Safer than a Human = Better than a Human?

There have been very many articles on these pages and other making and breaking the case for automation systems being safer? than a human operated system.? This metric is also being used more extensively as a headline for selling automation systems to replace human operated systems.

This short article looks at the case and gives an author’s interpretation on the case for and against this argument.

When speaking of risk in the piece the base definition is the severity of impact and frequency of occurrence for a given hazard event.? This is key to understanding the arguments here and should be born in mind.

The Case for Safer

When considering manually operated systems there are two elements to be considered.? The risks to the operator(s) within the system (internal risks) and the risks to this outside the system (external risks).? This equally applied to risks to the environment, property, etc.? When the automated system is looked at the internal risks to people are dramatically reduced.? This is down to two factors.? The first is where the human is involved.? In many cases the most hazardous elements are automated.? This almost eliminates the internal risks to people at this point.? Where people are used within the system the exposure time is greatly reduced in these cases.?

Addressing external risks, the argument is made the human operators are variable in ability and any human is variable from hour to hour depending on factors such as fatigue, mood, distractions etc.? by removing of dramatically reducing the role of people within the system these risks are reduced dramatically,? therefore the system is more consistent, predictable, less hazardous to these inside and outside.? The reduction in risk compared to a manual system is great enough to claim that automation is safer.

The Case Against

On the counter a concept of risk transposition is utilised.? Risk transposition is in essence the effect of automating a system on the risk profile.? Whist removing an operator reduces the internal risks these risks are transposed and magnified into external risks? as the automation system lacks the fine judgement of an experience human operator to? account for the stack up of small variables which allow the correct risk-based decision to be made.? This lack of judgement may be down to the lack of experience in these technologies at present, the limitations of current technology or simply the experience of those developing the systems.? In all this, how does one prove that a system is safer without real world data?? This may pose a 22 as to prove a system one must release a system and then prove it is a greater risk than a human.? The validation models and approached at this point, especially for high complexity and AI systems are often immature so the argument of safety is speculation at best.

Frequency vs Visibility

In all this one should consider that because an event is or is not widely reported it does not make it more of less frequent.? The media and regulatory bodies do a vital job in highlighting serious events which occur daily in our world.? However, the attention is often on the most serious, high profile or novel events which occur and not the most frequent.? A classic example is mortality on road, rail and air.? In the UK in 2022 there were:

·??????? ROAD:? 1711 fatalities (Dept of Transport, Reported road casualties Great Britain, Annual Report: 2022)

·??????? RAIL:? 263 fatalities of which 236 suicide (Office of Rail and Road https://dataportal.orr.gov.uk/statistics/health-and-safety/rail-safety/)

·??????? AIR:? 11 fatalities (Air Accidents Investigation Branch, Annual Safety Review 2022 )

What may be more interesting are injuries and near miss events.? However, in the case or road this is? near impossible to determine as a vast majority are not reported.? This is lees so in the case or rial and air transport where stricter regulation enforces reporting.

But of these the rail and air events were far more prominent.? Road incidents tend only to? be reported if extreme or involving high profile persons.? Therefore, we should be careful in allowing reported incident to be too influential in the perception of risk.

To safe or not too safe, that is the question.

When the argument s for and against are looked at they are in fact both true (I hear the cries of remove the fence!).? the question is one of where and when they are made.

The positive argument looks at high risk, close environments where the external risks can be controlled, and the internal risks are high.? There typically are mining and industrial applications.? The argument against loos at more public spaces where the internal risks are lower but the external risks are higher (taxis, delivery systems, domestic applications).? Here lays the key to? this question.? It is one of the applications.? We should consider the following factors when looking into making the claim of risk? reduction

·??????? Level of internal risk

·??????? Controllability of external risks

·??????? Severity of transposed hazards

·??????? Controllability of system

Level of internal risk is a measure of the unmitigated internal risks which automation can mitigate.? The lower this is the lesser the overall benefit of automation.? So in a relatively benign environment the internal risk levels may be lower than in a very hazardous environment.

The Controllability of external risk is a measure of the extent to which the hazards outside the system be controlled. This could be measures such as exposure or proximity to the system.? Separation, interaction, weather and ground conditions.? In a public space this is difficult as access cannot be controlled and hazards are difficult to predict at any given moment.

Transposed internal to external hazards will occur in every application.? The severity of the external hazards created by automation needs to be considered as the creation of sever hazards, even if the exposure time or frequency is low needs to? be assessed as cumulatively these could? raise the overall risk level.

System controllability is a measure of how predictably the system reacts to a given input.? Low controllability requires a greater degree of human judgement and experience.? As a rule lower controllability systems transpose a higher degree of risk than high controllability systems, reducing the overall risk reduction.

To Conclude

In conclusion the assessment of? overall risk reduction from manual to automated system is a valid measure of the effectiveness of automated systems.? At the more detailed level though this is not a valid measure as the risk profile changes.? This change is dependent on the application and the types and frequencies of hazards present.? Reported incidents are not a good measure though and actual statistics require a greater deal of consideration and the reliability of reporting is key.?

Safer than a human?? It depends!