Enhancing Workplace Safety with AI

Picture this: you have a birds-eye view of every single risk, incident and control within your organisation; you can peer into hidden corners and identify unseen hazards; you can predict possible scenarios and simulate new ways to safeguard people’s wellbeing… all from your desktop. This capability is mind-blowing, but it’s not a distant dream. It is, in fact, the reality unfolding before our eyes.

Artificial Intelligence (AI) is revolutionising our approach to safety, unleashing exciting (and sometimes alarming) opportunities. Innovative safety leaders are already analysing vast amounts of workplace data to uncover hidden incident patterns, predict risks, test controls and provide new insights to protect their people’s wellbeing. Read on to learn more about enhancing workplace safety with AI.

Aggregating diverse data

While traditional safety metrics continue to provide crucial insights into workplace health and safety, Artificial Intelligence raises the bar. The sheer variety and amount of the information that can be processed by AI is a game-changer.

Speaking at a recent HSE leaders forum hosted by The Safe Step, Downer Group’s General Manager of Zero Harm Risk, Mathew Hancock explained that AI has helped him to develop complex dashboards drawing upon hundreds of thousands of different data points, from all around the business. This detailed overview of Downers’ people and operations is beginning to drive deeper insight and learning in managing health and safety within a highly diverse organisation where a commitment to Zero Harm is a core principle. “Giving people visibility is the first step. If you can’t see it, you can’t fix it,” explains Hancock.

AI-driven HSE management systems may in future aid data-driven decision making by being able to overlay complex information received from incident reports, legislation, wearable devices, Internet of Things (IoT) sensors, weather patterns, facilities and asset management systems, HR systems and employee feedback. The intersection of all this information gives safety professionals a broad view of the business and sheds light on unseen trends, or root causes of accidents.

At Downer, Mathew Hancock uses machine learning to match control function structures with risk type taxonomies. “For ‘people at heights’ incidents, we can see what type of actions are implemented after these incidents. In this way, we’ve been able to improve education and refine the actions we take both before and after an event,” he says.

Furthermore, if monitoring occurs in real time, AI tools may be able to quickly detect anomalies, unsafe conditions, or deviations from safety protocols, allowing for immediate intervention.

Insights derived from unstructured information

AI allows you to uncover gems of information hidden not only within numerical data, but also within verbal, visual and written documentation too.

Artificial Intelligence applications are becoming much more effective a processing vast datasets of complex, unstructured content. As a result, resources such as CCTV footage, written incident reports, safety observations, or near-miss records may become much more valuable for safety professionals.

It’s currently difficult to analyse human language at scale because it’s complicated by context, dialect, levels of education, slang and cultural expressions. One challenge is to establish AI models that can disambiguate complex language to improve the identification of specific concepts to unlock insight embedded in narrative descriptions. Generative AI and advanced natural language processing techniques are beginning to be able to do this with a decent degree of reliability.

To make sense of natural language insights, it is important to overlay knowledge structures. WHS professionals are exploring the creation of new ways to categorise and organise incidents by developing taxonomies and hierarchical structures.

While Mathew Hancock concedes that there’s a long way to go, he believes Downer is making good progress. “Things aren’t mapped out perfectly yet, but we can now review hundreds of thousands of incidents by seasonal effects, time of month and day and risk taxonomy,” he says. The next steps are applying reasoning, but Hancock estimates this functionality is still some years away.

Simplifying safety

Natural language interactions are still a long way from being perfect, future AI tools may have advanced natural language processing capabilities, enabling seamless and intuitive interactions with health and safety professionals.

Those aiming to simplify safety should in time be able to strip-out bureaucracy - rather than dealing with lengthy manuals and checklists safety professionals can ask questions, receive real-time guidance, and access relevant information to make informed decisions swiftly.

Of course, the accuracy of this information is very much reliant on the quality of information the system is using. Mathew Hancock believes safety leaders need to prepare for more advanced AI by bringing structure and order to their systems and data using currently available technology. “Think about the foundational future advanced AI platforms will require. What do you need to set up now, so that you’ll have the right data and knowledge structures to draw upon in the future?” he suggests.

Once appropriate structures and models are put in place, AI systems should be able to be developed that can prompt and guide employees in generating more accurate and detailed safety reports – reducing the burden of data quality assurance and fostering a culture of continuous improvement.

Producing and revising content

Generative AI has various uses, notably re-writing, summarising, extracting details and generating example content. Generative AI applications that re-write content or extract details can be developed to deliver quite robust and consistent results.

More open-ended questions can be helpful, but are more likely to get things wrong - for example, you could ask ChatGPT to ‘write a paragraph about fall arrest system requirements in the style of a corporate report’, or request it to ‘summarise this policy in 100 words or less’, and it might work, but you will most certainly have to check your homework before submitting it as a number of highly publicised cases have shown (e.g. the lawyer who was caught using ChatGPT to prepare his case without fact checking what it said).

Downer Group has used AI to reduce time-consuming and technical tasks such as reviewing regulations, classifying actions and allocating relevant tasks. “We have used natural language technology to analyse documentation and procedures, aggregate and classify narrative data (e.g. incident reports), have applied Generative AI to produce content for our risk and control libraries and maps of concepts relating to our critical risks” Mathew Hancock says.

Be a cautious and responsible user of AI

Artificial Intelligence unleashes exciting possibilities, but Hancock recommends using caution when adopting Generative AI.

“Procedures written by ChatGPT can sound convincing, but the detail within might be completely wrong – especially when it comes to regulatory requirements at this point in time,” warns Mathew Hancock. “We also have to be careful about cyber security, privacy and data sovereignty rules / contractual terms,” he adds.

Many WHS professionals are being warned to consider privacy and ethical concerns and to take a responsible approach to safety innovation. Data sovereignty and security are paramount, especially with tools available on the open market and AI platforms where data is processed overseas. Be aware of and mitigate the risks.

Always cross-verify information with subject matter and legal experts, especially when it relates to complex issues or high-risk situations. You are the expert and must review and revise all information. Artificial Intelligence won’t replace human relationships, corporate knowledge, reasoning and years of expertise and training. Rather, it’s a tool to be used with care.

Hancock cautions against getting swept up in the hype. There are claims about what large language models / generative AI can do that may not turn out to be true – reasoning being a key example. “Large language models might look like they are reasoning, but they are actually fabricating language that looks like reasoning – this is a very important distinction. The models stack words together based on some pretty crazy statistical analysis at massive scales. Give it a lot of examples of what language expressing reasoning looks like, and it will be able to produce language that look like it is reasoning – but it doesn’t mean the models can reason reliably, at scale and under complexity. We need to learn what generative AI can be relied upon to do, and what it can’t.”

Getting started with custom AI solutions

Identify pain points. Think about specific areas where AI can be beneficial, such as automating tasks and analysing data to improve workplace safety. Gather relevant data and ensure its quality and accessibility for effective AI implementation.

Play and experiment. Familiarise yourself with AI techniques and tools, such as machine learning algorithms and natural language processing, to determine which ones align with your objectives. Collaboration with HSE professionals and IT experts is crucial to gather domain expertise and ensure a comprehensive approach.

Finance will also be critical – do you have the mandate to invest in this area? If not, how can you influence the business? With commercial applications of AI in their infancy, partnering with a university or a software provider might be another avenue for funding innovations.

Begin with small-scale projects, creating prototypes or proof-of-concepts to showcase AI's potential in addressing identified challenges. Continuously iterate and refine your approach, incorporating feedback and insights gained, to progressively build a custom HSE management system that leverages the benefits of AI technology.

Stay current. This is a rapidly evolving field. Things you can’t do now, may soon be possible (and more affordable).

“Don’t put your head in the sand. Understand where the future of your business, and the HSE profession, is going. It might be scary, and a bit overwhelming, but it’s very exciting too,” said The Safe Step’s Associate Director Stephen Colidcutt.

How we can help

WHS teams are only beginning to imagine the possibilities created by Artificial Intelligence (AI). Many will lack the required technical expertise needed to lead the way in this space.

Those with the necessary technical understanding are in strong demand and can be difficult to find and attract to your organisation.

Working with a specialist HSE recruiter will help you to identify and attract the right talent to your organisation. We also work with senior consultants who specialise in the technology space who can run projects and develop the capabilities of your HSE team.

With the right people by your side, you will have the ability to harness AI's power to process and organise complex data, organisations can uncover hidden insights, implement targeted interventions, and drive continuous improvement in their workplace safety practices.