What does Artificial Intelligence (AI) have to do with workplace safety and health? The National Institute for Occupational Safety and Health (NIOSH) has been at the forefront of workplace safety and robotics, creating the Center for Occupational Robotics Research (CORR) and posting blogs such as A Robot May Not Injure a Worker: Working safely with robots. However, much remains unknown regarding the related field of AI, specifically the application of AI at work. AI is a broad transdisciplinary field with roots in logic, statistics, cognitive psychology, decision theory, neuroscience, linguistics, cybernetics, and computer engineering. Machine learning (ML), a sub-discipline of AI, has led to the application of internet searches, ecommerce sites, goods and services recommender systems, image and speech recognition, sensor technologies, robotic devices, and cognitive decision support systems (see the blog AI and Workers’ Comp).
It is predicted that the impact of AI will be as globally transformative on economic and social structures as steam engines, railroads, electricity, electronics, and the Internet.   AI applications in the workplace of the future raise important issues for occupational safety and health. “Artificial Intelligence: Implications for the Future of Work” was recently published in the American Journal of Industrial Medicine. The commentary reviews the origins of AI, the use of machine learning methods, and emerging AI applications such as sensor technologies, robotic devices, or decision support systems.
Although still in their infancy, as AI-enabled applications are introduced in the workplace, occupational safety and health professionals need to develop a better understanding about AI methods and their potential effects on work and workers. Maximizing the potential occupational safety and health benefits of AI applications, while minimizing any potential challenges, is critical. The following summarizes AI workplace applications outlined in the commentary.
Advanced or “smart” sensors exhibit greater functionality than traditional sensors. Smart sensors can be surgically placed in the body (implanatables); worn on the body or embedded safety clothing (wearables); or attached to a workplace object to monitor different parameters (placeables).   Any device or object with embedded sensors can be connected to the Internet, and to other similar devices, forming an Internet of Things (IoT). A cloud-based IoT platform can collect, integrate, and analyze data from a distributed industrial network of IoT sensors to improve assessment of different workplace safety and health hazards.
AI-enabled sensors can provide both promising benefits for the practice of occupational safety and health and potential challenges. One benefit could be use of continuous data from workplace sensors for early intervention to prevent toxic exposures. Those data would allow practitioners to transition from traditional reliance on slower episodic area or breathing zone sampling. Large data sets produced by a 24/7 sensor network, analyzed by ML-enabled algorithms, have the potential to improve surveillance of safety and health effects from AI, decrease uncertainty in risk assessment and management practices, and stimulate new avenues of occupational safety and health research. Also, AI-enabled virtual reality training can be used to create dynamic, high-fidelity immersive environments to simulate hazardous situations and enhance a worker’s hazard recognition capabilities.
Among the challenges is the privacy dilemma associated with the use of AI-enabled sensor technology to monitor and track all aspects of worker performance. More businesses are managing their workforces using sensor technology, cloud-based human resource systems, and ML-enabled data analytics in an approach called “people analytics.” Proposed best practices for employer-sponsored worker monitoring programs include using only validated sensor technologies; ensuring voluntary worker participation; ceasing data collection outside the workplace; disclosing all data uses; and ensuring secure data storage.
Recently, there has been a shift from workplace robotic devices that do routine functions—automated robots—to the more advanced robots that are able to interact with people and their environment—autonomous robots. These newer AI-enabled robotic devices are called collaborative robots or “cobots”. The presence of a cobot and a human worker in the same work area raises a number of safety issues, primarily collision control. In 2016, the International Organization for Standardization (ISO) provided safety requirements to promote safe human-cobot collaboration. For industrial cobots equipped with AI-enabled sensors, the ISO recommended: (1) safety-related monitored stopping controls; (2) human hand guiding of the cobot; (3) speed and separation monitoring controls; and (4) power and force limitations.
AI methods are also enabling one robotic device to learn from the experience other robotic devices, since the sensors in robotic devices can be connected to the cloud. The learning experience of one AI-enabled robotic device can be uploaded to all other connected robots by means of “cloud robotics.”
Decision Support Systems:
Firms that collect and store large amounts of data, who have robust computational capabilities, and in‐house computer engineering expertise, are introducing AI to support financial, operational, and organizational risk decision‐making. AI applications can be used to mine knowledge from data for decision-making applications by using a decision support system (DSS)—a multi-purpose informational AI-enabled tool—that aids humans in finding information or making decisions. For example, AI-enabled DSSs have shown promise in medicine and can be used to detect lung cancer in x-ray screening.
DSSs may have a role in improving occupational risk assessment and risk management strategies. Can AI-enabled DSSs prevent catastrophic events such as chemical plant explosions by recognizing root causes of such events earlier? Can AI-enabled DSSs aid in determining the optimal placement of fire fighters during disasters like wildland fires to prevent them from being overtaken by the fire? Can AI-enabled DSSs aid in making risk control decisions under conditions of uncertainty? Can AI-enabled systems take control from a human to prevent a human action that will lead to severe injury or a fatality?
These and other questions about AI and the future of work deserve the attention of the occupational safety and health community. Concerns about ML-enabled DDSs, including algorithm transparency and algorithm bias, have arisen as they are introduced across industry sectors. The lack of methodological transparency inherent in ML methods (“black box”) can impair user trust in the outputs produced by a DSS.
Another implication of AI on work is automation. Several estimates have been published about the extent to which job tasks could be automated across industry sectors. Studies by Oxford University and by the McKinsey Global Institute indicate that about half of all job tasks in the U.S. economy could be automated with current AI-enabled technologies. However, not all studies agree that AI will be that much of a job eliminator. Some studies point to several economic, legal, or societal factors that could restrain a firm from adopting job-displacing AI technologies. Fears of technological disruption by AI may be exaggerated, as technology adoption is often slow which provides time for new task and job creation to offset job loss from automation. 
Human-machine interactions must also be addressed when considering AI in the workplace. Negative consequences can occur when system controls are not fully understandable to humans, or fully responsive in practice as they were in design. Managing risk as AI-enabled technologies are introduced to the workplace should start with a systems safety approach that focuses on system operation and controls to ensure the reliability and safety of AI technologies enabling autonomous systems. The introduction of AI-enabled technologies in self-driving vehicles, at a nuclear power plant, or in the avionics systems of a jet airliner,  raises issues of how to manage the uncertainties associated with human-machine interactions with AI-enabled systems.
Occupational safety and health practitioners, researchers, employers and workers must consider the ramifications of AI-enabled applications in the workplace. Before AI-enabled devices or systems are introduced into a workplace, a thorough preplacement safety and health review of their benefits and risks should be performed. We welcome your thoughts in the comment section below (the blog post) as we proactively address the potential advantages and challenges of this technology.
 International Labour Organization. Safety and health at the Heart of the Future of Work: Building on 100 Years of Experience. Geneva, Switzerland; 2019. https://www.ilo.org/safework/events/safeday/WCMS_686645/lang–en/index.htm. Accessed June 21, 2019.
 Manyika J, Bughin J. The Promise and Challenge of the age of artificial intelligence. The McKinsey Global Institute, October 2018. https://www.mckinsey.com/featured-insights/artificial-intelligence/thepromise-and-challenge-of-the-age-of-artificial-intelligence. Accessed June 21, 2019
 West DM, Allen JR. How Artificial Intelligence is Transforming the World. Washington, DC: Brookings. April 24, 2018. https://www.brookings.edu/research/how-artificial-intelligence-is-transforming-the-world/. Accessed June 21, 2019.
 Nag A, Mukhopadbyay SC. Wearable flexible sensors: a review. IEEE Sens J. 2017;17(13):3949-3960. https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7931559&tag=1. Accessed June 21,2019.
 Bray H. MIT Labor Shows Off Smart Treads That Can Send Messages, Change Color. Boston Globe. June 18, 2017. https://www.bostonglobe.com/business/2017/06/19/mit-lab-shows-off-smartthreads-that-can-send-messages-change-color/Kf8S15GtAWuZEc7TXprVkM/story.html. Accessed June 21, 2019.
 Yang G-Z. Implantable Sensors and Systems: From Theory to Practice. New York, NY: Springer; 2018
 Metz R. The company embeds microchips in its employees, and they love it. MIT Technol Rev. August 17, 2018. https://www.technologyreview.com/s/611884/this-company-embeds-icrochips-inits-employees-and-they-love-it/. Accessed June 21, 2019.
 Chui M, Loeffler M, Roberts R. The Internet of Things. McKinsey Quarterly. March 2010. https://www.mckinsey.com/industries/high-tech/our-insights/the-internet-of-things. Accessed June 21, 2019.
 Bellanca JL, Orr TJ, Helfrich WJ, Macdonald B, Navoyski J, Demich B. Developing a virtual reality environment for mining research. Min Metal Exploration. 2019;1-10. https://doi.org/10.1007/s42461-018-0046-2. Accessed June 21, 2019.
 Booth R. UK businesses using artificial intelligence to monitor staff activity. The Guardian. April 7, 2019. https://www.theguardian.com/technology/2019/apr/07/uk-businesses-using-artifical-intelligenceto-monitor-staff-activity. Accessed June 21, 2019.
 Collins L, Fineman DR, Tsuchida A. People Analytics: Recalculating the Route. 2017 Human Capital Trends. Deloitte Insights. February 28, 2017. https://www2.deloitte.com/insights/us/en/focus/human-capital-trends/2017/people-analytics-in-hr.html. Accessed June 21, 2019.
 Marchant GE. What are best practices for ethical use of nanosensors for worker surveillance? AMA J Ethics. 2019;21(4):E356-362. https://journalofethics.ama-assn.org/sites/journalofethics.amaassn. org/files/2019-03/stas1-1904_0.pdf. Accessed June 21, 2019.
 Vysocky A, Novak P. Human—robot collaboration. MM Science Journal. 2016:June:903-906. https://doi.org/10.17973/MMSJ.2016_06_201611. Accessed June 21, 2019.
 International Organization for Standardization. ISO/TS 15066:2016. Robots and Robotic Devices—Collaborative Robots. https://www.iso.org/standard/62996.html. Accessed June 21, 2019.
 Kohoe B, Patil S, Abbeel P, Goldberg K. A survey of research on cloud robotics and automation. IEEE T Autom Sci Eng. 2015;12(2):398-409. https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7006734. June 21, 2019.
 Bonini CP. Simulation of Information and Decision System in the Firm. Englewood Cliffs, NJ: Prentice Hall; 1963.
 Lahav O, Mastronarde N, van der Schaar M. What is interpretable? Using machine learning to design interpretable decision-support systems. June 11, 2019. https://arxiv.org/abs/1811.10799. Accessed June 21, 2019.
 Frey C, Osborne M. The future of employment: how susceptible are jobs to computerisation. Technological Forecasting and Social Change. 2017;114:254-280. https://doi.org/10.1016/j.techfore.2016.08.019. Accessed June 21, 2019.
 The McKinsey Global Institute. A Future That Works: Automation, Employment and Productivity. January 2017. https://www.mckinsey.com/~/media/McKinsey/Featured%20Insights/Digital%20Disruption/Harnessing%20automation%20for%20a%20future%20that%20works/MGI-A-future-that-works_Full-report.ashx. Accessed June 21, 2019.
 Arntz M, Greggory T, Zierahn U. The Risk of Automation for Jobs in OECD Countries: A Comparative Analysis. Organisation of Economic Cooperation and Development, Social, Employment and Migration Working Papers No. 189; 2016. https://dx.doi.org/10.1787/5jlz9h56dvq7-en. Accessed June 21, 2019.
 Atkinson RD, Wu J. False Alarmism: Technological Disruption and the U.S. Labor Market, 1850-2015. Information Technology & Innovation Foundation, May 8, 2017. http://www2.itif.org/2017-falsealarmism-technological-disruption.pdf. Accessed June 21, 2019.
 Cass O. The Once and Future Worker. A Vision for the Renewal of Work in America. New York, NY: Encounter Books; 2018
 Acemoglu D, Restrepo P. The race between man and machine: implications of technology for growth, factor shares, and employment. Am Econ Rev. 2018;108(6):1488-1542. https://pubs.aeaweb.org/doi/pdfplus/10.1257/aer.20160696. Accessed June 21, 2019.
 Autor D. Why are there still so many jobs? The history and future of workplace automation. J Econo Perspect. 2015;29(3):3-30. https://pubs.aeaweb.org/doi/pdfplus/10.1257/jep.29.3.3. Accessed June 21, 2019.
 Leveson NG. Engineering a Safer World. Systems Thinking Applied to Safety. Cambridge, MA:MIT Press; 2011.
 Rouhani BD, Samragh M, Javidi T, Koushanfar F. Safe machine learning and defeating adversarial attacks. IEEE Secur Priv. 2019;17(2):31-28. https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8677311. Accessed June 21, 2019.
 Weise E, Marsh A. Google self-driving van involved in crash in Arizona, driver injured (Update). USA Today, May 5, 2019. https://phys.org/news/2018-05-waymo-self-driving-car-collisionarizona. html. Accessed June 21, 2019.
 Walker JS. Three Mile Island: A Nuclear Crisis in Historical Perspective. Berkeley, CA: University of California Press; 2004.
 Clark N. Report on ’09 Air France Crash Cites Conflicting Data in Cockpit. New York Times, July 5, 2012. https://www.nytimes.com/2012/07/06/world/europe/air-france-flight-447-report-citesconfusion-in-cockpit.html. Accessed June 21, 2019.
 German K. FAA has no timetable for when Max 8 will fly again. CNET, June 10, 2019. https://www.cnet.com/news/boeings-737-max-8-all-about-the-aircraft-flight-ban-and-investigations/.