Algorithms and the Future of Work

An algorithm is a series of precise, step-by-step instructions used by a machine to perform a mathematical operation.[1] The use of algorithm-enabled systems and devices will bring many benefits to occupational safety and health but, as with many new technologies, there are also risks to workers. A new commentary in the American Journal of Industrial Medicine focuses on new sources of worker risk that algorithms present in the use of worker management systems, advanced sensor technologies, and robotic devices. Determining if an algorithm is safe for workplace use is rapidly becoming a challenge for manufacturers, programmers, employers, workers, and occupational safety and health practitioners. To achieve the benefits that algorithm-enabled systems and devices promise in the future of work, we must study how to effectively manage their risks. Key points from the commentary are highlighted below. An in-depth discussion is available in the article.

Algorithms in the Workplace

Machine learning algorithms are powering various occupational safety and health applications across several industry sectors.[2] Algorithm applications are found in manufacturing;[3, 4] construction;[5] agriculture;[6] extractive mining;[7] retail;[8] and public governance.[9] Data-driven insights powered by algorithm-enabled systems and devices can be conceptualized as future-of-work tools in occupational safety and health that may one day tell you what happened (descriptive systems) and why it happened (diagnostic systems); forecast what will happen (predictive systems); support decision-making based on present and future conditions (prescriptive systems); and take physical actions (semi-autonomous and autonomous systems).[10-13]

The expected benefits of integrating algorithms into workplace equipment, processes, conditions, and human management systems should be tempered by a full awareness and understanding of their risk profile. Understanding the risks and benefits of algorithm-enabled workplace systems should be based on a comprehensive risk evaluation.[14] Risks posed by algorithm-enabled systems generally originate in three areas:

(1) errors and biases in the input or training data;

(2) flaws in the design of the algorithm or mistakes in coding the algorithm into programming language; and

(3) user disregard of an algorithm’s limitations or underlying assumptions, leading to inappropriate application or incorrect interpretation of system outputs or decisions.[15]

The increasing complexity of proprietary algorithms—especially self-learning algorithms which can change their decision logic during operation—make it difficult for designers, manufacturers, and users to gain an operational understanding about how an algorithm works.[16,17] Lack of algorithmic transparency can be a major impediment to the assessment and control of new occupational safety and health risks.¹⁴ As algorithmic decision making is increasing in various societal systems,[18] and in worker management systems, advanced sensor technologies, and robotic devices, [2] attention is focused on ways to attain greater algorithm transparency.[19 –21]

Algorithmic Management

Close physical supervision has been the traditional way that employers have monitored their workers. Employers can now monitor workers by means of video surveillance; track a worker’s physical movements through geolocation algorithms; monitor an employee’s use of email, social media, and web browsing; assess a worker’s productivity, level of engagement, propensity to leave the organization, and adherence to workplace safety behaviors.[22-24] These new data-driven approaches to human resources management are referred to as “people analytics” and are touted as helping employers make better decisions.[25]

Algorithmic management techniques can collect and store worker data on a continuous basis, potentially without express purpose or worker disclosure.[26] In some algorithmic management technologies, the observer of the worker and the decision-maker can both be non-human agents.[27–29] Algorithmic-enabled productivity and performance systems often represent a type of management control without worker consent when surveillance is not prospectively disclosed to workers.[30] When algorithms are given power over a worker’s job, and when the worker has no information or understanding of what data the algorithm is collecting, how the data are being used, and for what purpose, workers report feelings of powerlessness.[31, 32] This is not surprising since under algorithmic management workers often have no meaningful interaction with their “digital supervisor.” [27, 33]  This algorithmic management can be associated with the erosion of worker autonomy, work intensification, psychosocial stress, and a decline in worker well-being.[34–36]

Algorithmic management is especially pervasive in the gig economy,[37–39] but digital surveillance and management technologies are also seen across other industry sectors.[40, 41]  Shift allocation algorithms, delivery route algorithms, warehouse workers movement algorithms, continuous performance algorithms, and other work productivity algorithms are being applied not only to manufacturing workers, but also to service workers, knowledge workers, warehouse workers, and even to first-line supervisors. [32]

New algorithmic technologies have the potential to significantly transform organizational control by affecting the employer-employee relationship.[42] The diffusion of algorithmic management systems will affect the future of work but may do so in unintended and undesirable ways.[43]  While the limits of what is private about a worker under algorithmic management are currently uncertain, the presence of large amounts of information about a worker within an algorithmic-enabled system presents a potential security risk. Additionally, algorithms that are used to automate organizational management systems may produce discriminatory outcomes that can reproduce and reinforce society’s historical age, racial and ethnic, and gender biases, among others.[44,45]

Advanced Sensor Technologies

Sensors are at the heart of technology-managed work as they provide the data inputs for algorithmic controls. Advanced sensor technologies are being commercialized and entering the workplace as new exposure science tools.[46] Advanced sensor technologies using miniaturized algorithm-embedded microprocessors have the potential to greatly accelerate advances in occupational exposure science by continuously sensing the ambient work environment for hazardous substances or a worker’s proximity to known hazards.[47] However, they can also raise worker concerns over intrusive worker surveillance, algorithmic bias, and violation of personal privacy.[48]

Robotics

Algorithms are critical components of all robotic devices. From an AI architectural perspective, robotic devices can be physically-embodied robots or digital-decision assistants. Robots exhibit three major functions: they sense, plan, and act. Algorithms are involved in all three of these essential robotic functions. As with algorithmic management systems, work intensification may occur when a robotic system under algorithmic control causes a mismatch between a human worker’s physical or cognitive capabilities and work demands. When robotic systems are designed to maximize productivity without adequately considering the impact on human workers’ performance, their risk profile increases. While integration of robotics into work processes promises many productivity benefits, workers may face the risks of work intensification[49] and job displacement[50] from their use.

Governance

Introduction of algorithm-enabled AI systems and devices for workplace use is accelerating faster than algorithm-specific risk assessment and risk management strategies can be developed. [2, 51] When integrated into workplace systems, algorithms can present a unique taxonomy of risks that may not be addressed in an organization’s traditional occupational safety and health risk management approaches.[14] New methods are needed to detect biases in input data, find design errors in proprietary algorithms, and ensure that output decisions are logically derivative of the input data. [15] The risks arising from invasive surveillance, algorithmic bias, loss of autonomy and privacy, inaccurate decision outputs, and work intensification should be added to existing risk assessment and management approaches. Workers should also have latitude and method to challenge algorithm-generated decisions.[52]

Conclusion

In the future of work, algorithms will provide many beneficial applications in occupational safety and health. While algorithm-enabled systems and devices may reduce sources of human error and enhance worker safety and health, algorithms may also introduce new sources of risks to worker wellbeing. Determining that an algorithm is safe for use in a worker management system, in advanced sensor technologies, in robotic devices, and in other workplace systems, tools and machinery will challenge the risk assessment and management capabilities of algorithm designers and software programmers, algorithm-enabled equipment manufacturers, employers, workers, and occupational safety and health practitioners. To ensure that the benefits of algorithm-enabled systems and devices have a prominent place in the future of work, now is the time to study how to effectively manage their risks.

John Howard, MD, Director, National Institute for Occupational Safety and Health

References

[1] Louridas P. Algorithms. Cambridge, MA: The MIT Press; 20206

[2] Pishgar M, Issa SF, Sietsema M, Pratap P, Darabi H. REDECA: a novel framework to review artificial intelligence and its application in occupational safety and health. Int J of Env Res Pub He. 2021; 18: 6705-6742. https://doi.org/10.3390/ijerph/18136705

[3] Wurst T, Weimer D, Irgens C, Thoben K-D. Machine learning in manufacturing:advantages, challenges, and applications. Prod Manu Res. 2016;4(1):23-45.  https://doi.org/10.1080/21693277.2016.1192517

[4] Raia R, Tiwarib MK, Ivanovc D, Dolgui A. Machine learning in manufacturing and industry 4.0 applications. Int J Prod Res. 2021;59(16):4773-4778. https://doi.org/10.1080/00207543.2021.1956675

[5] Baduge SK, Thilakarathna S, Perera JS, et al. Artificial intelligence and smart vision for building and construction 4.0: machine and deep learning applications. Automat Constr. 2022; 141: 104440. https://doi.org/10.1016/j.autcon.2022.104440

[6] Benos L, Tagarakis AC, Dolias G, Berruto R, Kateris D, Bochti D. Machine learning in agriculture: a comprehensive updated review. Sensors. 2021; 21: 3758-3813. https://doi.org/10.3390/s21113758

[7] Wojtecki L, Iwaszenko S, Apel DB, Bukowska M. Use of machine learning algorithms to assess the state of rockburst hazard in underground coal mine openings. J Rock Mech Geotech Eng. 2021; 14: 703-713. https://doi.org/10.1016/j.jrmge.2021.10.011

[8] Huber J, Stuckenschmidt H. Daily retail demand forecasting using machine learning with emphasis on calendric special days. Int J Forecast. 2020; 36(4) :1420-1438.  https://doi.org/10.1016/j.ijforecast.2020.02.005

[9] Zuiderwijk A, Chen Y-C, Salem F. Implications of the use of artificial intelligence in public governance: a systematic literature review and a research agenda. Gov Inform Q. 2021; 38: 101577-101596. https://doi.org/10.1016/j.giq.2021.101577

[10] Farias H. Machine learning vs predictive analytics: what’s the difference? Concepta. October 10, 2017. Accessed August 30, 2022. https://perma.cc/6QAS-MB47

[11] Lepenioti K, Bousdekis A, Apostolou D, Mentzas G. Prescriptive analytics: literature review and research challenges. Int J Inform Manage. 2020; 50: 57-70. https://doi.org/10.1016/j.ijinfomgt.2019.04.003

[12] Harel D, Marron A, Sifakis J. Autonomics: in search of a foundation for next-generation autonomous systems. P Natl Acad Sci USA. 2020; 117(30): 17491-17498.  https://doi.org/10.1073/pnas.2003162117

[13] Saveski M, Awad E, Rahwan I, Cebrian M. Algorithmic and human prediction of success in human collaboration from visual features. Sci Rep. 2021; 11: 2756-2769. https://doi.org/10.1038/s41598-021-81145-3

[14] Steimers A, Schneider M. Source of risk of AI systems. Int J Env Res Pub He. 2022; 19: 3641-3673.  https://doi.org/10.3390/ijerph19063641

[15] Krishna D, Albinson N, Chu Y. Managing algorithmic risks, safeguarding the use of complex systems and machine learning. 2017. Deloitte & Touche, LLP. Accessed August 30, 2022. https://www2.deloitte.com/content/dam/Deloitte/us/Documents/risk/us-risk-algorithmic-machinelearning-risk-management.pdf

[16] Davenport TH, Ronanki R. Artificial intelligence for the real world. Harvard Bus Rev. 2018. January‐February. Accessed August 30, 2022. https://www.hbsp.harvard.edu/product/R1801H-PDFENG

[17] Tsamados A, Aggarwal N, Cowls J, et al. The ethics of algorithms: key problems and solutions. AI Soc. 2022; 37: 215-230. https://doi.org/10.1007/s00146-021-01154-8

[18] Barfield W, Barfield J. An introduction to law and algorithms. In: Barfield W, ed. The Cambridge Handbook of the Law of Algorithms. Cambridge, UK: Cambridge University Press; 2020:1-15. https://doi.org/10.1017/9781108680844

[19] Diakopoulos N. Accountability in algorithm decision making. Commun ACM. 2016; 59(2): 56-62. https://doi.org/10.1145/2844110

[20] Polack P. Beyond algorithmic reformism: Forward engineering the designs of algorithmic systems. Big Data Soc. 2020; January-June: 1-15. https://doi.org/10.1177/2053951720913064

[21] Koene A, Clifton C, Hatada Y et al. A Governance Framework for Algorithms Accountability and Transparency. European Parliamentary Research Service, Panel for the Future of Science and Technology. April 2019. Accessed August 30, 2022. https://www.europarl.europa.eu/RegData/etudes/STUD/2019/624262/EPRS_STU(2019)624262_EN.pdf

[22] Montealegre R, Cascio WF. Technology-driven changes in work and employment. Commun ACM. 2017; 60(12): 60-67. https://cacm.acm.org/magazines/2017/12/223043-technology-driven-changes-inwork-and-employment/fulltext

[23] West DM. How employers use technology to surveil employees. Brookings. January 5, 2021. Accessed August 30, 2022. https://www.brookings.edu/blog/techtank/2021/01/05/how-employers-use-technology-to-surveilemployees/

[24] Christenko A, Jankauskaitė V, Paliokaitė A, van den Broek L, Reinhold K, Järvis M. Artificial intelligence for worker management: an overview. Bilbao, Spain: European Agency for Safety and Health at Work. June 2022. Accessed August 30, 2022. https://osha.europa.eu/en/publications/artificialintelligence-worker-management-overview

[25] Bodie MT, Cherry MA, McCormick ML, Tang J. The law and policy of people analytics. Saint Louis University Legal Studies Research Paper No. 2016-6. May 10, 2016. Accessed August 30, 2022. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2769980#

[26] Ravid DM, Tomczak DL, White JC, Behrend TS. EPM 20/20: a review, framework, and research agenda for electronic performance monitoring. J Manage. 2020; 46(1): 100-126. https://doi.org/10.1177/014206319869435

[27] Ajunwa I, Crawford K, Schultz J. Limitless worker surveillance. Calif Law Rev. 2017; 105: 735-776. https://doi.org/10.15779/Z38BRMF94

[28] Kalischko T, Riedl R. Electronic performance monitoring in the digital workplace: conceptualization, review of effects and moderators, and future research opportunities. Front Psychol. 2021; 12: 633031. https://doi.org/10.3389/fpsyg.2021.633031

[29] Newlands G. Algorithmic surveillance in the gig economy: the organization of work through Lefebvrian Conceived Space. Organ Stud. 2021; 42(5): 719-737.  https://doi.org/10.1177/0170840620937900

[30] Brione P. My boss the algorithm: an ethical look at algorithms in the workplace. United Kingdom, Advisory, Conciliation and Arbitration Service (ACAS). March 6, 2020. Accessed August 30, 2022. https://www.acas.org.uk/my-boss-the-algorithm-an-ethical-look-at-algorithms-in-the-workplace

[31] Niehaus S, Hartwig M, Rosen PH, Wischniewski S. An occupational safety and health perspective on human in control and AI. Front Artif Intell. 2022; 5: 868382. https://doi.org/10.3389/frai.2022.868382

[32] Kim PT, Bodie MT. Artificial intelligence and the challenge of workplace discrimination and privacy. ABA J Lab Employ. 2021; 35(2): 289-315.  https://www.americanbar.org/content/dam/aba/publications/aba_journal_of_labor_employment_law/v35/no-2/artificial-intelligence.pdf

[33] Strauß S. From big data to deep learning: a leap towards strong AI or “intelligentia obscura’? Big Data Cogn Comput. 2018; 2(3): 16-35. https://doi.org/10.3390/bdcc2030016

[34] Kantor J, Streitfeld D. Inside Amazon: wrestling with big ideas in a bruising workplace. New York Times. August 15, 2015. https://www.nytimes.com/2015/08/16/technology/inside-amazon-wrestlingbig-ideas-in-a-bruising-workplace.html

[35] Digital Taylorism. The Economist. September 9, 2015.  https://www.economist.com/business/2015/09/10/digital-taylorism

[36] Liu HY, Digital Taylorism in China’s e-commerce industry: a case study of Internet professionals. Economics & Industrial Democracy. 2022; 1-18. https://doi.org/10.1177/0143831X211068887

[37] Bucher EL, Schou PK, Waldkirch M. Pacifying the algorithm—anticipatory compliance in the face of algorithmic management in the gig economy. Organization. 2021; 28(1): 44-67. https://doi.org/10.1177/1350508420961531

[38] Rahman H. The invisible cage: workers’ reactivity to opaque algorithmic evaluations. Admin Sci Quart. 2021; 66(4): 945-988. https://doi.org/10.1177/00018392211010118

[39] Lenaerts K, Waeyaert W, Gillis D, Smits I, Hauben H. Digital platform work and occupational safety and health: overview of regulations, policies, practices, and research. Bilbao, Spain: European Agency for Health and Safety at Work (EU-OSHA), 2022. Accessed August 30, 2022. https://osha.europa.eu/en/publications/digital-platform-work-and-occupational-safety-and-healthoverview-regulation-policies-practices-and-research

[40] Stacey N, Ellwood P, Bradbrook S, et al. Foresight on new and emerging occupational safety and health risks associated with digitalization by 2025. Bilbao, Spain: European Agency for Safety and Health at Work (EU-OSHA), 2018. Accessed August 30, 2022.  https://osha.europa.eu/en/publications/summary-foresight-new-and-emerging-occupational-safety-andhealth-risks-associated-digitalisation-2025

[41] Jarrahi MH, Newlands G, Lee MK, Wolf C, Kinder E, Sutherland W. Algorithmic management in work context. Big Data Soc. 2021; July-December: 1-14. https://doi.org/10.1177/20539517211020332

[42] Kellogg KC, Valentine MA, Christin A. Algorithms at work: the new contested terrain of control. Acad Manag Ann. 2020; 14(1): 366–410. https://doi.org/10.5465/annals.2018.0174

[43] Giermandl LM, Strich F, Christ O, Leicht-Deobald U, Redzepi A. The dark sides of people analytics: reviewing the perils for organizations and employees. Eur J Inform Syst. 2022; 31(3): 410-453. https://doi.org/10.1080/0960085X.2021.1927213

[44] Bottomley E. Data and algorithms in the workplace: an overview of current public policy strategies. University of California, Berkeley Center for Labor Research and Education. Working Paper, Technology and Work Program. November 17, 2020. Accessed August 30, 2022.  https://laborcenter.berkeley.edu/wp-content/uploads/2020/12/Working-Paper-Data-and-Algorithms-inthe-Workplace-An-Overview-of-Current-Public-Policy-Strategies.pdf

[45] O’Neil C. Weapons of Math Destruction: How Big Data Increases Inequality and  Threatens Democracy. New York, NY: Broadway Books; 2017.

[46] Goede H, Kuijpers E, Krone T, et al. Future prospects of occupational exposure modelling of substances in the context of time-resolved data. Ann Work Expos Health. 2021; 65(3): 246-254.  https://doi.org/10.1093/annweh/wxaa102

[47] Ozanich R. Chem/bio wearable sensors: current and future direction. Pure Appl Chem. 2018; 90(10): 1605-1613. https://doi.org/10.1515/pac-2018-0105

[48] Moore PV. OSH and the Future of Work: Benefits and Risks of Artificial Intelligence Tools in Workplaces. Bilbao, Spain: European Agency for Safety and Health at Work (EU-OSHA); 2019. Accessed August 30, 2022. https://osha.europa.eu/en/publications/osh-and-future-work-benefits-and-risks-artificial-intelligencetools-workplaces

[49] Mauno S, Herttalampi M, Minkkinen J, Feldt T, Kubicek B. Is work intensification bad for employees? A review of outcomes for employees over two decades. Work & Stress. Ahead of Print. 2022. https://doi.org/10.1080/02678373.2022.2080778

[50] Acemoglu D. Harms of AI. National Bureau of Economic Research, Working Paper No. 29247, September 2021. Accessed August 30, 2022. https://doi.org/10.3386/w29247

[51] Olhede SC, Wolfe PJ. The growing ubiquity of algorithms in society: implications, impacts and innovations. Phil Trans R Soc A. 2018; 376(2128): 20170364. https://doi.org/10.1098/rsta.2017.0364

[52] Nair L, Stevens J. Algorithms in the workplace—the rise of algorithmic management. Future of Work Hub. July 26, 2021. Accessed August 30, 2022.  https://www.futureofworkhub.info/comment/2021/7/26/algorithms-in-the-workplace-the-rise-ofalgorithmic-management-hcd4f