How-to guide: How to use AI to improve workplace safety (USA)

Introduction

This guide will assist in-house counsel, private practitioners, and compliance professionals to understand and implement AI solutions to improve workplace safety. By identifying safety needs suited for AI improvement, implementing AI solutions, and continuously refining safety systems, businesses can create safer work environments. However, AI also carries a unique set of risks that must be considered when determining what AI will be used for and the policies and procedures governing its use.

This guide covers how to:

Identify workplace safety needs suited for AI improvement
Implement AI to improve safety
Mitigate AI risks
Continuously improve AI safety systems

This guide can be used in conjunction with How-to guide: Risks and liabilities of AI algorithmic bias, Checklist: Steps to mitigate risks associated with AI use in business, and Quick view: Key data privacy and data security terms (USA).

Section 1 – Identify workplace safety needs suited for AI improvement

1.1 Understand workplace safety hazards

Workplace safety hazards vary depending on the industry, organization, and the nature of the work. For example, construction projects on active highways have hazards related to traffic and heavy equipment, while a steel manufacturing facility may have fumes and particulates that contain metals such as lead, chromium, or nickel.

Common types of hazards that employers should be aware of include:

hazards related to the physical environment that may cause accidents such as slips, trips, falls, burns, crushing injuries, or electric shocks;
biological hazards that can cause infections or illnesses in humans, such as bacteria, viruses, fungi, mold, and parasites;
chemical hazards, which may show up in virtually any type of work environment, including manufacturing facilities, laboratories, and even office buildings. These hazards can come in many forms, such as chemical spills, leaks, or exposure to hazardous materials such as asbestos or lead paint;
psychosocial hazards that may relate to workplace stress, violence, harassment, and other factors that can negatively impact mental and emotional well-being. This includes issues like excessive workload, lack of control, bullying, and discrimination; and
mechanical hazard, involving machinery and equipment that can cause injuries through moving parts, sharp edges, or other mechanical actions. This includes things like unguarded machinery, and faulty tools.

1.2 Identify workplace safety hazards through an organizational safety audit

Organizations should conduct an organizational safety audit to identify specific hazards using the following steps.

1.2.1 Review safety violation/workplace accident reports

As an initial step in assessing hazards specific to an organization (or a specific facility within the organization), the organization’s Safety Assessment Team (SAT) should review the history of safety violations and accident reports that are specific to the organization or facility. This will provide a preliminary view of areas that require additional safety focus. An organization may also want to interview employees about potential safety hazards that may have gone unreported.

1.2.2 Conduct a corporate safety audit

In a safety audit, the SAT should be on-site to observe workers as they conduct their daily duties. It is important to observe how the workers utilize the machinery and equipment needed to perform their job functions, and their job site conduct in general.

The audit should identify potential hazards and risks of specific job functions and evaluate the need to implement additional safety procedures or employee training. The SAT should engage with employees to discuss their observations regarding safety behaviors, job functions, and any other specific safety concerns. In conducting the audit, the SAT should be mindful of common safety hazards, such as:

slips and falls;
machinery hazards;
chemical and biological exposure;
temperature stress;
workplace violence; and
non-compliance with personal protective equipment (PPE) rules.

The individuals performing the audit should evaluate company safety manuals, standard operating procedures (SOPs), and training records to ensure the documentation aligns with current legal requirements and industry standards. Individuals should also look for gaps in the company materials, such as outdated procedures or missing PPE requirements.

1.3 Review industry standards and practices

An organization should first identify the regulatory standards that apply to its industry. There are government-issued health and safety standards applicable to most industries. In terms of national standards, the Occupational Safety and Health Administration (OSHA) has issued safety requirements that apply to most general industry employers, and has also issued a Compliance Assistance Quick Start guide. The guide provides a useful roadmap on how to identify the major OSHA general industry requirements and guidance materials that may apply to a specific workplace.

There are additional national safety standards and practices. While not compulsory, these standards provide robust frameworks for organizations in their respective industries to maintain safety, compliance, and business operations. Common safety industry standards include:

ISO 45001: International occupational health and safety standard.
SSIP: Safety Schemes in Procurement, recognized in the construction industry.
ISO 9001: International standard for quality management systems.
ISO 50001: International Standard for energy management.
ISO 39001: International standard for road traffic safety management systems (RTSMS).
ISO 14001: International standard for environmental management systems.
ANSI/ASSP Z10: Occupational Health and Safety Management Systems. This is a U.S. standard that provides a framework for managing occupational safety and health risks.
ISO 27001: Information security, cybersecurity and privacy protection — Information security management systems.
ISO 22301: Security and resilience — Business continuity management systems.
ISO 31000: Risk management guidelines.

Many industry groups and trade associations also have their own industry safety guides.

Section 2 – Implement AI to improve safety

There are a number of ways in which AI can be used to improve workplace safety.

Replacing human analysis with AI technology can reduce human errors that may occur because of inattention or improper analysis. Use of AI can also improve efficiency by dealing with routine tasks and allowing workers to focus on more complex issues.

2.1 Types of AI that can improve workplace safety

Some of the key types of AI that are being used in workplace safety are set out below. Further detail about how each specific type of AI can be used in workplace safety is provided in section 2.2.

2.1.1 Computer vision

Computer vision technology monitors workplace video footage and images, allowing for the detection of a wide variety of objects to identify risks. Such risks may include equipment malfunctions, hazards, or non-compliance with PPE procedures.

2.1.2 Predictive analytics

Predictive analytics allows for the use of AI to determine cause and effect from historical safety data to help organizations avoid safety incidents before they occur. It may also mitigate the damage to employees or property if safety incidents occur. Some AI technologies have the capability to analyze large datasets of past events and may provide mitigation strategies to detect potential risks.

2.1.3 Robotic process automation

Robotic process automation (RPA) can be combined with AI to allow for the automation of repetitive or dangerous tasks in the workplace. The automation of these tasks allows workers who would otherwise be assigned to these duties to focus on more complicated activities. RPA using AI can also offer predictive quality control in manufacturing to identify defective products before the entire manufacturing process is complete, thereby saving time, improving quality, and reducing waste.

2.1.4 Natural language processing

Natural language processing (NLP) technology is now being used in workplace safety analysis to reduce laborious manual processes and streamline safety reporting. NLP rapidly analyzes large volumes of data and provides feedback that SAT and human resources departments can utilize in developing workplace safety practices.

2.1.5 Virtual reality

Virtual reality can be used in employee training programs to present realistic simulations of hazardous tasks, allowing for better workforce training without risk of injury. Virtual reality can also be used to conduct simulated emergency response drills, making employees better prepared to react when emergencies occur.

For more information on different types of AI, see Quick views: Key AI terms (USA) and Overview of AI in business (USA)

2.2 Using AI to improve workplace safety

Some of the key ways in which AI can improve workplace safety are summarized in the table below, alongside the type of AI that can assist.

Benefit	Type of AI that can assist
Real-time threat detection AI can monitor the workplace in real-time and immediately alert managers and workers to potential safety issues. Because workers can quickly respond to the alerts, the technology may prevent accidents and injuries that would otherwise occur.	Computer vision
Early warning systems Monitoring equipment can be used to provide early warnings prior to breakdown, providing for faster responses to accidents or breakdowns.	Computer vision Predictive analytics
Monitor environmental hazards Technology can be utilized to monitor for environmental hazards. For example, filters that monitor air quality can be used to detect the presence of contaminants that would be harmful to the workforce.	Computer vision
Monitor and analyze employees’ behavior and health AI technology provides an opportunity to monitor employee conduct. This type of surveillance allows for assessing employee performance both individually and as a group. This can identify problematic or hazardous behavior and provide an opportunity to correct it. Using insights provided by monitoring employees and machinery can help predict when hazards become present. For example, if the rate of accidents rises dramatically after employees have performed a task for an extended period of time, alerts can be built into the system that call for the employee to be relieved from the task prior to issues occurring.	Computer vision Predictive analytics
Predictive prevention Using AI allows organizations to analyze data from sources that would otherwise be unavailable, such as machinery sensors, in conjunction with historical data, to identify patterns that may predict potential safety risks. This may allow employers to take proactive measures to prevent accidents and injuries.	Predictive analytics
Predictive equipment maintenance AI can be used to determine when certain equipment maintenance should be performed. This enhances the organization’s overall productivity and prevents breakdowns that may cause hazards to the employees who operate it.	Predictive analytics
Manage PPE inventory Managing PPE can be an exhaustive, manual process. With AI technology, the useful life of equipment can be monitored so that the determination of when equipment will need to be replaced can be predicted.	Computer vision Predictive analytics
Perform dangerous tasks One of the most important ways AI will improve workplace safety is by performing dangerous tasks. Certain functions within the workplace are inherently dangerous, such as the handling of hazardous materials or work done in small or confined spaces. AI can be used to perform some of those tasks to eliminate risks to employees from identified hazards.	Robotic process automation
Perform repetitive tasks Certain tasks within a business may be repetitive in nature. Because repetitive tasks tend to result in higher error rates when performed by employees – and since repetitive tasks may give rise to injuries (eg, carpal tunnel syndrome) – to the extent these tasks may be performed by AI, the overall safety of the workplace may be enhanced.	Robotic process automation
Improve efficiency by automating inspections AI allows for the automation of safety inspections and risk assessments, allowing workers to focus on more complex tasks.	Natural language processing
Improve training and education AI allows for personalized training for employees that is specific to their job requirements and learning styles. Workers who are trained and equipped to better understand workplace safety can naturally reduce the risk of accidents and injuries.	Virtual reality
Improve emergency response Technology may be implemented that immediately notifies response teams of an emergency situation. The AI system may provide instructions on where the accident has occurred and the nature of the danger (eg, hazardous materials spills).	Computer vision Virtual reality
Post-incident analysis Having the appropriate technology implemented can assist with analyzing data after an accident or mishap that will help determine the cause of the issue and how to prevent future occurrences. The system may provide a record of when and where the incident occurred, and how quickly the response happened. An AI system may also show where a response could have been more effective or efficient.	Predictive analytics

Section 3 – Mitigate AI risks

There are various risks associated with the use of AI in the workplace. For further information, see How-to-guide: Understanding AI-driven risks (USA). Some of the risks that apply in relation to AI and workplace safety are covered below.

3.1 Worker privacy

Using AI to monitor the workplace gives rise to the potential for invasion of worker privacy. Federal privacy laws, as well as most state privacy laws, give discretion to employers regarding the extent of their employee monitoring programs. Employers should use caution in implementing surveillance policies, and when using technology such as video or other types of monitoring of employees and the workplace.

3.1.1 Federal workplace privacy laws

Federal workplace privacy and employee monitoring regulations stem primarily from the Electronic Communications Privacy Act of 1986 (ECPA, 18 USC Ch 119), which protects the privacy of wire, oral, and electronic communications. In addition, the National Labor Relations Board (NLRB) has issued a Memo on Unlawful Electronic Surveillance and Automated Management Practices. The memo sets out the concern on the part of the NLRB that employee monitoring may be intrusive or abusive even if activities are not explicitly prohibited.

3.1.2 State workplace privacy laws

State privacy laws often place restrictions on employer electronic monitoring. The employer must balance the types of monitoring needed to serve their business needs with the need to respect employees’ reasonable expectations of privacy at work. Employers should also be aware that legal monitoring is not always optimal monitoring: excessive monitoring can demotivate employees and cause a drop in productivity, or even increase the likelihood of employee turnover.

As a matter of caution, employers should consult with specialist counsel regarding applicable state laws on surveillance cameras in workplaces before installing and using such devices.

For further information, see IT and Data Protection USA Practical Resources and US Data protection and privacy (state-by-state).

3.2 Bias detection

AI bias, sometimes referred to as machine learning bias or algorithm bias, occurs when biases are embedded into the original training data or AI algorithm that can then lead to distorted outputs and potentially harmful outcomes. When AI bias goes unaddressed, it can impact an organization’s success. Bias reduces AI’s accuracy, and therefore its potential.

In attempting to detect bias, computer programmers examine the set of outputs that the algorithm produces and review it for anomalous results.

For further information, see How-to guide: Risks and liabilities of AI algorithmic bias.

3.3 Ethical and legal compliance

Using AI in the workplace, especially for purposes such as employee monitoring, is likely to have ethical and legal ramifications. See How-to guide: Understanding AI-driven risks for further information.

3.4 Cybersecurity

Incorporating AI in the workplace raises cybersecurity issues. As with any technology used in connection with data relating to a person, sensitive, private or personal information is potentially at risk. See Checklist: Steps to mitigate risks associated with AI use in business.

Section 4 – Continuously improve AI safety systems

4.1 Regular evaluation and updates

While every attempt may be made, through extensive research and testing, to prevent foreseeable risks before implementing AI, there is a limit to what can be learned in a laboratory. Learning from real-world use is a critical component of creating and releasing increasingly safe AI systems. This should include implementing formal processes to gather feedback from technology support, the SAT, and the users of the AI systems.

4.2 Stay up to date with technological changes

Changes in technology are inevitable, so organizations and the SAT should develop policies and procedures on conducting reviews of the technology being used and what can be done to improve upon it.

4.3 Stay up to date with regulatory changes

AI in business is a new and sometimes controversial subject. Organizations should ensure that regular reviews are conducted with the SAT, technology specialists, and legal and regulatory specialists, both internal and external. For example, employee monitoring law is constantly evolving, and many states are regularly enacting new laws and regulations in this area. The review process should be performed regularly – at least annually and perhaps even quarterly or monthly.

Additonal resources

Protex AI, The Complete Guide to AI Safety in the Workplace
SHRM, Companies Turn to AI to Improve Workplace Safety

Filed under

Topics

Artificial intelligence