Imagine driving a high-tech car where artificial intelligence helps you brake. Everything seems perfect, until the car decides to brake and you, confused, hit the accelerator. What went wrong? It is not just the software's fault, but the way the human and the machine interacted.
Many experts are realizing that assessing AI safety is a bit like looking at only one piece of a giant puzzle. We often focus solely on code and algorithms, forgetting that AI does not live in a vacuum. It works alongside real people and within organizations with specific rules and procedures. This ensemble is known as a 'ocio-technical system.'
A new study has highlighted that current risk assessment methods often overlook this very aspect. To solve this, researchers have proposed a new variable called 'Socio-Technical Alignment' (STA). Think of it as a thermometer that measures how well the AI, the human operator, and organizational processes are 'in sync.' If the alignment is low, the risk of accidents increases, even if the AI seems perfect.
To test this idea, scientists analyzed a complex system managing liquid hydrogen. They demonstrated that including human and organizational elements in the calculations makes risk assessment much more accurate and reliable. In short, to make AI safe, we must stop looking only at the chips and start looking at how we interact with them.
