̽����ѡ

Text: Johan Sievers, first published in Swedish in the magazine Medicinsk Vetenskap No 3/2018

What is the issue you are trying to solve?

“In order to avoid people undergoing demanding psychological treatments that do not help them, all treatment providers should really evaluate if their patients are receiving the right kind of therapy or if it needs to be modified. But treatment providers are not that good at making prognoses, often being a bit too optimistic. Previous experience show that the prognoses improve if they are based on standardized self-assessments of well-being which are analysed in a more systematic way. We now want to take this idea one step further and investigate if learning machines, that is AI, can help us improve the prognoses.”

How will you make a learning machine outperform an experienced psychologist in making treatment prognoses?

“With the help of AI, you can find connections in large bodies of information. We have chosen to focus our study on online CBT treatment because of the vast amounts of data available in the form of about 6,000 treatments carried out at the county council’s Internet Psychiatry Unit. At the moment, we are making the data available for processing by a learning machine, it is an immense task.”

Can you give an example of what the AI can find?

“A learning machine can find connections that would take a treatment provider a very long time to find. For example, it can analyse messages sent between the patient and the treatment provider and then find patterns which point to a certain outcome. At a basic level, this would involve counting the number of negative and positive words. If it turns out that there is a connection between the success of the treatments, the information can then be used as part of a treatment prognoses.”

How would a treatment provider make use of recommendations from a learning machine?

“To begin with, we want to find out if a learning machine can make good treatment prognoses. Then, we want to create an interface so it can give the treatment provider comprehensible and useful information, and then create routines for usage of that information. Finally, we want to study if this work method really improves the patient’s treatment.”

Are there any ethical implications involved?

“If you, for example, were to let a learning machine make changes to a treatment plan it would raise questions about the ̽����ѡ responsibility. But we will only allow the machine to give advice to the responsible treatment provider and only modify the treatment so that patients receive more, not less, help and support than what they usually receive during internet treatment. In that way, we are able to address the ethical concerns regarding this project.”