Reliability of human point of work risk assessment vs artificial intelligence risk assessment for construction health and safety

The significance of construction workers being involved in Point of Work Risk Assessment (POWRA) cannot be overemphasised. It is a quick, on-the-spot assessment performed by workers before starting a task, focusing on the current conditions and any hazards identified. There are anecdotal claims that AI is capable of providing robust risk assessment similar to or better than those developed by construction workers if adequate prompts are provided to the AI tool. However, there is no credible evidence to support these suggestions. The aim of this paper is to explore the reliability of AI generated construction project risk assessment vs human developed POWRA. This pilot study compared independent risk assessments conducted by 25 Postgraduate Construction Project Management students against AI Prompt (ChatGPT). It conducted a human POWRA vs AI risk assessment for the task with the following information – project description, tools and equipment, processes and activities, workers involved, and the work environment. The findings indicated that the students identified 21 hazard types whilst the AI identified 20 hazards that were rated with the required control measures, and then rating the risk after control measures were applied. The AI identified nine hazards that were not included in those identified by the students whilst 10 hazards identified by the students were missed by AI. The AI provided control measures and 15 of those were missed by the students whilst the control measures provided by the students had 16 controls missed by the AI tool. The implication is that humans should not become too overly reliant on information that is provided by technology as dependable without any sense-checking. Without adequate checks and balances and fail safes, organisations could increasingly be dehumanizing construction work as we lean more into technology. Technology is already advancing construction OSH and AI is a work in progress that should still require human oversight at this time.