Employment law and AI
these tools in both the recruitment process and deciding who deserves a promotion may lead to firms unknowingly perpetuating discrimination in both instances. For instance, Amazon was forced to abandon its resume-reviewing software due to its prejudice against women applying for executive positions. Because more men had applied in the past, the software had developed a preference towards men, even penalizing applications which used the word ‘women’ and favouring words such as ‘executed’ or ‘captured’ 13 which were more commonly used by men. AI tools have also been accused of prejudice against women when evaluating performance by counting them as less productive due to the number of maternity leave days they take being factored in as time away from work, 14 inevitably meaning many AI algorithms are more partial to promoting what they see as more efficient men, rather than women. Physical characteristics are not the only ones subject to discrimination. Since AI is primarily trained on quantitative data, it often overlooks soft skills such as communication, leadership and emotional intelligence, favouring measurable attributes like previous job titles or qualifications instead. 15 This creates a deficient assessment process that prioritizes technical credentials over holistic skills which might be far more valuable in the workplace. For instance, a candidate whose CV lacks particular terms or accomplishments valued by the algorithm may be passed over – even if they have exceptional interpersonal skills or leadership potential. This narrow approach can result in unjust outcomes that compromise equality and fairness in the workplace by discounting highly competent candidates from jobs and promotions, especially those from diverse or non-traditional backgrounds who may not have pursued traditional qualifications. In the UK, the Equality Act 2010 (section 39) is pivotal in addressing discrimination in relation to whom employers choose to offer a job, who not to offer a job to, and who to promote and mandates that applicants and employees must not be disadvantaged based on traits like age, gender, sex and race. 16
Nonetheless, the lack of opacity of many of these tools, dubbed ‘black box’ algorithms, presents an obstacle that makes this legislation by itself inadequate in preventing discrimination. These algorithms are often too complex to be understood, even by their developers, operating with little to
13 Dastin, J. ‘Insight - Amazon Scraps Secret AI Recruiting Tool That Showed Bias against Women’, Reuters , Reuters, 11 Oct. 2018, www.reuters.com/article/world/insight-amazon-scraps-secret-ai- recruitingtoolthatshowedbias-against-women-idUSKCN1MK0AG/. Accessed 5 Dec. 2024. 14 ‘Women May Pay a ‘MOM PENALTY’ When AI Is Used in Hiring, New Research Suggests | NYU Tandon School of Engineering’, Nyu.edu , 2023, www.engineering.nyu.edu/news/women-may-pay-mom-penalty-when-ai-used- hiring-newresearch-suggests. Accessed 5 Dec. 2024. 15 Rodion T. ‘Council Post: What AI Can and Cannot Do for Recruiting Today’, Forbes , 12 Aug. 2024, www.forbes.com/councils/forbestechcouncil/2024/07/24/what-ai-can-and-cannot-do-for-recruiting-today/. Accessed 12 Dec. 2024. 16 ‘Equality Act 2010’. Legislation.gov.uk , 2010, www.legislation.gov.uk/ukpga/2010/15/section/39. Accessed 8 Dec. 2024. See also Grosvenor, A. ‘The Interplay between the Equality Act 2010 and the Use of AI in Recruitment.’ Pump Court Chambers , 19 June 2023, www.pumpcourtchambers.com/2023/06/19/the-interplay- between-the-equality-act-2010-and-the-use-of-ai-in-recruitment/. Accessed 8 Dec. 2024.
94
Made with FlippingBook - PDF hosting