Diagnosing burnout – can AI help?

Fatigue is pretty much what it sounds like. Exhausted people often feel tired, irritable, depressed, and tense. You don’t get any rest and no relaxation even if you already have time. Physical symptoms can also occur. According to the Max Planck Institute (MPI) of Psychiatry, up to 20 percent of the world’s population suffers from various forms of fatigue.

However, fatigue is often recognized too late, even though it shortens life expectancy as a risk factor for other common serious illnesses such as depression, heart attack or stroke if left untreated. Therefore, early detection is especially important.

In both interviews and questionnaires, there is a risk of false results due to so-called response tendencies. These occur when respondents not only interact with the content of the question, but are also influenced by other factors in their answer.

PeopleImages, GettyImages

Why is diagnosis often difficult

However, fatigue is often recognized too late because symptoms often seem nonspecific and resemble those of depression or anxiety disorders. Fatigue is also manifested in the psyche and cannot be read from objectively measurable values ​​or vital signs. Therefore, the diagnosis can only be made by discussing with the sufferers. In addition, standardized questionnaires are often used.

However, there is a risk of false results in both interviews and questionnaires. Similar to personality tests, people subconsciously tend to choose ‘extreme’ answer options such as ‘never’ or ‘every day’, rather they tend to answer ‘occasionally’ with caution and with frustrating effect. In addition, some people are good at describing their inner lives and sensibilities, while others are not.

How can artificial intelligence help?

This is where AI comes into play: researchers at the Bern University of Applied Sciences had the idea to use machine learning as a diagnostic tool. This should enable effective analysis of freely written texts and identification of signs of fatigue among authors.

With this type of AI, a computer is able to “learn” without being specially programmed. It receives none or just a few of the predefined rules, but the learning algorithm automatically recognizes the regularities it finds in the examples given to it for training. In the course of training, he learns to assign repetitive patterns to specific categories. This form of AI-powered analysis now works with both written and spoken language. Experts refer to this as Natural Language Processing (NLP)in German “machine processing of natural language”.

This linguistic assessment is already being used by language assistants and translators, for example, or to analyze the moods of residents in social media. However, NLP has also been used to identify various mental illnesses based on texts from social media, and the research group led by Mascha Kurpicz-Briki at Berne University of Applied Sciences used it in their experiments to analyze freely written texts for signs of being burned.

Reddit icon on smartphone screen

The freely accessible texts from the discussion platform were used to “teach” the AI.

Unsplash.com, Brett Jordan

Reddit posts as diagnostic material

In our case, the researchers wanted to train the algorithm to recognize signs of fatigue in written texts. To do this, they used texts from the Reddit discussion platform, where large data sets categorized by topic are publicly available. In this way, they were able to find specific posts with or without reference to the word exhaustion and analyze over 13,000 posts.

For training, 70 percent of the posts were divided into the different training datasets “fatigue vs. non-fatigue”, “fatigue vs. depression” and “fatigue vs. non-fatigue – no keywords”. This is how the algorithm should learn to assign posts to the correct categories. This means recognizing the signs of burnout in texts, distinguishing them from depression, or assigning them to the category of burnout without using the buzzword “burnout.” The researchers then used the remaining 30 percent of the articles to test how well the algorithm recognized text fatigue.

It is 90 percent correct

And what was the result? The researchers found that their algorithms are indeed suitable for such a task. In up to 93 percent of cases, the AI ​​system was right on the task. Even without keywords, i.e. the direct indication of fatigue in the text, the algorithm was able to correctly assess the writer’s condition and distinguish between burnout and depression.

According to scientists, artificial intelligence systems can provide valuable assistance in the difficult diagnosis of fatigue in the future. “By processing machine language, fatigue can be diagnosed accurately and quickly. This is a very promising result,” says Mascha Kurpicz-Briki. However: Psychologists who support diagnosis and who receive extensive advice and care from real people will remain indispensable despite this computer-aided diagnostic assistance and will be a bottleneck in accessing professional help.

Leave a Comment