VTT, one of Europe's leading research institutions, has developed an AI-based tool that can recognise when knowledge workers are under stress. The tool can be used in organisations to be aware of stress levels, identify root causes, and act early on, which helps to reduce or avoid stress-related sick leaves.
Work-related stress and its follow-up effects have become a major issue globally. Employers are actively seeking ways to address this issue, but even the detection of stress poses challenges because few tools can do it reliably. In collaboration with the Finnish Institute of Occupational Health, VTT has developed a tool to distinguish between stressful and non-stressed conditions of knowledge workers by analysing behavioural data from computer usage and converting this data into easily understandable metrics.
From Research Challenges to Practical Solutions
The AI-based algorithm assesses an individual's mouse movements and detects changes in behaviour, as unusual behaviour can be an indication of stress. Since not all deviations from the norm are due to stress, especially because behaviours vary significantly between individuals, the algorithm needs to be calibrated for its user.
"Different approaches could be used for this", says Atte Kinnula, Senior Scientist at VTT, "in our research, we opted to ask subjective self-reports of whether users felt the day was stressful or not. Currently, our AI learns with about 30 reports, which is about 1/3 of the amount comparable studies have used. For the solution to be practical, this learning phase must still be improved."
Over the past four years, VTT researchers have collected data from real-life workplaces to teach AI. At this point, it can detect stress with a 71% accuracy daily and 84% accuracy over three months. By adding other behavioural data, such as typing tempo, the scientists at VTT think it is likely possible to improve the accuracy beyond this. One of the challenges the research team faced was that algorithms used in laboratory studies didn't give good results with real-life data.
"Our ability to get these results with data from actual regular work is a major step towards making such solutions practical", Kinnula says.
To present the results easily, positively and engagingly, VTT developed the concept of an organisation barometer. This well-being data map shows the stress status over time at a glance.
"We researched to determine the most effective approach for presenting data, particularly the potential of visual metaphors for this purpose and how they affect the user experience", says Senior Scientist Julia Kantorovitch from VTT. "It was great to see comments such as '(This) tool makes you reflect on how you use your time, so you benefit immediately' and 'this kind of data presentation is useful for the organisation when it can be discussed openly to collectively think of correcting actions' from workers who participated to the study", she continues.
Privacy and Acceptance are Paramount
The barometer aims to give the organisation a tool to view the workforce's stress situation while preserving the privacy of individuals. Preventing data misuse and ensuring complete transparency in data handling are crucial. One of the key premises is that if you measure an individual's behaviour, that individual must own the data and the results and employees should have the autonomy to choose whether their results are shared or not. Furthermore, all collected data must undergo thorough anonymisation. This means that not only should it be impossible to identify individuals from the shared data, but it should also be impossible to know who has or has not shared their results.
VTT has collaborated with cybersecurity company Nixu to examine the system's operational framework, safeguard privacy, and implement principles such as MyData, which revolve around the sovereign ownership of data. Cybersecurity company Silverskin has also audited the code of the solution to meet the security requirements of organisations that could take it into use.
While privacy is critical, acceptability is equally important – if employees don't find such tools acceptable, they will not be taken into use. To study how willing people would be to share their stress data anonymously with their organisations, the research team carried out a large international survey with over 1200 responders among knowledge workers. The survey data shows that 80% of responders would be willing to do so if it is used to support their own and colleagues' work well-being, a figure sufficiently high to give organisations a reliable view of the situation.
Hope On the Horizon
While technology is frequently held responsible for the hectic atmosphere of modern workplaces, it can also be harnessed to help. The increasing rate of absences related to stress-related illnesses is concerning. Still, research is exploring novel methods to address this issue.
"With the utilisation of AI-based solutions, VTT has demonstrated one of the more promising approaches", Kinnula says. "The results have been auspicious, and while the technology is still in early development, we can foresee that it will reach maturity in 3-5 years. Currently, we are looking for partners to develop the technology further".
The research thus far has primarily focused on knowledge workers. Still, VTT is expanding its testing to various work environments, including factory settings, aiming to bring out tools to monitor and manage stress across different walks of life for employers and workers alike.
New Algorithm Boosts Athlete Posture Detection Accuracy