When artificial intelligence (AI) began to prove its worth as a solid and reliable tool for undertaking repetitive workplace processes such as scanning images, text, and other data formats, it emerged as a powerful tool for scaling productivity and spotting anomalies in large data sets. 

In the medical field, AI is commonly used to help diagnose diseases. A 2019 study led by notable researchers including Professor Alastair K Denniston, Dr Eric Topol and Dr Pearse Keane showed that when compared to human doctors, AI-powered models are just as effective in detecting diseases from medical images. The technology is increasingly being used for early detection and other preventative measures against diseases such as aneurysms, cancer, multiple sclerosis and Parkinson’s, to name a few. As of October 2022, the United States Food and Drug Administration (FDA) has already approved more than 500 AI and machine learning-enabled medical devices.

But can we translate the successful use of AI in medical imaging – which benefits from an abundance in data and clearly standardised processes – and apply it into a more dynamic environment found within the health and safety sector? There are several ways that AI can positively impact health and safety practices and help address some of the biggest hazards to workers within manufacturing, construction, industrial and logistics:

Actionable detection of slips, trips and falls

According to the Royal Society for the Prevention of Accidents (RoSPA), slips, trips and falls are the most commonly reported injury in the workplace, on average causing 40 per cent of all reported major injuries and costing employers more than £500 million per year. However, many more go unreported.

Advanced sensors such as accelerometers can detect if a person has tripped or fallen. However, they require the person to be wearing the sensor on their body, making them costly and vulnerable to user errors. But when combined with computer-vision software, advanced behaviour AI – an AI system that directly interacts with humans to understand human behaviour for further decision making – gives cameras the power of sight with added context; an important tool to have when needing to detect slips, trips, and falls as soon, or even before, they happen. 

By using behaviour AI, employers will be able to see when and where workers have fallen, decipher the difference between a slip, trip or a deliberate bend forward, and understand the context of the situation – for instance, if there was crowding at the time, uneven material on the ground or if it was due to being distracted.

Computer vision-based behaviour AI can also be used for preventative and predictive measures. In addition to spotting incidents quicker and being able to respond to it with necessary actions (faster triage means speedier escalation or de-escalation), behaviour AI can extract useful insights on what is causing the falls and enable companies to make changes to their environments to prevent future incidents. Many construction and industrial settings already have CCTV or infrastructure cameras in place, so the behaviour AI can tap into existing video feeds to process the data, meaning setup costs would be kept relatively low.

Flagging insufficient PPE gear

The International Safety Equipment Association defines personal protective equipment (PPE) as items “worn to minimise exposure to hazards that cause serious workplace injuries and illnesses”. According to the United States Department of Labor’s Occupational Health and Safety Administration (OSHA), these include “gloves, safety glasses and shoes, earplugs or muffs, hard hats, respirators, or coveralls, vests and full-body suits”. In addition to ensuring workers wear PPE every time they arrive on site, employees are also responsible for inspecting and replacing them regularly to meet stringent compliance requirements, especially when the equipment becomes damaged, dirty or ineffective over time. 

Thanks to its ability to spot anomalies from large amounts of data, the use of AI to flag employees wearing insufficient or incorrect PPE could help prevent them entering a hazardous environment without the proper protection. If exposed beforehand to images and video showing appropriate PPE, computer-vision software would be able to detect and match items, while more advanced AI can pick up on more detailed information, such as ill-fitting PPE. 

Having this technology could also help address secondary safety challenges in the sector, such as the difficulties women face with poorly designed PPE, by measuring statistics and capturing data on the number of times employees enter a construction site with improperly fitted PPE. This will give employers data-based insights on which to enact change – in this case, to explore the root issue of why there are a high number of employees with ill-fitting PPE – without requiring them to report or submit complaints. 

Preventing “struck-by” injuries

One of the stats found on the OSHA website says that “approximately 75% of struck-by fatalities involve heavy equipment such as trucks or cranes, with one in four ‘struck by vehicle’ deaths involving construction workers, more than any other occupation”. There are a several ways that AI can help prevent struck-by injuries: 

• Companies using semi- or fully automated machinery in their construction, warehouse or manufacturing site can install AI within the machinery to improve its safety around human workers. Some AI technologies combine sensor hardware with pre-mapped pathways to help guide machinery automations; others use computer vision through connected dashcams or other cameras placed on the machinery to respond to visual stimuli.
• AI can help map optimal route options for humans in the workplace. This would be more of a historical analysis of current work sites and the directional pathways of every moving object in the space and would be useful in unchanging environments such as warehouses. For ever-changing environments, like a construction site, it might be better to use something more flexible like behaviour AI, which pre-maps zones of interest and tracks humans’ physical behaviour.
• Behaviour AI can help detect and predict risky interactions between people and moving objects as they happen, using cameras on moving vehicles. Dashcams with built-in behaviour AI models can predict a person’s crossing intent in front of a vehicle so that the vehicle’s driver can be alerted to a potential crash. Behaviour AI would be the most effective way of preventing struck-by injuries due to its use of camera footage as a sense of sight, and the need for swift action in the aftermath of an incident. 

Enabling and automating better incident reporting

Although not a direct way of keeping people physically safe, AI’s ability to complete mundane tasks at scale means it can be a reliable tool for automating incident reporting. According to the Survey of Occupational Injuries and Illnesses (SOII), almost 69% of incidents go unreported for various reasons, and without this feedback from employees on what happened, when, where, how, and why, it is more difficult for employers to make larger health and safety issues to better protect their employees. If using Behaviour AI, incident reporting can be made even easier, as there would be video footage to go along with the report, or it could be automatically flagged to the employer.

Conclusion

AI can be used on its own as a technology or in tandem with edge computing, sensors, cameras or on the cloud. Its flexibility means it can support a larger system or be the main algorithm in a process. It depends on what the end goal or objective is, what current systems companies have in place, and the specific outcomes and metrics they want to achieve.

As AI and machine-learning-enabled technologies continue to evolve, the entire health and safety ecosystem will transform in correlation as more parts of the supply chain will be impacted. Its applications include being a tool to scale productivity using fewer resources, a preventative and protective enhancement for keeping workers safe on site, and even aiding post-incident reporting by helping workers and employers complete long administrative reports quickly and efficiently.

The desire, intent and demand is getting there: according to Peak.AI’s 2022 Decision Intelligence Maturity Report, “manufacturing ranked third highest in commercial AI maturity, with a score of 53, significantly higher than associated industries, construction and architecture (49) and transport, warehousing and logistics (40)”. 

But how ready are these industries to adopt and implement AI technology?

Regardless of when health and safety leaders are ready to implement AI to keep their workers safe, like with any adoption of new technology, they should always keep in mind the key objectives they need to achieve, including specific and tangible metrics. They should also have clarity on how easy (or difficult) it is to integrate within their current systems or if an entire new infrastructure is required. However, the most important point – and one that business executives need to be aware of the most – is how usable, transparent or explainable and ethical the AI model is designed to be. If the algorithm ends up perpetuating unfair bias, or incorrectly triggers inappropriate automations, then it will be costly to backtrack and fix. It’s better to have this clarity and understanding of how the AI system is meant to work at the beginning to ensure complete success in the long run.