How artificial intelligence searches texts for signs of burnout

Work is an important factor in our society. It creates the economic basis for many people’s existence, giving their life structure and purpose. At the same time, work is demanding ever more of people due to increasing productivity pressure and connectivity, and in some cases this can become overwhelming. The latest government figures indicate that 30 percent of the working population in Switzerland say they’re emotionally exhausted. Ten years ago, this figure was just under 24 percent. Overwork and chronic stress can lead to burnout. A research team at BFH is hoping to use computer-aided analyses to help identify burnouts more quickly. In an interview, project manager Mascha Kurpicz-Briki explains where she’s at with her research on burnout diagnosis.

You have set yourselves the goal of using computer-aided analyses of texts to identify whether someone is suffering from burnout. How did you come up with the idea?

Burnout has become a problem in our society. Ever more people are under constant stress, and if it gets to a point where they feel overwhelmed, this can lead to burnout. I asked myself how the rapid pace of developments in digital technology can be exploited effectively to benefit people who have suffered burnout.

What benefits can artificial intelligence bring to the diagnosis of burnout?

Multiple-choice questionnaires that simply involve ticking boxes are the accepted method for diagnosing burnout. Although these tests work well, the disadvantage is that patients can influence the result by not answering a question truthfully. By analysing answers to open-ended questions or evaluating the transcripts of interviews, this can improve the accuracy of a diagnosis. However, such procedures are complex and time-consuming for the experts. This is where the new technological possibilities come in: Computational Linguistics applications can analyse such texts in a matter of seconds, i.e. recognise patterns in the answers that are typical of a person affected by burnout.

Can you briefly explain how Computational Linguistics works?

Computational Linguistics, also called Natural Language Processing, is able to process language or texts automatically. Not only can the programs extract certain information or structures from large volumes of text or extensive voice recordings, but – and this has recently made headlines – they are also able to create texts, as with ChatGPT, for example. The applications benefit greatly from the principle of machine learning, in which the computer learns from large amounts of sample data.

How exactly does artificial intelligence recognise a burnout pattern in texts or conversations? Does the programme search for specific words or sentence structures?

In the project, we are trialling various technologies. Programs based on machine learning don’t simply work on the basis of individual words, but also include other aspects such as stylistic text elements in the analysis. However, more research is needed before we get to the point where the system can tell with a fair degree of certainty whether the person who has written a text is suffering from burnout.

Can’t people fool the system by deliberately making certain statements or avoiding certain words?

With traditional questionnaires, it is easy to falsify the results. For example, if a question asks how often they feel depressed, they can simply tick “rarely” instead of “often”. Even with automated systems, manipulation can never be ruled out entirely. But unlike with conventional questionnaires, patients do not know what signs the programme will be looking for when evaluating their answer. This makes it much more difficult to influence the result.

Will burnout diagnoses soon be made by machines and no longer by people?

No, that wouldn’t be appropriate We want to empower people through technology, not replace them. Machine systems are tools designed to support people. In our case, the actual diagnosis will be left to a specialist, using not just elements such as personal interviews, but also the findings of Computational Linguistics.

Can artificial intelligence also be used to diagnose mental illness or dementia?

In principle, yes. We have another project running in which we are trying to recognise eating disorders with the help of computers. Wherever psychology involves analysing questionnaires, texts or conversations, there is potential to support and underpin diagnoses with technological applications.

When do you think computer-aided analyses of texts or conversations will become common practice to detect burnouts?

It will still be some time before that happens. The results of our research to date are promising: we’ve been able to train the system to a point where it is basically able to distinguish whether a text comes from a person with burnout or not. But further validation of the methods and the results is needed. We also have to clarify with clinical institutes in what form and in what procedures the applications could be used.

And then we would need to adapt it to different language versions. We can’t develop a programme in English and simply translate it into German, French and Italian. That wouldn’t do. You have to take into account the subtleties, the particular idiom of the individual languages and also the cultural differences that are sometimes reflected in the languages.