Funding supports research into human error in AI

Research into the role of fatigue, workload, and stress in artificial intelligence is the subject of a GB£150,000 ($292,100) grant from Thales to academics and PhD students at UniSA.

The research will study how the effect of sleep deprivation impacts wellbeing and cognitive performance.

The data will then be used by Thales to ensure that their AI products optimise and enhance, instead of degrading, an operator’s performance, particularly in marine environments.

According to Professor Siobhan Banks, co-direct of the Behaviour Brain Body Research Centre, at UniSA, noted that the approach taken by Thales is distinct from other AI projects.

“By considering the human early on in the design process, we can manage a lot of the errors that arise when human factors aren’t taken into consideration and which can result in the development of a tool that doesn’t work. This is a much smarter approach.”

Thales provides services that are part of the upgrade of the Collins Class submarines, and is competing for work on the next generation Future Submarine and Future Frigate programs, both of which will utilise the latest technologies to improve the performance of sailors on each platform.

As long shifts are common in maritime environments, the Behaviour Brain Body Research Centre will utilise insights from adjacent fields such as mining, transport, and other military sectors to understand how to best improve the environment for sailors operating submarines and surface vessels. Ensuring that seamen and women are best prepared to oversee AI operations will be part of the effective roll-out of predictive tools on vessels, said Banks.

“More research into AI is needed to understand how people can use it effectively. Removing or changing tasks with AI could also have unexpected negative effects, especially when operators are fatigued or stressed,” she said.

One recommendation, Banks suggested, would be the machine encouraging the operator to have a cup of coffee.

“There are opportunities to tie the monitoring of human states into the AI systems. By tracking operators, we could know when they become too stressed or fatigued and spot early signs of errors so we can intervene, perhaps switching them out of a role, using augmented reality to improve information flow, or suggesting the person take some caffeine.”