AI in EHS: A Powerful Tool, But Not a Legal Shield

By Doug McLean, LinkedIn

Artificial intelligence (AI) is rapidly transforming Environmental, Health, and Safety (EHS) management, offering the potential to revolutionise risk assessment, monitoring, and training. However, businesses must understand that relying on AI doesn't absolve them of their legal responsibilities. In fact, over-reliance on AI can create significant legal vulnerabilities, especially in the event of an incident or fatality.

AI offers compelling advantages in EHS, from analysing vast datasets to predicting potential hazards. It can streamline compliance, personalise training, and provide valuable insights. But these benefits come with risks, particularly when it comes to legal scrutiny.

#EmexSoftware #EHSSoftware #EHS

The Perils of Over-Reliance:

Imagine a scenario where a workplace incident results in a loss of life. The subsequent investigation reveals that the company relied heavily on AI-generated risk assessments, checklists, and method statements. Suddenly, the seemingly advantageous AI system becomes a liability. Here's why:

• "Suitable and Sufficient" Remains the Standard:? Key legislation like the UK's Management of Health and Safety at Work Regulations 1999 requires risk assessments to be "suitable and sufficient."? An AI-generated assessment, however sophisticated, is only as good as the data it's trained on. If the AI misses a crucial hazard due to incomplete or biased data, or if it oversimplifies a complex process, the company could be found in breach of this fundamental requirement. Simply pointing to the AI's output isn't a valid defence. The onus remains on the business to demonstrate they conducted a thorough assessment, not just that an algorithm did. Similar requirements exist across jurisdictions, often stemming from broader frameworks like the EU's Framework Directive 89/391/EEC.
• Explainability is Key: Courts will want to understand the reasoning behind risk assessments. If the company can only say, "The AI didn't flag it," that's a red flag. Judges need to see a clear, understandable process. "Black box" AI models, where the decision-making process is opaque, are particularly problematic in this context. The company must be able to explain why the AI made certain choices, which requires human understanding of the AI's logic and the data it was trained on.
• Unique Circumstances Matter: AI models are trained on historical data. They might struggle with novel situations, "black swan" events, or deviations from standard operating procedures. Human experts, on the other hand, can apply critical thinking and adapt to unforeseen circumstances. Over-reliance on AI creates a vulnerability to situations outside the training data.
• Human Expertise is Essential: If employees become overly dependent on AI, their own risk assessment skills can atrophy. This "deskilling" makes it harder to identify errors in the AI's output or to respond effectively to situations where the AI fails. A court might see this as a failure to maintain a competent workforce. Human oversight of AI is crucial, and that oversight must be informed and skilled.
• Bias and Discrimination: AI models can perpetuate or even amplify existing biases in data, leading to inadequate risk assessment for specific groups of workers or certain tasks. This can expose the company to claims of discrimination, potentially violating equality legislation, in addition to safety violations.
• "Reasonably Practicable" Still Applies: The legal standard often revolves around "reasonably practicable" measures (e.g., as defined in the UK's Health and Safety at Work etc. Act 1974). If the AI suggests a mitigation measure that's impractical, the company can't simply hide behind the AI's recommendation. They have a duty to critically evaluate the AI's output and implement measures that are actually reasonable and practicable.
• Data and Training Matter: If the company used flawed or incomplete data to train the AI, or if they failed to validate its performance, they could be seen as negligent. A court will likely scrutinise the entire AI development and deployment process, not just the final output.? Furthermore, the data used to train the AI can raise significant GDPR concerns.? If the training data includes personal data (e.g., employee health records, near-miss reports containing identifying information), the company must ensure they have a lawful basis for processing this data, have implemented appropriate security measures, and are transparent with employees about how their data is being used.? Failure to comply with GDPR can result in substantial fines.
• Responsibility, Not Liability Shifted: Using AI doesn't absolve a company of its legal responsibility for workplace safety. The company is still ultimately accountable, even if the AI made an error. They can't simply blame the algorithm.

The Takeaway:

AI is a powerful tool for EHS management, but it's not a legal shield. Companies must maintain human oversight, critically evaluate AI outputs, and ensure their risk management processes are robust and compliant, regardless of whether AI is involved. AI should augment human expertise, not replace it. Failure to heed this warning can have devastating legal consequences, especially in the event of a fatality. The key is responsible AI implementation, where technology enhances, but never supplants, human judgment and accountability.? This includes careful consideration of data privacy and compliance with regulations like GDPR.