Written by Fredrik Löfberg
Recent years have seen the rise of Artificial Intelligence (AI) evangelists. We've seen incredible advancements, such as computers beating humans at Go using new strategies. The most fascinating part of this achievement was that the AI actually learned how to play the game. It couldn't win by just using raw computing power to predict all moves, as there are so many variations. The same can be observed in robotics when robots aren’t programmed to walk, jump or climb – but learn and train just like a toddler does.
Yet, in all these situations, the AI was designed and trained to accomplish a very specific task, such as to win Go games. This is where we are now. We can teach machines to be better than humans at very specific tasks. And these are usually tasks that require capabilities that are analytical in nature, like visual observation, logical decision making, and mathematical operations. But as powerful the AI is at the GO game, it would lose to a child if the task were to organize colored pencils by their place in the rainbow.
How AI will change safety and risk management
For health and safety and risk management tasks, humans often employ visual concepts for data in order to better observe the conditions, compliance, best practices, and apparent hazards. We might even take it so far as to say that we mostly use visual observations to do this. We use it to both understand and eliminate the risks we find in our workplaces. With proper training, the human brain is pretty good at accomplishing this, even with a limited amount of data. By training, we refer to knowledge about risks that a person has acquired as a result of their experiences and learning. And with a limited amount of data we are discussing, that one human can maybe analyse hundreds of pictures per day, but there is a lot of room for human error when the focus is lost, or the awareness levels are not at our best.
Visuals are a case where machines, applying first generation AI machine learning, consistently deliver powerful performances. AI models are taught to recognize hazards from many single pictures, which is the same way EHS professionals learn to do their jobs. We use large-volume image data sets and feed the machines information about all the elements depicted in the image and which are relevant. With this training, the machine learning model learns to identify the point of interest and will outperform even experienced EHS professionals in spotting potential risks. However, humans will remain the final decision makers, hence the technology can be utilized as an Augmented Intelligence that on the one hand raises alerts when needed and on the other hand frees the human users from boring routine tasks thus allowing them to focus on other more engaging topics.
First steps towards artificially intelligent EHS
The automation level using intelligent solutions will rise rapidly in the coming years. It is crucial that every EHS manager is aware of this trend and stays ahead of the game to be able to drive an organization’s EHS strategy towards the right direction. AI-based tools will soon act as assistants in processing and analysing data, both visual and written, and then gradually take over routine tasks such that straightforward actions can be automated. In doing so, machines will help humans to focus on more conceptual topics, deliver a better performance on routine tasks and thus help us do our jobs better and safer in the future.