Artificial intelligence still needs accountable intelligence

Artificial intelligence is already capable of making fire-safety consultancy more efficient. It can organise survey notes, interrogate large document sets, compare revisions, identify apparent inconsistencies and help turn technical information into clear client communication. Used carefully, it releases the consultant from repetitive processing and creates more time for investigation, analysis and engagement with the design team.

That is a genuine opportunity. It is not a transfer of professional responsibility.

AI can produce an answer that is fluent, structured and technically plausible while relying on an outdated standard, inventing a requirement or overlooking a critical qualification. In fire safety, these are not simply editorial defects. An apparently authoritative error can influence the design of an escape route, the specification of fire-resisting construction or the priority attached to a risk.

Verification is therefore essential, but technical verification is only part of the challenge.

Some questions have relatively definite answers. A provision may comply with a stated benchmark, or it may not. The source, edition and application can be checked. Other decisions occupy the more difficult territory of professional judgement: whether a departure creates a material risk, whether an alternative measure provides adequate protection, and whether further expenditure or disruption would be proportionate to the benefit achieved.

AI can assist that assessment. It can test assumptions, present options and identify factors that deserve consideration. But who decides what is proportionate?

The answer cannot be “the system”. Proportionality depends upon context: the severity and likelihood of harm, uncertainty in the evidence, the vulnerability of those exposed, the effectiveness and practicality of possible measures, and the consequences of getting the decision wrong. These matters require competent human judgement, openly reasoned and capable of being challenged.

Nor can the consultant’s conclusion silently become the decision of an algorithm. Under fire-safety legislation, the Responsible Person must arrange a suitable and sufficient assessment and implement appropriate measures. Government guidance also makes clear that both dutyholders and the competent professionals appointed to assist them may carry responsibility for inadequate work. Fire-safety duties do not disappear when part of the process is delegated.

The ethical use of AI therefore requires more than checking its arithmetic or citations. It requires control over where it is used, transparency about its influence, protection of confidential information, and a clear record of who reviewed and accepted the resulting advice. The UK’s AI framework similarly emphasises safety, transparency, accountability and governance, not merely technical performance. Those principles are particularly relevant where AI informs safety-critical work.

The Engineering Council’s updated ethical principles ask professionals to consider the risks and implications of AI and to communicate uncertainties and trade-offs proportionately. That is an appropriate test for fire-safety practice.

AI should make competent consultants more effective. It should not make professional judgement anonymous. The essential question is not whether AI contributed to the advice, but whether an identifiable, competent person understood the evidence, tested the output and remained willing to take responsibility for the conclusion.

Pyrology Insight provides commentary on contemporary fire safety topics. It is intended to stimulate professional discussion and should not be relied upon as project-specific advice.