Making algorithms safe for workers: occupational risks associated with work managed by artificial intelligence

Abstract

It is increasingly common for companies to use artificial intelligence mechanisms to manage work. This study examines the health hazards caused by these new forms of technological management. Occupational risks can be reduced if they are taken into account when programming an algorithm. This study confirms the need for algorithms to be correctly programmed, taking account of these occupational risks. In the same way as supervisors have to be trained in risk prevention to be able to perform their work, the algorithm must be programmed to weigh up the occupational risks – and when such features do not exist, steps must be taken to prevent the algorithm being used to direct workers. The algorithm must assess all (known) factors posing a risk to workers’ health and safety. It therefore seems necessary to incorporate a mandatory risk assessment performed by specialists in the programming of algorithms so that all ascertained risks can be taken into account.

Perceived COVID‐19 health and job risks faced by digital platform drivers and measures in place to protect them: A qualitative study

“…It is possible to engineer algorithms to account for occupational risks. As argued by Todolo‐Signes, 44 algorithms can and should be programmed to take into account all known occupational health risks. The content and transparency of digital platform algorithms is an issue taken up by the European Commission's proposed directive on improving the working conditions of people working through digital labor platforms.…”

Prevalence of and risk factors for burnout and occupational stress among couriers: A systematic review

“…Both qualitative research from France (Apouey et al, 2020) and China (Lian & Zhou, 2019) found that managerial controls realised through platform technologies, such as digitalised piece rate and pay linked to customer ratings, may be associated with increased work stress among couriers. In contrast, a large number of research from sociology, human resources management, information technology and labour protection have highlighted the potential mental health risk associated with platform technology features such as gamification, algorithmic management and behavioural control (Bérastégui, 2021; Todolí-Signes, 2021; Woodcock, 2020; Woodcock & Johnson, 2018; Zhou, 2020). This factor needs to be investigated by public health research using more robust methods.…”

Algorithms and the future of work

“…96,100 By implementing algorithmic bias detection and mitigation procedures-algorithmic hygiene-and subjecting algorithms to extensive testing before use, consequential decisions negatively affecting workers can be lessened. 96,101,102…”

“…121 In the case of algorithm-enabled decision systems, the risk control strategy characterized by a human review of algorithmic outputs, called "humans-in-the-loop," is touted as an effective risk management tool. 102,122 However, humans may not be that effective at algorithmic oversight. 71 Many algorithms are inherently opaque and therefore difficult to audit.…”

“…Reliance on indiscriminate data mining may reproduce forms of discrimination set in motion by previous decision‐makers 99 and may lead to occupational inequalities in hiring, retention, and termination, primarily adversely affecting minority workers 96,100 . By implementing algorithmic bias detection and mitigation procedures— algorithmic hygiene —and subjecting algorithms to extensive testing before use, consequential decisions negatively affecting workers can be lessened 96,101,102 …”