Artificial intelligence learns to improve occupational safety

In the Finnish Institute of Occupational Health’s AI Safety project, artificial intelligence was trained to identify various things and events called human factors in the background data of accidents.

The identification of background data would make it possible to shift the analysis not only to the individual but also to the organisation’s actions and work practices, which would enable a genuine and effective development of safety.

“So far, we have been able to identify more than 80 per cent of the human factors behind accidents correctly with AI. This is a fairly good result, even though artificial intelligence is not yet unparalleled in terms of identification accuracy,” says Maria Tiikkaja from the Finnish Institute of Occupational Health.

In order for artificial intelligence to be able to detect these human factors, they should be recorded in companies’ safety observations, near miss reports and accident reports.

The Finnish Institute of Occupational Health’s human factors tool consists of four components that influence the background of human activities: the actions and characteristics of the individual, work activities – including the quality and quantity of work and work methods – work characteristics, as well as group factors and organisational factors. Each component has a number of subcomponents.

In order for artificial intelligence to be able to detect these human factors, they should be recorded in companies’ safety observations, near miss reports and accident reports.

“Instead of just recording ‘I slipped on the stairs’, employees should record more extensively the background of the event, for example, ‘I slipped on the stairs because they were wet and I was busy’,” Tiikkaja describes.

The next goal of the Finnish Institute of Occupational Health’s research group is to teach artificial intelligence to predict the development of safety, i.e. identify risks more accurately.

For example, if human factors are repeatedly recorded, artificial intelligence may detect that companies experience a lot of slipping, which may be caused by employees being in a rush or water on the floor.

“Then the company could think about why there was water on the floor three times in a month,” Tiikkaja says.

The next goal of the Finnish Institute of Occupational Health’s research group is to teach artificial intelligence to predict the development of safety, i.e. identify risks more accurately.