The ability to predict safety-related events can be a useful tool in the campaign for companies preventing incidents and protecting the safety of their employees and the environment. Receiving a notification prior to a pending event and having the power to act toward preventing it is an ideal scenario for safety management. However, though this seeming near omnipotence is an appealing notion, there are several drawbacks to “the power of prediction” that companies should keep in mind. This article will walk through 6 of the most common drawbacks of attempting to predict safety-related events.
- False Positives: One of the most significant drawbacks of predicting safety-related events is the potential for false positives. False positives occur when a system or process identifies a potential safety issue or event that does not actually exist. False positives can lead to unnecessary work, wasted resources, and decreased confidence in the accuracy of predictive models.
- False Negatives: In addition to false positives, predicting safety-related events can also result in false negatives. False negatives occur when a system or process fails to identify a potential safety event that does occur. For example, a predictive model might fail to identify an event due to incomplete or inaccurate data. False negatives can be even more dangerous than false positives, as they can lead to preventable incidents.
- Over-reliance on Technology: Predictive models and other technologies can be powerful tools for predicting safety-related events. However, there is a risk that companies may become over-reliant on technology and fail to use other important safety tools, such as simply talking to employees about their concerns. Over-reliance on technology can lead to a false sense of security and may result in complacency regarding safety risks.
- Lack of Context: Predictive models and other technologies may be effective at identifying potential safety hazards based on historical data, but they generally lack application of the context for the situations utilized in their training. For example, a predictive model may identify a potential hazard based on previous incidents, but that hazard may not be relevant in the current situation due to changes in processes, equipment, or personnel. Companies must ensure that predictive models are context-specific and have input for the current conditions and factors that may be contributing to or influencing safety.
- Potential Ethical Concerns: Predictive models can raise ethical concerns related to privacy and data security. Predictive models rely on data, and companies must ensure that they are collecting and using data responsibly and ethically. Related to this, predictive models may lead to unintended consequences, such as biased decision-making or discrimination. When bringing work to a (temporary or prolonged) halt is the most effective safety measure to prevent an incident, it inherently means setbacks. To diminish the impact of these setbacks, utilizing an understanding and communicative manager/supervisor is still the best means of instigating a pause in operations.
- Black-box Explanations: In a black-box system for predicting safety-related events, complex algorithms and machine learning techniques are used to make predictions based on historical data. One of the main benefits of black box systems is that they can make predictions based on large amounts of data and algorithms that would be difficult for humans to process. However, the downside of black box systems is that they can be difficult to interpret or explain, which is problematic in the context of safety-related decisions where stakeholders need to understand how predictions were made and what factors were considered to mobilize prevention efforts effectively. Will a management team be convinced to act on predictions without the ability to tap into the reasoning used to flag the situation as dangerous?
The desire to be able to predict the outcome of safety-related events is very understandable. However, there are significant drawbacks to predictive technology that companies must consider before adopting this type of system. To mitigate these risks, companies must take a comprehensive approach to safety that incorporates multiple tools and approaches, including relying on the expertise and input of humans already practicing safety within the organization.
Cary comes to the SafetyStratus team as the Vice President of Operations with almost 30 years of experience in several different industries. He began his career in the United States Navy’s nuclear power program. From there he transitioned into the public sector as an Environmental, Health & Safety Manager in the utility industry. After almost thirteen years, he transitioned into the construction sector as a Safety Director at a large, international construction company. Most recently he held the position of Manager of Professional Services at a safety software company, overseeing the customer success, implementation, and process consulting aspects of the services team.
At SafetyStratus, he is focused on helping achieve the company’s vision of “Saving lives and the environment by successfully integrating knowledgeable people, sustainable processes, and unparalleled technology”.