1_Kandel.docx Bellarmine Law Society Review Volume XV | Issue II Article I Equal Access or Algorithmic Barriers?: AI and the Fight for Disability-Inclusive Hiring Valerie Kandel Cornell University, vmk29@cornell.edu EQUAL ACCESS OR ALGORITHMIC BARRIERS?: AI AND THE FIGHT FOR DISABILITY-INCLUSIVE HIRING VALERIE KANDEL1 Abstract: This paper examines how artificial intelligence (AI) hiring tools, while marketed as objective and free of bias, perpetuate structural discrimination against individuals with disabilities. By tracing the historical legacy of ableism in employment, from early personality testing to modern algorithmic screening, the paper situates AI-driven recruitment and selection within a broader pattern of exclusion. It argues that biases embedded in AI design, misrepresentative training data, and inaccessible application processes reproduce barriers that the Rehabilitation Act and Americans with Disabilities Act sought to eliminate. Through the case of Mobley v. Workday, the paper highlights the legal and ethical challenges of algorithmic discrimination, including diminished transparency, accountability, and informed consent. Ultimately, it proposes a four-part framework for disability-inclusive AI governance: increasing diversity in AI development and training, mandating auditing and impact assessments, enforcing privacy and consent protections, and requiring human oversight in employment decisions. The paper concludes that equitable AI hiring demands proactive policy intervention and renewed enforcement of disability rights principles to ensure true inclusion in the digital labor market. I. Introduction When applying for a job, most individuals are already nervous about being judged by a hiring manager. But for an individual with a disability, this feeling may be heightened by fears of discrimination in the hiring process. Biases, stereotypes, and accessibility challenges that exist in traditional hiring processes can cause many barriers for otherwise qualified individuals. When one introduces artificial intelligence (AI) hiring technology into the mix, another layer of 1 Valerie Kandel is a junior at Cornell University’s School of Industrial and Labor Relations, pursuing a Bachelor of Science with intended minors in Business, Law and Society, and Information Ethics, Law, and Policy. She serves as an Undergraduate Research Fellow under Professor Virginia Doellgast, studying the impacts of artificial intelligence on labor in the telecommunications and video game sectors. Valerie’s academic and professional interests focus on employment and disability law, as well as the governance of emerging technologies in the workplace. She plans to attend law school to further explore the intersection of law and policy with labor, technology, and workers’ rights. 6 uncertainty and the possibility of exclusion are added. AI software vendors market these tools as a solution to eliminate human biases and move towards equal treatment, but this may not be the case in practice. Take the example of a model individual, who has been diagnosed with an anxiety disorder, depression, or any other mental-health related disability. Just like anyone else looking for a job, they check popular listing websites such as LinkedIn, Indeed, or ZipRecruiter until they finally find a position that seems interesting, pays well, and matches their qualifications. The individual applies, submitting all the necessary materials, and then has to go through a sequence of hiring processes, such as a video interview, where they are scored on their answers. The “key attributes for the role, such as collaborative teamwork skills, patience in customer service, and past managerial experience” are all evaluated, not by a hiring manager, but instead by an AI algorithm.2 These tools are supposed to analyze data from application materials, including any videos or other personality assessments that candidates are invited or sometimes required to take. Based on their use of key words, phrases, or microexpressions in the written or video materials, candidates are given a score that is then used by the hiring manager or even another AI tool to select who will ultimately get the job.3 For a person with a disability, however, the experience and outcomes during the hiring process may be different. For instance, in a video interview, they might struggle to make eye contact or answer all the questions within a time limit because this creates a stressful situation or they could not focus due to an attention deficit disorder.4 Or, some video interviews or personality tests may ask about how optimistic the candidate is, which candidates with 4 Ibid.,144. 3 Ibid., 144. 2 Moss, Haley. “Screened Out Onscreen: Disability Discrimination, Hiring Bias, and Artificial Intelligence.” Disability Law Journal 4, no. 1 (January 1, 2023). https://research.ebsco.com/linkprocessor/plink?id=306c7863-ba12-35f2-8d89-4d221120d4bd. 7 depression may struggle with. The AI may then pick up on any microexpressions that display fear or a lack of confidence or comfort. Then, when it comes to deciding which candidates should be invited back for the next round of interviews, or ultimately be selected for the position, this individual may not be chosen. Disparate outcomes like this are unfortunately becoming all too common for individuals with disabilities, especially as AI becomes more ingrained within every step of the recruitment and selection processes. This was the case for Derek Mobley, an African American male over the age of forty with diagnosed anxiety and depression. Mobley had been looking for a job and continuously faced rejections, despite being qualified for the jobs he applied for. Mobley began to realize that he and likely many other candidates had been discriminated against based on factors such as race, age, and having a disability by Workday’s AI-powered applicant screening tools that many firms use for recruitment. Mobley’s legal team argued that an AI hiring tool must have been used in his application process because he received rejections outside of business hours and decisions were made a very short time after he applied. For instance, “the opinion notes one allegation that ‘Mobley received a rejection at 1:50 a.m., less than one hour after he had submitted his application.’”5 Mobley’s legal case focused on the level of human involvement compared to AI algorithms throughout these processes. His situation raised awareness about the very real and undiscussed issues with implementing AI systems in the hiring process without understanding the breadth of problems it may introduce, including the consequences of limited human oversight. Mobley’s case, which became a large-scale class action lawsuit, has the potential to 5 See, Rachel V., and Annette Tyman. “Mobley v. Workday: Court Holds Artificial Intelligence Service Providers Could Be Directly Liable for Employment Discrimination Under ‘Agent’ Theory.” Employee Relations Law Journal 50, no. 3 (December 1, 2024): 41–43. https://research.ebsco.com/linkprocessor/plink?id=6563a2aa-dcea-3ef6-8a0a-5d7c4e01570e. 8 disrupt a rapidly growing industry that has been creating disparate outcomes for minority candidates, while asserting that this is a fair and unbiased method. As of now, existing employment laws and other regulations fall short of effectively protecting marginalized groups, especially the disabled community, from the effects of discriminatory hiring practices when AI technologies are used in the hiring process.6 This has continued due to a lack of research, political discourse, and legal cases being filed against companies using AI. Since many individuals face negative outcomes related to AI, often unknowingly, there needs to be a deeper analysis and understanding of this topic. While many AI vendors and employers claim that these technologies allow for unbiased and objective sourcing, assessment, and selection of candidates, I will argue that the current uses of these tools disparately impact candidates with disabilities in hiring practices and employment outcomes. In my paper, I consider the current landscape of hiring discrimination, beginning with barriers that the disability community has faced historically, which led to the initial creation of laws aimed to protect their employment rights. Afterwards, I will critically evaluate the state of these existing laws through their gaps or failures to adequately address the needs of this community. From there, I transition to a discussion about how these existing conditions and laws fit into the new context created by AI-based hiring tools. I will examine the emergence of AI use in hiring, starting with factors that contributed to the popularity of AI for hiring purposes, and then provide a broad overview of how these technologies work, focusing on their learning or training processes. Before transitioning to how discrimination can occur through these tools, I will briefly address how companies have been implementing these AI technologies into their 6 Marshall, Romaine C., et al., “Artificial Intelligence and Employment Law.” Employee Relations Law Journal 50, no. 1 (June 1, 2024): 27–33. https://research.ebsco.com/linkprocessor/plink?id=37dcfcf4-a3e5-38ca-a0b5-63f14a317c7e. 9 hiring practices. In addressing the potential for disability discrimination, I focus on algorithmic biases that emerge from the embedded values within AI development, the effects of misrepresentative or biased data introduced during the learning process, and specific accessibility issues that may arise when these tools are used. At the end of this section, I will introduce the argument of how companies’ choices made when implementing AI hiring tools also have an impact on hiring outcomes. Subsequently, I will discuss the relevant ethical considerations behind these disparate outcomes, including a loss of privacy and a lack of informed consent leading to a power imbalance, a lack of transparency and accountability, a lack of sufficient human oversight, and a general lack of inclusivity and diversity in employment outcomes. By analyzing legal, technical, social and policy-related factors, this paper will ultimately advocate for a set of potential solutions that can be a part of a disability inclusivity framework in AI-powered hiring practices. To move toward equality in hiring practices outcomes, I will first review the recent policy initiatives to fill legal gaps, as well as guidelines by the Equal Employment Opportunity Commission (EEOC), to see how these materials set the stage for my proposed guidance or framework to address any more specific gaps. Afterwards, I intend to outline my proposed solution framework, which has been created with the previously evaluated ethical considerations in mind. This framework will focus on inclusivity and ethical application of AI technologies in hiring processes, but specifically emphasize participation and more accountability by the government, companies and AI vendors. My guidelines will prioritize and reflect the values of accessibility, transparency, and accountability for companies utilizing these technologies. Furthermore, through my chosen approach, I will promote more expansive and enforceable reforms, including mandated auditing and AI impact assessments and the requirement of human 10 involvement in hiring practices, to reinforce equity and inclusion. By addressing these critical gaps and proposing more actionable reforms, my framework aims to create a more equitable hiring landscape that upholds the rights of all members of the disability community. II. Historical Barriers to Hiring for Individuals with Disabilities This section will involve a review of the barriers that individuals with disabilities have faced in the hiring process. First, I will provide a brief overview of the legacy of selection assessments such as personality tests that have historically discriminated against members of the disability community. Though I will specify that ableist notions have existed for centuries, the majority of this analysis will focus on key legislation that aimed to prevent hiring discrimination for individuals with disabilities. Sections of the Rehabilitation Act, Americans with Disabilities Act, and its added amendments will be situated into the proper historical context. Then, I will briefly review the strengths and limitations of this existing legislation in how they protect the rights of disabled individuals. After considering all of these relevant barriers and legislative attempts to rectify them, I will transition to a discussion about how these historical issues associated with traditional hiring methods have persistent legacies in modern AI-powered hiring methods. Historically, individuals with disabilities have faced a legacy of negative stigma from “biased assumptions, harmful stereotypes and irrational fears,” causing the disability community to experience pervasive “social and economic marginalization.”7 These perceptions, which have unfortunately existed for centuries, not only questioned disabled individuals’ ability to care for 7 Anti-Defamation League. “A Brief History of the Disability Rights Movement.” ADL, November 22, 2024. https://www.adl.org/resources/backgrounder/brief-history-disability-rights-movement#:~:text=In%20the%201800s %2C%20people%20with,entertainment%20in%20circuses%20and%20exhibitions. 11 themselves, but on a larger scale, to contribute to society.8 Employment discrimination can be traced back to the industrial turn of the 19th century, when individuals with disabilities were either not permitted to work due to biases, fears, and stereotypes, instead being institutionalized for most of their lives, or being forced to “serve as ridiculed objects of entertainment in circuses and exhibitions.”9 While this overview of the profound struggles and exclusion faced by generations of people with disabilities is brief, it allows us to consider how individuals with disabilities, who are otherwise perfectly qualified to work, were historically marginalized in employment sectors and how the legacies of this have persisted within the 21st century. For the most part, disparate treatment persisted unaddressed until the 1960s when the civil rights movement began. This movement brought more awareness to the fact that minority groups, including people with disabilities, faced unfair challenges and treatment in employment due their membership in a protected class. A major contributing factor for unequal outcomes for minority job candidates was due to the fact that “in the 1960s, virtually all hiring procedures were designed with white middle-class men in mind and policymakers and testing experts recognized that new instruments needed to be created to facilitate equal access.”10 These new hiring methods emerged in the form of personality assessments, which were part of a lucrative industry that benefited employers. These tests were portrayed as a scientific hiring method, despite lacking empirical support, and instead were accompanied by a number of ethical concerns that eventually caught the attention of lawmakers.11 The first employment protections for marginalized groups were in the Civil Rights Act (CRA) of 1964, specifically Title VII, which prohibited employers “from engaging in two forms 11 Ibid., 847. 10 Kassir, Sara, Lewis Baker, Jackson Dolphin, and Frida Polli. “AI for Hiring in Context: A Perspective on Overcoming the Unique Challenges of Employment Research to Mitigate Disparate Impact.” AI and Ethics 3 (2023): 845–68. https://doi.org/10.1007/s43681-022-00208-x. 9 Ibid. 8 Ibid. 12 of discrimination: disparate treatment (e.g., intentional exclusion of a person because of their identity) and disparate impact (e.g., unintentional disadvantage of a protected class via a facially neutral procedure).”12 This act was the first of its kind to address discriminatory personality tests and disparate hiring outcomes in general, but unfortunately, its protections did not expand to cover individuals with disabilities. Continued anger with disparities in hiring was a major motivating factor for disability rights activists, who fought in the years following the CRA’s passage to create legislation that would protect the disability community. The Rehabilitation Act of 1973 was the first legislation of its kind specifically aimed at protecting the rights of individuals with disabilities to help better integrate them into the workforce. Section 504 of the Act “prohibits discrimination against individuals with disabilities in any program or activity receiving federal financial assistance.”13 While this legislation was revolutionary for its time and did have some positive effects, for the most part it fell short in regards to compliance and positively integrating individuals with disabilities. This legacy of hiring discrimination did not lessen into the 1990s, as many companies still made decisions based on the negative perceptions of hiring managers.14 Employers exhibited much “discomfort” towards job candidates with disabilities, expressing concern that hiring these individuals would lead to increased costs in accommodations.15 Despite disability rights activism and Section 504, persistent ableism within society continued to create barriers to employment for individuals with disabilities, highlighting the need for further legislation to protect the rights of the disability community. This more comprehensive law came in the form of the Americans with Disabilities Act 15 Ibid., 110. 14 McFarlin, Dean, James Song, and Michelle Sonntag. “Integrating the Disabled into the Work Force: A Survey of Fortune 500 Company Attitudes and Practices.” Employee Responsibilities & Rights Journal 4 (June 1991): 107–23. https://doi.org/10.1007/BF01390353. 13 Moss, “Screened Out Onscreen,” 167. 12 Ibid., 845. 13 (ADA) of 1990, which “made discrimination in hiring, terminations, promotions, and wages based on disability illegal [and] also required employers to provide reasonable accommodations.”16 The ADA prohibits an employer from using any selection criteria that unfairly discriminates against or screens out disabled candidates, “unless the criteria is ‘job-related’ and ‘consistent with business necessity.’”17 Additionally, sections of the ADA have made illegal standardized tests or personality assessments that discriminate against individuals with disabilities.18 After the ADA was passed, courts narrowed the intention of the Act by focusing more on individuals who could be covered by the law’s language, rather than more important protections from discrimination. The ADA Amendments Act of 2008 (ADAAA) overruled these cases and instead “expanded the definition of disability under the ADA.”19 The ADAAA made the disability definition requirement less strict and offered coverage to other impairments as they were experienced without “mitigating measures.”20 Yet, despite all this progress, research still shows that gaps persist in protections for individuals with disabilities in the workplace and in hiring. While all of these laws had positive intentions to bolster hiring and employment outcomes for individuals with disabilities, unfortunately enforcement and implementation is still inconsistent due to the presence of stigma. Employers still express concern over “the added cost of reasonable accommodations which imposes additional hiring costs,” which is not a challenge faced by any other minority group in this sector.21 Despite the many years these laws have existed, hiring discrimination has persisted, as evidenced by the U.S. unemployment rate, which 21 Ibid., 263. 20 Ibid., 262. 19 Armour et al., “Disability Saliency,” 262. 18 Ibid., 190. 17 Moss, “Screened Out Onscreen,” 189. 16 Armour, Philip, Patrick Button, and Simon Hollands. “Disability Saliency and Discrimination in Hiring.” AEA Papers and Proceedings 108 (2018): 262–66. 14 remains at 7.2% for individuals with disabilities, and is double the 3.5% rate for those without a disability.22 Statistics like this show evidence of a discrepancy in employment outcomes persisting even in the modern era. A major contributing factor in this inequality can be found due to gaps in the protections of disability discrimination legislation, which largely involve self-evaluation and voluntary compliance and therefore are difficult to critically evaluate.23 This also has repercussions for the modern employment landscape with its own unique challenges. As AI-powered recruitment and selection tools become more prevalent in hiring practices, the existing perspectives and tools remaining from the past hiring landscape must be considered and questioned in how they apply to modern practices. Traditional hiring methods were largely based on problematic hiring practices and personality tests that originally created unfair circumstances for minority candidates. Now, these traditional tools have been built into modern AI hiring systems, which use subjective measures while referring to them as objective. As these tools become increasingly popular, however, research shows they tend “to have an outsized discriminatory effect on job seekers with all types of physical and mental disabilities.”24 Such evidence demonstrates that while the technology may seem facially neutral, it encompasses the historically problematic validity and disparate impact concerns that the aforementioned laws were intended to protect against.25 There are currently few mechanisms for applying existing laws, which already have their own set of gaps and issues, to the emerging problems within AI-powered technologies. Once again, voluntary compliance seems to be the only enforcement measure in place. This is 25 Kassir et al., “AI for Hiring in Context,” 848. 24 Brown, Lydia X. Z. “Hiring Discrimination by Algorithm: A New Frontier for Civil Rights and Labor Law.” Human Rights 49, no. 1/2 (October 1, 2023): 16–18. https://research.ebsco.com/linkprocessor/plink?id=9f496cb2-c514-3553-9499-8faf2d73e14c. 23 Kassir et al., “AI for Hiring in Context.” 22 Bureau of Labor Statistics, U.S. Department of Labor. “Persons with a Disability: Labor Force Characteristics - 2023,” News Release, (2024). 15 especially concerning when we consider that most employers “had never tried to articulate their job performance goals in a systematic fashion, to develop selection devices carefully targeted to serve those goals, or to measure the success of such devices by validity studies.”26 Therefore, not only is an analysis of the existing legal protections for job candidates with disabilities necessary, but so is a critical review of these new AI hiring tools to see how policymakers can tackle new barriers that current laws fall short of addressing. III. Emergence of AI Use in Hiring This section will offer some background information into the AI-related nature of this topic, beginning with a discussion of the contributing factors that have led to AI adoption in recruitment and selection practices by top firms. Additionally, I will also briefly introduce the AI neutrality argument, before explaining, in broad terms, how AI hiring tools function, including their learning and training processes. This background emphasizes how even data-driven AI systems are rooted in human decision-making processes, spelling issues with presumed neutrality from these tools. Lastly, I will explore the different ways that AI is utilized by hiring managers, before addressing how disparate outcomes can occur along each step of the hiring process. IIIa. Contributing Factors of AI Adoption for Hiring Research from industry research leaders, such as the Society for Human Resource Management, shows that “about 79 percent of employers were using some kind of automated tool in their hiring process as of February 2022—and that was before generative artificial intelligence (AI) tools like ChatGPT were in the headlines.”27 There are many reasons why AI use has skyrocketed in recent years, but especially for recruitment purposes. AI tools have 27 Brown, “Hiring Discrimination by Algorithm.” 26 Ibid., 848. 16 benefits for improving efficiency in time-to-hire and filling empty positions rapidly, which reduces costs.28 These tools have also allowed recruiters to act more strategically and focus on “big picture items,” such as “building valuable relationships,” while AI systems focus on more menial, time-consuming tasks.29 Furthermore, AI can be implemented in all stages of the hiring process, which allows for more flexibility, personalization, and overall coordination. There are also many competitive advantages of AI, which allows hiring managers to have a larger applicant pool and use autonomous systems that can identify the most qualified candidates most efficiently.30 For top firms that want to stand out by bringing in top talent, this is a major motivating factor. Employers and AI vendors also cite another argument in favor of using AI hiring systems. They claim that “automated hiring tools increase equity by neutralizing the human factor in biased, discriminatory treatment.”31 Essentially, they argue that decreasing human involvement in decision-making also limits human biases. However, it is important to consider that “many vendors rely on poorly defined concepts of bias that obscure how AI can reflect and even exacerbate bias.”32 While many employers and “vendors contend that their tests are predictively valid and in line with business necessity, little independent evidence supports these claims,” causing AI and employment researchers to call these systems into question.33 Reviewing the AI creation and learning processes is instrumental in understanding the concerns with this neutrality theory and seeing if they actually have merit. If so, firms may be implementing a 33 Ibid., 20. 32 Tilmes, Nicholas. “Disability, Fairness, and Algorithmic Bias in AI Recruitment.” Ethics and Information Technology 24, no. 2 (n.d.). https://doi.org/10.1007/s10676-022-09633. 31Brown, “Hiring Discrimination by Algorithm.” 30 Cruz, Ignacio Fernandez. “How Process Experts Enable and Constrain Fairness in AI-Driven Hiring.” International Journal of Communication (Online) 18 (January 1, 2024): 656. https://research.ebsco.com/linkprocessor/plink?id=e514767f-4700-3f15-9a31-7d7a9f9fbb61. 29 Ibid. 28 Beaumont-Oates, William. “AI Recruitment and How It Works.” Thomas.Co, 2024. https://www.thomas.co/resources/type/hr-blog/ai-recruitment-and-how-it-works. 17 biased technology that discriminates against marginalized individuals, like those with disabilities, in violation of existing laws. IIIb. How AI Hiring Tools Work Early AI development began in the 1950s with “problem solving and symbolic” intentions and in the 1960s, expanded to creating autonomous systems that could “mimic basic human reasoning.”34 These early models developed into the advanced algorithms we see in AI today, which work by “[automating] repetitive learning and discovery through data.”35 Modern AI tools have been created to perform “frequent, high-volume, computerized tasks,” without human oversight.36 As many different types of AI systems have evolved over recent years, their capabilities in data processing and pattern recognition have proven themselves limitless in their application to almost any task. This functionality is made possible by its learning and training process. Developers teach these AI technologies by introducing essentially infinite amounts of data to their early systems. The AI gains knowledge “by combining large amounts of data with fast, iterative processing and intelligent algorithms,” and “learn[s] automatically from patterns or features in the data.”37 This is a simplified explanation of how AI tools function, but it is necessary to understand that AI systems are never truly able to create or process information or make judgments on their own. Even newer generative AI tools simply transform existing data created by humans into new forms by using existing components. Other functionalities of AI systems that may seem self-sustaining are still built on existing data and human decision-making processes that have been codified. For instance, 37 Ibid. 36 Ibid. 35 Ibid. 34 SAS Institute. “Artificial Intelligence: What It Is and Why It Matters.” SAS Institute. Accessed November 16, 2024. https://www.sas.com/en_us/insights/analytics/what-is-artificial-intelligence.html. 18 “natural language processing (NLP) is the ability of computers to analyze, understand and generate human language, including speech.”38 While developers may be building human qualities into the code of AI systems, their ability to make decisions or analyze data is still based on how data shows a human would act. Furthermore, all current AI abilities, such as computer vision, still rely on “pattern recognition and deep learning to recognize what’s in a picture or video.”39 Due to its ability to rapidly mirror human behavior in various tasks, implementation of these technologies has been increasingly widespread. Not only can their impacts be found within many different industries and companies, but also in various parts of their employment practices. IIIc. Use of AI in Recruitment and Selection Processes With the endless possibilities for AI, companies have found many ways to integrate AI technology into their hiring practices to cut costs and increase productivity. These technologies range from those developed in-house by companies for their own personal use to those by third party companies or vendors that create hiring platforms for employers. Hiring managers have been incorporating various AI systems into every stage of recruitment and selection, using these tools to assist them with initial candidate identification, as well as “outreach, screening, assessment, and [even] facilitation” throughout the hiring process.40 First, companies may utilize AI to create job descriptions to help them locate the best candidates that could potentially match a role they are looking to fill. AI tools are often used to “identify the pool of active and passive candidates (e.g., via LinkedIn) or to (re-)discover top talents in the pool of former candidates via their internal automated tracking system.”41 This may also include the “targeted advertisement of open positions” based on patterns they recognize in 41 Ibid., 992. 40 Hunkenschroer and Luetge, “Ethics of AI-Enabled Recruiting and Selection,” 991. 39 Ibid. 38 Ibid. 19 individuals on popular job platforms or from past applicants.42 Once top candidates are identified, resume reviews are used to thin out the number of potential interviews. AI implementation in this stage may involve scanning documents for key terms and recognizing important qualities which can be used “to score or rank candidates” and “[match] candidates [to] job openings to identify best fit.”43 However, newer AI technologies may “go even further and use ML to make predictions about a candidate’s future job performance based on signals related to tenure or productivity, or the absence of signals related to tardiness or disciplinary action.” 44 Next, companies will likely perform screening, for which AI is the perfect tool, since it can “complete laborious and repetitive tasks which also means it is perfect to do things such as background checks which lowers both errors and bias,” at least according to AI vendors.45 AI tools can also be used for social media screenings, which involve “scraping [and] analytics of social media postings for psychological profiles” or anything else that the employer may find problematic.46 Once the system identifies and screens the best potential candidates, companies will likely employ a variety of practices to further evaluate them before making final decisions. Candidates may be assessed in many ways, such as directly being asked to take personality tests, the results of which the employer will assess.47 There may also be “simulations, games, [or] tests” used to “assess certain skills, capabilities and traits.”48 Hiring managers may also perform a “linguistic analysis of writing samples [and] web activity.”49 AI tools may even adapt traditional interviews with more modern methods. Not only does AI increase efficiency for communication and facilitation purposes such as “setting up interview times and using chatbots,” 49 Ibid., 992. 48 Hunkenschroer and Luetge, “Ethics of AI-Enabled Recruiting and Selection,” 992. 47 Beaumont-Oates, “AI Recruitment and How It Works.” 46 Hunkenschroer and Luetge, “Ethics of AI-Enabled Recruiting and Selection,” 992. 45 Beaumont-Oates, “AI Recruitment and How It Works.” 44 Ibid.,992. 43 Ibid., 992. 42 Ibid. 20 it can even help automate interviews.50 Hiring managers can use “structured video interviews, [where] AI technology replaces a human interviewer and asks the candidate a short set of predetermined questions.”51 The AI is then used not only to “evaluate the actual responses, but also make use of audio and facial recognition software to analyze additional factors such as the tone of voice, microfacial movements, and emotions to provide insights on certain personality traits and competencies.”52 These and other methods allow employers to build an entire profile of a candidate and see how they would fit as a potential employee within their organization. By using information obtained throughout all of the prior steps of the recruitment process, employers may even use AI tools to help them ultimately select candidates. All of the aforementioned data can “feed into algorithms, and are weighed and statistically [analyzed] to make predictions about job performance.”53 This may involve candidates being scored to see how they compare to “past employees [and] testing for personality traits associated with strong performance” in that specific company.54 Companies often use this information to make final decisions about which candidates to hire. Even after firms utilize AI to help select their candidates, the involvement of AI technologies does not end there. The tools may be further used to communicate “where applicants stand in the [hiring] process” and explain their next steps as well as in “scheduling of interviews [and] sending of job offers.”55 Once candidates are selected, some firms even implement these technologies in employee onboarding, further reinforcing how prevalent AI-powered practices are within every part of the hiring process. 55 Hunkenschroer and Luetge, “Ethics of AI-Enabled Recruiting and Selection,” 992. 54 Tilmes, “Disability, Fairness, and Algorithmic Bias”, 20. 53 Kelan, Elisabeth. “Algorithmic Inclusion: Shaping the Predictive Algorithms of Artificial Intelligence in Hiring.” Human Resources Management Journal 34, no. 3 (2023): 694–707. https://doi.org/10.1111/1748-8583.12511. 52 Ibid., 992. 51 Hunkenschroer and Luetge, “Ethics of AI-Enabled Recruiting and Selection,” 992. 50 Beaumont-Oates, “AI Recruitment and How It Works.” 21 IV. Potential for Disability Discrimination in AI-Driven Hiring In this section, I intend to lay out how the algorithms, AI systems, and practices introduced in the prior sections may directly contribute to disparate outcomes for members of the disability community. First, I will begin with algorithmic biases that may inadvertently screen out disabled candidates due to designer biases that are then built or embedded into the programs, either intentionally or unintentionally. Next, I will focus on how problematic data may lead to discrimination in the hiring process as well. Additionally, I will introduce an under-researched argument against AI hiring practices, which is that these tools create accessibility issues that are not being sufficiently accommodated. I will then address how companies’ choices in how to implement AI may also contribute to disparate effects. Lastly, I will transition to a review of the ethical repercussions of this technology and how they situate the necessity for policy initiatives. IVa. Algorithmic Biases Against Disabled Candidates While AI has been intentionally designed to become largely self-sustainable, humans remain a necessary part of their functions and learning processes. Therefore, they are still responsible for values that become inherently built into the technologies, as per the embedded values theory. This theory asserts that all technology is “not morally neutral and that it is possible to identify tendencies in them to promote or demote particular moral values and norms.”56 This viewpoint considers that all technological systems contain “built-in consequence[s]” created by human decision-making and thought processes.57 If we connect this viewpoint to AI tools, because they are modeled after and taught from human data, they should be subject to critical ethical analysis. This is even more urgent if we consider that these AI tools, specifically for 57 Ibid., 42. 56 Brey, Philip. “Values in Technology and Disclosive Computer Ethics.” In The Cambridge Handbook of Information and Computer Ethics. Cambridge: Cambridge Univ Pr, 2010. https://research.ebsco.com/linkprocessor/plink?id=1dfb4e1c-59f9-31aa-8953-ac4f34ca4887. 22 hiring and decision-making, are being labelled as objective and free from “human biases” when this is very far from the truth. In practice, research shows that “AI may equally replicate and amplify such bias and embed it in technology.”58 So, while the neutrality argument examined earlier in this paper sounds ideal in theory, we cannot ignore that “technology—even and especially algorithmic technology—does not exist apart from the social, cultural, and political context in which it is created.”59 Moreover, we must realize the very real possibility that “algorithmic technologies are built on and reflect the pre-existing biases and prejudices of the people and companies that create and purchase them.”60 These biases do not just manifest themselves as AI systems automatically discriminating against candidates with certain protected characteristics. Instead, this may involve “a disproportionate distribution of prediction errors, a faulty design of the AI architecture,” or other problematic search terms or processes.61 Essentially, algorithmic biases exist when developers’ personal stereotypes or biases may have been unintentionally built into the code of AI systems. This result leads to facially neutral decision-making processes having adverse effects on a certain group. There are many striking examples of these types of algorithmic biases within AI used for hiring purposes. For instance, in one automated resume screening tool, “the two characteristics the algorithm most strongly associated with successful job performance were having the first name Jared (a name coded as white and male) and having played high school lacrosse (a sport that often connotes access to wealth privilege).”62 This shows how the biased judgements of AI developers rooted in traditional perceptions of a successful candidate may unintentionally screen 62 Brown, “Hiring Discrimination by Algorithm.” 61 Buyl, Maarten, et al. “Tackling Algorithmic Disability Discrimination in the Hiring Process: An Ethical, Legal and Technical Analysis.” Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency, June 20, 2022, 1071–82. https://doi.org/10.1145/3531146.3533169. 60 Ibid. 59 Brown, “Hiring Discrimination by Algorithm.” 58 Kelan, “Algorithmic Inclusion,” 694. 23 out marginalized candidates who do not share these experiences or qualities that are completely irrelevant to the job at hand. Furthermore, for individuals with disabilities in particular, these resume review systems may “penalize candidates for long gaps between jobs” or “for lacking leadership experience, even though people from marginalized communities might be less likely to obtain that very experience due to discrimination and exclusionary workplace cultures.”63 These few examples already show how these algorithms constantly perpetuate existing biases unintentionally held by developers within their workplaces. This also applies specifically to the topic of intersectionality, as “discriminatory patterns evidenced in automated hiring tools can impact people in every marginalized community, with an exponentially negative impact on those who belong to more than one marginalized group.”64 Currently, there are many unseen issues with current algorithms because most operate behind the scenes and candidates are not usually aware that they are being penalized by these supposedly “unbiased” AI systems for experiences that are simply a part of having a disability. As we continually realize the problems present within these systems, we should consider other ways that AI hiring methods are perpetuating problematic practices of the past. IVb. Effects of Misrepresentative or Inaccurate Data Aside from personal biases that developers may unknowingly code into AI systems, the data these tools learn from might also lead to discriminatory outcomes for the disability community. As previously explained, AI tools are designed to make decisions and recognize patterns based on data introduced during their learning processes. However, this could be problematic if we consider that the “datasets used for machine learning may contain historical biases, unrepresentative data and collection bias” and may also result in candidates with 64 Ibid. 63 Ibid. 24 disabilities being unfairly screened out.65 Issues occur when algorithms are created or trained using historical data from past or current employees from that company that does not positively or accurately reflect their experiences. For instance, if a company does not have any workers with disabilities or other protected characteristics, the system may not match these types of candidates with this position, and may rank them lower or even screen them out. Another related issue is when “underlying data may be unrepresentative of the wider population,” which means that the AI tools may be unsure how to score candidates and may just screen them out instead.66 Since white individuals’ faces are predominantly used to train AI for video interviews, research shows that these tools did not easily recognize the faces of black women and often scored them lower.67 This could also be the case for those with impairments affecting their facial appearance or expressions. Or, if interview tools do not contain data from individuals with speech or communication disorders, AI may interpret them incorrectly, which could lead to negative employment outcomes. This could also connect to collection bias issues, especially if individuals with disabilities are not being included or positively represented in these datasets because data collection methods are inaccessible. Data-related issues are a pretty large area of concern as “AI-supported hiring may thus give rise to biases owing to the domination of underlying datasets by specific groups,” once again perpetuating existing disparities.68 IVc. Review of Specific Accessibility Issues Another problem with AI-powered hiring methods are accessibility issues that could pose barriers to individuals with disabilities using such systems. It is becoming apparent that “some automated hiring tools—like gamified tests that assume a neurotypical, sighted candidate with an 68 Kelan, “Algorithmic Inclusion,” 697. 67 Ibid., 699. 66 Ibid., 699. 65Kelan, “Algorithmic Inclusion,” 697. 25 ordinary range of motion—are outright inaccessible for users with disabilities.”69 Despite little research being put into investigating this issue further, it seems rather obvious that some “disabilities may render the standard hiring process simply inaccessible, e.g., mutism if verbal communication is part of the assessment.”70 Likely due to the fast-paced and often impersonal nature of AI hiring methods, employers are clearly not putting time into finding alternative means of assessing these candidates. Since “employers may not provide adequate alternative assessments, modifications, or accommodations,” oftentimes these applicants slip through the cracks.71 This is frankly unacceptable, especially when employers are legally obligated to provide job candidates with accommodations, “even if the tools are procured through outside vendors.”72 However, once again, there is a clear lack of research investigating accessibility challenges that may result from AI hiring methods, which also presents a challenge for employers being aware of further potential of disparate impact. IVd. Implementation by Companies and its Impacts Even though algorithms are a concern, regardless of whether or not they are mitigated, “the promise of equity-driven, nondiscriminatory hiring algorithms is still a promise rather than a reality because bias can creep in consciously and unconsciously from employer practices rather than the technology itself.”73 Therefore, it is important to remember that “AI-based hiring decisions in organizations are context dependent and blend the capabilities of algorithmic- powered tools with choices and judgments made by process experts.”74 How employers and 74 Ibid., 656. 73 Cruz, “How Process Experts Enable and Constrain Fairness,” 656. 72 Engler, Alex. “The EEOC Wants to Make AI Hiring Fairer for People with Disabilities.” Brookings, 2022. https://www.brookings.edu/articles/the-eeoc-wants-to-make-ai-hiring-fairer-for-people-with-disabilities/. 71 Brown, “Hiring Discrimination by Algorithm.” 70 Buyl et al., “Tackling Algorithmic Disability Discrimination,” 13. 69 Brown, “Hiring Discrimination by Algorithm.” 26 hiring managers decide to utilize AI, such as by setting specific search or evaluation criteria, influences how their candidate pool is created. Additionally, we can also consider that the “the applied pressure for efficient, fast, and quality candidate sourcing, recruiters often trade off systematic and fair sourcing practices for inconsistent […] and implicit personal judgments or stereotypes about candidates.”75 When employers trust AI recruitment methods wholly without checking their validity, many additional ethical risks can arise. This is especially true if we consider that most existing legal protections involve some level of voluntary compliance, which is certainly not being fulfilled within the current AI-powered hiring landscape. IVe. Ethical Considerations and Necessity for Policy Change Considering all of the potential for discrimination present within these AI-powered hiring methods, a comprehensive ethical review is necessary before we can begin to address how to fill in the relevant gaps in law and policy. Now that we have reviewed the direct effects of how these tools can screen out candidates with disabilities, it is important to focus on other ethical considerations that emerge from these practices. First, many researchers discuss the loss of privacy or “lack of informed consent” that may exist surrounding AI tools.76 For AI regulation on its own, but especially in hiring, “the informed consent requirement is not yet well implemented [...] rendering the protection of personal privacy an ethical challenge.”77 One dimension of this is the “active debate about the extent to which it is ethically appropriate to use social media information for personnel selection purposes.”78 Of course, “legally, social media content is public data, but it is questionable whether it is ethical to mine social media data for hiring purposes” when consent has not been given for analysis in an 78 Ibid., 995. 77 Ibid., 995. 76 Hunkenschroer and Luetge, “Ethics of AI-Enabled Recruiting and Selection,” 995. 75 Ibid., 657. 27 employment context.79 Additionally, there could be an issue with private medical information that is submitted to these companies for accommodation purposes being utilized for AI training purposes without providers’ permission or knowledge and causing a confidentiality concern. The overall ethical issue that is created here is that “applicants in the job market generally hold less power than employers.”80 It is becoming increasingly obvious that “even if applicants are informed enough to consent to the process, they may not be able to opt out without being disadvantaged in the process.”81 Another large part of this issue is whether employees are being made aware of the fact that AI is being utilized by employers. Another ethical consideration that has been raised surrounds firms’ “ability to establish transparency by providing applicants with updates and feedback throughout the process and in a timely fashion” when they use AI hiring technologies.82 Despite the fact that creating feedback opportunities for candidates to see why they were rejected should be much easier when utilizing AI, a lack of transparency still prevails. A clear challenge presented is that “the predictive and decision-making processes of algorithms are often opaque, even for the programmers themselves.83 It has been uncommon for AI vendors or firms to provide qualitative or quantitative reports to show how these decisions are made. Researchers have expressed how this transparency is “ethically critical in the personnel selection context, due to its high relevance for people’s lives, and because this kind of black-box system may remain unchallenged, thereby obscuring discrimination.”84 The secondary part of this ethical concern is that employers feel a lack of responsibility and accountability when they use these tools, especially when they were created by vendors. Then the larger issue is that nobody is ensuring that these tools are being 84 Ibid., 997. 83 Ibid., 997. 82 Ibid., 997. 81 Ibid., 995. 80 Ibid., 995. 79 Ibid., 995. 28 fairly used for decision-making or questioning who would be liable for disparate outcomes.85 Since AI tools themselves obviously cannot be held accountable, many ethicists assert that “it should be a human agent who is ultimately responsible for the decision made when selecting an employee.”86 This shows another underlying issue, which is the lack of human oversight in these procedures. Though it is difficult to identify whether a human or a computer is making the final decision in most of these companies’ practices, many claim “that AI has already taken over the automated decision-making process, forwarding or rejecting candidates.”87 If human intervention in these procedures continues to decline, this could mean other issues for individuals facing disparate outcomes, as there would be even less oversight for equality in hiring practices. The overarching concern accompanying all of these considerations is how these tools can result in a decreased level of diversity and representation within companies. AI ethicists are concerned that “a systematic bias through AI could result in more homogeneity in organizations.”88 This is especially true if we consider that “a single decision-making algorithm” is making decisions based on code created by developers or data that is not necessarily inclusive or representative of diverse populations.89 Also, these tools may be replacing a team of “several human decision makers with potentially differing views” which may also lead to less workplace diversity.90 The prevailing worry from “disability advocates is that people with disabilities will be discouraged by digital assessments and drop out of the application process” completely.91 Since firms may also be losing out on top talent if qualified applicants are being unfairly 91 Engler, “The EEOC Wants to Make AI Hiring Fairer.” 90 Ibid., 994. 89 Ibid., 994. 88 Ibid., 994. 87 Ibid., 997. 86 Ibid., 997. 85 Ibid., 997. 29 screened out for having disabilities, this creates a situation in which everyone faces a loss that should certainly be rectified by policy initiatives. V. Potential Solutions and Framework for Improvement In this final section of the paper, I will consider how all the arising issues evaluated previously and their accompanying ethical considerations spell out the need for policy initiatives that protect job candidates with disabilities during the hiring process. From there, I will briefly review recent policy and guidance frameworks which will assist me as I begin to create my own framework that would promote equitable hiring practices by adapting existing research into a set of recommended policy initiatives. Va. Review of Recent Policy and EEOC Guidance There have been attempts to fill the existing gap in legal protections surrounding AI-powered hiring methods; however, these have largely fallen short of addressing the needs of individuals with disabilities. For instance, in 2021 “the New York City Council passed the Automated Employment Decision Tool Law (AEDT),” which was credited as “a first-of-its-kind law on AI hiring discrimination.”92 While this law “stipulates that employers must conduct third-party bias audits for discrimination,” which is a great start, the language only includes “race, ethnicity, or sex,” which evidently fails to protect individuals with disabilities.93 Based on the existing laws, “employers have no obligation to provide any meaningful notice, explanation, or opt-out process to job seekers and no obligation to regularly audit their software for discriminatory impact using external experts or to report the results of these audits and remediate 93 Ibid. 92 Brown, “Hiring Discrimination by Algorithm.” 30 accordingly.”94 Further, a concern of disability activists is that when laws fall short of protecting marginalized groups, companies are “proliferating the market with tools that comply with the letter of the law but nonetheless discriminate.”95 So, it is important to consider the strengths of newer laws that, in theory, could fill in gaps left by existing legislation, and to examine how they can be improved to protect individuals with disabilities during all stages of the hiring process impacted by AI tools. In addition to frameworks within these insufficient but well-intentioned recent laws, we can also consider guidance present within the EEOC’s most recent set of AI-related employment guidance, which reviews how these systems may violate the ADA’s accommodation and equal treatment requirements and how they plan on addressing them. Within this framework, “the EEOC recommends that employers train staff to quickly recognize and respond to accommodation requests with alternative methods of candidate evaluation, and notes that outsourcing parts of the hiring process to vendors does not automatically relieve the employer of its responsibilities.”96 While most of the EEOC recommendations address employer-related practices, as they are “ultimately responsible for ADA compliance,” there is some guidance or discussion surrounding the practices of AI vendors.97 This involves their assertion that even “when an algorithmic tool is ‘validated’ according to a vendor, it does not provide inculpability from discrimination” and employers should be sure to still question vendors to ensure they are also being accountable.98 Furthermore, a review of this guidance shows that these recommendations alone may “help employers make fairer choices, but the EEOC does not seem to be purely counting on the 98 Ibid. 97 Ibid. 96 Engler, “The EEOC Wants to Make AI Hiring Fairer.” 95 Ibid. 94 Ibid. 31 good graces of employers to execute the changes it thinks are necessary.99 Instead, they “[provide] recommendations for job applicants who are being assessed by algorithmic tools,” which involves encouragement to “file formal charges of discrimination with the EEOC if a candidate feels they were discriminated against by an algorithmic hiring process.”100 From there, they can conduct an investigation, then “[try] to negotiate an agreement, and failing that, may file a lawsuit against the employer” on the grounds that their tools were discriminatory and candidates may be entitled to damages.101 This guidance is certainly a good start for policy, but more strict regulations expanding upon them are necessary. Vb. Proposed Solution Framework It is clear that organizations’ and vendors’ accountability is important and growing in discussion due to a sense of urgency in creating equality in an AI-powered hiring environment that is growing more popular by the day. Therefore, to address the root issues present in widespread use of AI-powered technologies in hiring, there needs to be a set of policies or initiatives implemented by the government, companies, and AI vendors. First, I recommend that AI vendors increase diversity and inclusivity in the development of AI tools and through the resources used during learning processes. Initially, it would involve creating teams of more diverse individuals to make sure different backgrounds are represented in systems created.102 This would directly target algorithmic biases, especially if these individuals are educated about the risks of implicit and explicit biases that may be built into their programs so they can prevent these issues ahead of time or raise questions when they arise through internal auditing procedures.103 One successful approach has been seen in “AI software vendors 103 Moss, “Screened Out Onscreen,” 195. 102 Hunkenschroer and Luetge, “Ethics of AI-Enabled Recruiting and Selection,” 999. 101 Ibid. 100 Ibid. 99 Ibid. 32 [removing] any wording or phrases that can unconsciously predict the gender of a candidate from CVs to circumvent unconscious bias and improve equity.”104 If developers are able to be self-aware about potential problematic areas within their tools to protect individuals with disabilities in particular, this could have many positive outcomes. Additionally, there should be a stronger effort to use more relevant and accurate data sources. However, this may connect to another issue of limited existing data surrounding the disability community, but it is also possible for them to weigh the data differently so it does not unintentionally discriminate against marginalized groups. For instance, there has been success in using “inverse weight propensity scores to re-balance groups for instance by taking into account how many black women older than 30 are in the dataset and then balance results internally.”105 Overall, more awareness of disability justice and intersectionality considerations during the early stages of AI development, even before the issues with implementation begin, would be an asset to preempt discriminatory outcomes.106 For the second prong of my recommended framework, external laws and internal policies must be created requiring accountability and transparency from companies. Many of the current issues surrounding AI in hiring could only be “addressed through rigorous coding protocols, job analysis and regular auditing of algorithms.”107 Therefore, stricter legislation and company policies should be created and enforced, requiring AI impact assessments and regular auditing to prevent disparate outcomes for the disability community. Research has suggested that risks can never be completely eliminated from AI tools, even with additional accountability from vendors. Therefore, as an additional check to ensure equality with these tools, AI ethicists assert that 107 Kelan, “Algorithmic Inclusion,” 701. 106 Moss, “Screened Out Onscreen,” 194. 105 Kelan, “Algorithmic Inclusion,” 701. 104 Hunkenschroer and Luetge, “Ethics of AI-Enabled Recruiting and Selection,” 999. 33 “technical due diligence regarding algorithmic design and implementation is crucial to keep this risk low.”108 Even when companies buy AI tools from vendors, “practitioners are strongly encouraged to refer to professional test standards and obtain critical information about the tools: for example, evidence that informs psychometric reliability, criterion-related validity and bias implications.”109 In simple terms, companies need to provide reports that explain what search terms they use, “why a candidate has been selected and the causality regarding which specific attributes can be associated with their success in a role.”110 A consistent theme among these recommendations involves ensuring explainability, transparency, and overall accountability from companies along every step of the process, as well as enforcing the legal requirement to provide accommodation, which should be more strictly regulated even within AI contexts. For the third recommendation, there should be stricter laws and policies surrounding privacy and informed consent created and implemented by companies and vendors, just as they have to comply with traditional hiring practices.111 Part of this involves private data being kept private by companies and not being included in AI hiring practices to evaluate or analyze candidates. This includes social media information and medical data alike, which should not be used for hiring purposes unless candidates are made aware and give their explicit consent. Lastly, “it should be always transparent to applicants whether they are communicating with another human or with AI” or if AI is being used to evaluate them, to increase the overall level of informed consent within these processes.112 This would also help candidates remain vigilant so they can take appropriate action, such as reporting to the EEOC, if they feel that they have been unfairly discriminated against. 112 Ibid., 999. 111 Hunkenschroer and Luetge, “Ethics of AI-Enabled Recruiting and Selection,” 999. 110 Ibid., 1000. 109 Ibid., 999. 108 Hunkenschroer and Luetge, “Ethics of AI-Enabled Recruiting and Selection,” 994. 34 Lastly, for the fourth prong, laws should reinforce human involvement in hiring processes. Cases such as Mobley v. Workday reinforce that allowing an AI tool to make final employment decisions is unacceptable and should be illegal. Therefore, a human review of all decisions made using AI would be necessary to ensure that ultimate candidate selection is not being unfairly determined by problematic AI tools. The most useful method would likely be the requirement of “AI ethics board with an oversight function[s], consisting of representatives of relevant stakeholders who debate the data and ethical dimensions of AI algorithms and agree on boundaries for AI technology in the company.”113 These boards, consisting of a diverse group of individuals, would perform “ethical audits” to ensure that ethical standards are being met and that no one faces disparate outcomes. Another element involves the requirement that AI vendors build into their tools the “ability to judge whether it can grant adequate accommodation.”114 If not, vendors should be required to inform companies using their tools to stay vigilant and make sure all users are accommodated.115 This would limit adverse effects arising out of the impersonal and fast-paced nature of AI and confirm that these methods align with existing legal protections. VI. Conclusion As I have argued in this paper, the topic of AI and its rapidly growing negative impact on the disability community must be addressed. It is evident that the current state of legislation, from more outdated legislation to recent attempts to fill gaps, still falls short of making any real, lasting, and positive change. Based on the prevalence of AI in many companies, along every step of the hiring process, and on the legacies of human biases and personality tests that have been 115 Ibid., 14. 114 Buyl et al., “Tackling Algorithmic Disability Discrimination,” 14. 113 Ibid., 999. 35 built into these tools, there is a clear potential for discriminatory outcomes. Due to the clear barriers of perpetuated bias, improper data being used, and accessibility concerns, there needs to be a reaffirmed emphasis on inclusive hiring practices when AI is utilized during hiring. With this in mind, it is important that nobody slips through the cracks and is unfairly screened out by AI systems or practices. To review the example of Derek Mobley’s case, the plaintiff succeeded in showing that Workday was still liable for the disparate outcomes, despite not being an employer but instead a vendor for businesses, which “opens the door for a significant expansion of liability” for AI vendors in the hiring process.116 Still, this decision reinforced how candidates may not even realize anything is amiss when instances of disparate impact are occurring behind the scenes. Drawing from Mobley’s example, individuals with disabilities should also be aware of how AI tools can potentially discriminate against them based on their protected characteristics. This case also opens the doors for a very important discussion about how existing laws and regulations protect individuals from discrimination that may occur due to AI technology. Furthermore, it is imperative for lawmakers to review the functions of AI, including how they work, and how implicit biases may be built into their algorithms, or how disparities may occur when they are implemented by companies. The potential for inclusive and fair hiring practices using AI-powered tools exists but requires more transparency, accountability, and use of positive data. Further, it necessitates more mindfulness by every involved party in considering how it could have disparate outcomes on the disability community. Additionally, it is important that we remember that concerns about biases within AI “[ignore] the fact that the original source of algorithmic bias is the human behavior it is simulating.”117 If we are concerned about the decisions being made by AI, we should first 117 Ibid., 994. 116 See and Tyman, “Mobley v. Workday,” 43. 36 address the errors in human behavior within society that it unintentionally mirrors.118 While my framework of suggestions would mitigate discriminatory effects from these tools, only once individuals with disabilities experience equal outcomes in hiring and employment can these biases be fully eliminated from AI-powered methods. 118 Ibid., 994. 37