, explainable, and accessible. While some countries have begun implementing AI ethics guidelines, there is still a need for stronger policies that protect individuals from biased and opaque AI systems. Regulatory frameworks should require companies to disclose how their AI models function, conduct impact assessments, and implement safeguards to prevent discrimination. Organizations should also provide training for employees on how to use AI responsibly and interpret AI-driven insights accurately. Governments can play a critical role by setting industry standards, funding research into ethical AI practices, and encouraging innovation in responsible AI development. In addition, AI literacy initiatives should be promoted among the general public to ensure that individuals nderstand their rights regarding AI-driven decisions.
In conclusion, AI has significant potential to improve decision-making, but it must be designed carefully to ensure that it is fair, transparent, and accessible. AI should work for everyone, including persons with disabilities, rather than creating new barriers. By following the recommendations outlined in this report—improving transparency, ensuring fairness, enhancing accessibility, incorporating diverse perspectives, and implementing regulations—AI can become a more inclusive and trustworthy tool. Organizations, policymakers, and AI developers must work together to create AI systems that are ethical and beneficial for all members of society. Ensuring a collaborative approach where stakeholders from different sectors, including advocacy groups, technology companies, and policymakers, engage in discussions will further help shape an AI-driven future that is fair, understandable, and accessible for all.
Consent and Inclusion, Diversity, Equity and Accessibility (2022-2023)
Project Overview
To consent or not consent is a question that has perplexed individuals with disabilities for some time. There is much information about privacy and data protection in the media. The project aims to explore the experiences of individuals with disabilities relating to consent and privacy. The goal of the project is to draft a list of best practices and recommendations to improve privacy protection and the garnering of consent for individuals with disabilities, whether it be at the grocery store or in a bank. An environmental scan of the literature in the area was completed. Co-design sessions and interviews using the Delphi method were undertaken. Recommendations were developed to help privacy professionals shape their approach to ensure privacy protection for people with disabilities is front of mind in strategic planning and policy development. The following themes were highlighted using qualitative research methods, safety, authentication and verification, accessibility, plain language and self-determination. These themes provided the impetus to craft the recommendations contained within the report. It is hoped that the recommendations will lead to a systematic change in the way privacy is considered and protected for individuals who experience sight loss and people with disabilities as a whole.
Introduction
This project began in mid-September 2022 and is set for completion at the end of March 2023. The project explores the intersectionality of privacy and consent with the IDEA principles. The practical focus is on improving the methods by which consent is obtained and privacy is protected for people with disabilities, particularly those with sight loss.
Project Overview
- Conduct a literature review around conceptualizations of consent and privacy through the lens of intersecting identities.
- Carry out a sequence of stakeholder meetings and focus groups with experts in privacy and consent who have special interests in intersectionality (a Delphi method approach).
- Carry out stakeholder consultations with other community groups or organizations that represent and/or serve people with different identities, geographies, or abilities (a co-design approach).
- Carry out focus groups of CNIB’s clients, persons who are blind or partially sighted in Canada (also a co-design approach).
- Synthesise research findings from the Delphi method, co-design sessions, and literature review.
- Create a series of recommendations for implementation and dissemination to privacy professionals and the broader community of stakeholders.
- Develop a framework of recommendations for the integration of IDEA principles within CNIB’s risk management framework. In developing this framework, the anticipated outcome is the development of best practices that can be integrated into this framework.
Guide Dog Access Barriers in Canada (2022-2023)
Between March 2022 and March 2023 the CNIB Research Team undertook a study on guide dog access barriers in Canada. This study took the form of a jurisdictional scan, co-design sessions, focus groups, the formation of a 15-member Advisory committee of guide dog handlers, as well as surveys of guide dog users and the Canadian public. During the guide dog handler’s Survey respondents were asked to record their experiences with respect to accessing 12 different public spaces. 52.9% of them said that they had experienced access issues with their guide dogs. 86.6% said that they had experienced these barriers within the last 5 years, while for 13.4% of respondents the barriers had occurred more than 5 years ago. Out of the 12 public areas where access barriers had occurred, the Top 3 most impactful locations were transportation providers at 13.4%, restaurants at 9.4%, and health care services at 9.1%
As a result of this work, 23 recommendations were formulated, targeting both the governmental and private sector levels. These included asking federal standards setting organizations to ensure that they keep their “hands off our harnesses” whenever standard-setting exercises which impact guide dogs are undertaken. They asked that federal organizations such as Accessibility Standards Canada play a leadership role in resolving guide dog barriers. They called for reduced reliance on certification and identification as a way of proving the legitimacy of a working guide dog. They also recommended that inequities be eliminated for individuals who train their own guide dogs. These recommendations challenged the Canadian Human Rights Commission to lead human rights agencies across the country in exploring ways of improving the inefficiency of the current human rights process. They requested that the Canadian Transportation Agency as well as taxi and ride-share businesses step up to improve public access for guide dogs. The recommendations asked for higher penalties and more training of local police forces, while proposing that a national public education campaign on guide dog issues be undertaken, including core curriculum at all levels of the education system from primary to post-secondary programs. Finally, they asked that Canada implement the same kind of “white cane” driving laws which exist in the United States to protect pedestrians travelling with white canes and guide dogs.
Accessible and Inclusive Procurement and Supply Chain Model (2021-2024)
Project Overview
As Diversity, Equity and Inclusion become a key priority, a foundational element of accessibility is often glossed over and essential for the inclusion of persons with disabilities in society. The products and services we purchase as an organization have a direct impact on the overall participation of the stakeholders. Considering the accessibility of new products and service and supply chain diversity, whether by the public, private, not-for-profit and small and medium enterprise, while procuring them, has the potential to drive and improve participation of persons with disabilities not only in employment and community, but also help in their economic inclusion.
Project Goals
- To understand current accessible and inclusive procurement and supply chain diversity models practiced by organizations in Canada.
- To identify gold standard practices, policies, and procedures within the Canadian public and private sectors
- To build evidence and provide recommendations for the development of standards in accessible procurement
Major Accomplishments
- Completed the Literature Review and Environmental Scan of standards, policies, and legislations on international and national level.
- CNIB and some of our collaborators conducted Key informant interviews with procurement professionals' vendors/entrepreneurs with disabilities.
- Conducted Value-stream mapping exerciseswith Federal stakeholders to build evidence and provide recommendations for the development of standards in accessible procurement
- Ran a Co-design session with vendors/entrepreneurs with disabilities.
- Developed an industry survey for different stakeholders involved in procurement process
Results
The findings from the research have enabled the CNIB research team to develop recommendations aimed at standardizing accessible procurement practices across Canada. These recommendations will support the development of robust accessibility standards that ensure inclusive procurement processes, ultimately fostering an environment where individuals with disabilities can thrive professionally.
International levels of employment study (2018)
Our goal was to understand the nature of employment of persons with sight loss, as well as barriers and reasons for not participating in the labour force.
The international levels of employment study closed Canada-wide on May 7, 2018, with 1,205 survey responses (825 online, 370 by phone).
Findings from Canadian data
Employment rate: Two in five people with sight loss surveyed are employed; 3 in 10 are employed full-time. Younger people are more likely to be employed than older people and people with mild/moderate sight loss are more likely to be employed than those with severe/total sight loss. Use of braille and screen-reader usage is not associated with success in employment.
Self-employment: Older persons with sight loss are more likely to be self-employed. There is a significant segment of the population interested in self-employment as a potential career route.
Barriers to employment: Attitudes and transportation are significant barriers to being hired and promoted in the workplace for persons with sight loss.
Unemployment rate and reasons for not looking for work: The unemployment rate for persons with sight loss is 14.5% – triple the Canadian general unemployment rate. Older persons with sight loss are more likely to be out of the labour force.
Educational Outcomes Survey (2019)
Our goal was to understand the educational experiences of 18-21 year olds who experience sight loss.
High School Graduation Rates
The overall HS graduation rate of youth blind or partially sighted, 18-21 years of age, is 75%.
For youth who experience blindness or partial sight only (64% of the survey respondent population), the HS graduation rate is 88%.
For youth who experience blindness or partial sight in addition to other disabilities (36% of the survey respondent population), the HS graduation rate is 58%.
Of all high school graduates, 87.5% graduated with a high school diploma; 12.5% graduated with a certificate of completion or a GED.
76% of high school graduates reported completing high school at 18 years of age or younger; 24% reported finishing high school older than 19 years of age.
Planned Postsecondary Education
79% of respondents indicated that they planned to proceed to postsecondary education.
Of this cohort, 49% planned to attend university, 38% community college, 7% trade school and 6% private career college.
Only 62% of high school graduates were already enrolled in postsecondary education, representing 37% of those with postsecondary enrolment plans.
An addition 44% of those with plans to attend postsecondary would enroll in the next two academic years (2019 or 2020 entry); the remaining 19% of individuals planned to attend in the next 3-5 years.
Braille Use
26% of respondents indicated they used Braille.
Of those who used Braille, 34% read for pleasure, 24% used for labeling, 21% read for school/work, 10% for note taking, and 10% did not use Braille in any of these contexts.
Only 15% of students reported receiving assistance for educational specialists for Braille instruction, and 14% for Braille transcription.
Guide dog advocacy priorities study (2019)
In collaboration with CNIB Guide Dogs, the department conducted a series of focus groups to understand the advocacy challenges, barriers and opportunities that guide dog teams graduating from the program will face in the Canadian landscape.
Key Findings
- 45.3% respondents were guide dog handlers for less than 10 years and 54.7% were for ten or more years.
- Most respondents (33.3%) are between ages 51 – 63.
- Technology such as apps was used frequently by guide dog users – 96% of respondents used smartphones.
- Guide dog handlers frequently found access issues in shopping centre and avoided them.
- Guide dog handers avoided going to hotels and found that this impacted on daily life.
- Most guide dog handlers have been employed in their lifetime.
- Guide dog handlers had a current employment rate of 55% compared to white cane users who had a current employment rate of 36%.
Cost of Vision Loss in Canada Update (2019)
This is an updated study that was based on the original Cost of Vision Loss 2008 study conducted by Deloitte Access Economics, commissioned by CNIB and the Canadian Ophthalmological Society.
The total financial cost of vision loss in 2017 was $23.5 billion, consisting of a total of $10.8 billion direct costs and $12.7 billion indirect costs. The financial cost of vision loss is variable by province. The provinces that had the highest direct costs were Ontario, Quebec, and British Columbia.
Money habits survey (2018) (Funded by Bank of Canada)
This study examined the use of bank note readers, shopping habits, financial literacy and attitudes toward money of blind and partially sighted Canadians.
The money habits study closed in April 2018, with 188 responses collected over the phone in French and English.
Denomination methods
- 83% of the sample use the large numeral feature, denomination (colour), and the tactile feature as their primary method determining the face value of bank notes.
- Only 14% of smartphone users have installed an app to determine the value of currency. The main ones are Seeing AI and TapTapSee.
Bank note reader
- One in five own a bank-note reader.
- 44% of bank note reader owners were 65-plus.
- Of these owners, 9 in 10 use their reader only sparingly and mainly at home.
- Owners of bank-note readers use tactile feature, the bank note reader and third-party verification as the main ways of determining face value.
Smartphones
- Almost 3 in 4 own a smartphone.
- Those under 55 are much more likely to have a smartphone or tablet than those 55 and older.
Financial activity and literacy
- Two-thirds of households have a budget, and 9 in 10 say they have a reliable and regular income.
- The median household income was $50,000. The Canadian median household income is $70,366.
- About one in four have experienced income shortfalls in the past 12 months. While four in 10 claim they can cover living expenses for 12 months, three in 10 could only cover three months or less.
- Most have a careful and conservative approach to finances.
- Half achieved “financially literate” and half achieved “moderately literate” on the financial literacy test.
Shopping behavior and payment method
- In the past 12 months, 40% of shopping payments were made in cash, in comparison to 33% in the Canadian population. Most of the sample used cash, credit and debit as the main ways to pay.
- Most withdraw cash once a month, or more often.
- Most shop at brick-and-mortar stores. Online shopping accounts for 16% of past 12-month shopping
- Smartphone users also shop online (20% of their shopping volume is done online), more than those who do not own a smartphone (7% of their shopping volume is done online).