I would like to say everything is working out wonderfully, but it is not true. Not many applications in the field of mental health work really well. Monika Redzisz sits down with Bartek Skorulski from Alpha Company
Monika Redzisz: Depression and other mental disorders are on the rise in Poland, particularly among young people. Access to specialists – psychologists, psychiatrists, – was never sufficient, now it has got worse dramatically. Can therapeutic applications serve as real help in this case?
Bartek Skorulski*: Yes, to an extent. Therapy, particularly the cognitive-behavioral therapy, can be partially done on the phone. Our research shows this kind of therapy can be partially automated. But it is a big challenge.
Why?
Applications are like drugs – we do not always want to use them. What can we do so that people use this particular tool, and more so – use it regularly? The application probably has to be a part of the whole system, it has to be sold as part of a package. Telefonica gives us this option.
Telefonica is the fourth telekom in the world, in terms of size; what does a cell phone operator have to do with therapeutic applications?
Telefonica operates in Spain and almost the whole of South America. Additionally, in the UK and Germany where it is known as O2. They have a gigantic number of clients. At some point, they came up with the idea to use their data not only as a source of training advertisement models for phone sales and the service, but also, to put it in general terms, for the good of the users. They founded Telefonica Alpha which deals with projects linked to mental health: addictions, sleep problems, chronic illnesses or depression. I work there.
How did you, a Polish mathematician, end up there?
I think sciences are overly theoreticized. Undoubtedly, many mathematicians will take offense at such a statement but I do not think that mathematics, even though it is a marvelous and engrossing discipline, is a bit of an art for art’s sake – axioms are corrected, there are not many ground-breaking things. That is why, at one point, I have decided I would like to do something more practical.
With applications, just like with drugs – we do not always want to use them. What can we do, so people use this tool, and what is more, use it regularly?
Besides, at the university, even before getting my Ph.D., I was linked to a project which checked whether mathematical proof was correct. I was really interested in artificial intelligence but the leader of our team said it would be better to do a Ph.D. in “normal” mathematics. Twenty years ago, a Ph.D. in abstract mathematics seemed more practical than artificial intelligence…
So, I did a Ph.D. in mathematics in Poland and left the country. I worked at universities in Chile and the USA.
But when I read about the new, unusually effective methods of image recognition, it got me really interested. I also did one of the first artificial intelligence courses available online: by Sebastian Thrun and Andrew Ng. The courses began a gigantic revolution in academic education known as MOOC, massive open online course. Their creators founded two educational platforms: Udacity and Courser which democratize knowledge in an unusual way. It also influenced my perception of university as an institution which, on one hand, is no longer able to actively follow the changing world, and on the other – lives up to the role of knowledge provider for the general population. In countries where university education is highly expensive, academia deepens the social stratification. It is visible in the USA, but much more so in Chile. It points to the fact that the latest protests in Chile, as with many previous ones, were started by students and pupils aware of their difficult situation.
Summa summarum, I came to the conclusion that I would like to leave the university and focus on artificial intelligence, or data science in general. Leaving was a big relief for me.
I moved to Spain where I worked in a company, among others, King which creates video games for cell phones (its most famous game is Candy Crash) and in Lidl where I came up with recommendation systems and sales forecast. I also collaborated with the University of Barcelona in the design of data engineering – online courses. I finally ended up in Telefonica Alpha.
An opportunity to create tools which influence human life in a concrete way, gives enormous power at work.
Were you able to create an operational tool that works well at Alpha?
On one hand, we have many nice prototypes. We are working on numerous projects. We are pragmatic. There is a lot of research in the world, and a few implementations. What we want to do are exactly that: implementations. To test the tool as fast as possible. If it works, then great, we throw more resources, people at it; if it does not work, then we try to do another thing.
These projects are what we call ‘’subclinical.’’ Sleep problems, chronic stress, burnout. You can work on this with the so-called regular people, not patients. On the other hand, we develop clinical projects, but they take a long time.
Such as?
For example, a tool for alleviating the symptoms of the body dysmorphic disorder. BDD makes you think you look terrible, we could say physically deformed. A surprising number of people suffer from this condition. For example, 2.5 percent of United States citizens. For this reason, these people often do not get out of the house.. the application is currently undergoing clinical trials.
Why are clinical projects such a challenge for you?
We deal with patients – people who suffer from mental illnesses. We need to be particularly careful, we are responsible for them. Additionally, clinical tests are strictly regulated by governmental institutions.
Psychological therapy can be partly automated. But it is a big challenge
What is the efficiency of the application?
First trials indicate that even a bit better than that of the doctors. We will know the exact results when we finish the clinical trials. It is worth noting that our application has the same success rate as a medical professional. It will mean that a psychiatrist will be partially replaced which will give us an opportunity to make treatment more accessible, democratic.
Do you use artificial intelligence here?
To a degree. We use artificial intelligence when we want to personalize therapy, recognize emotions, or predict an incoming crisis. AI is just a tool, it has its limits. It is in fashion now, but I think it is best to start with simpler methods and use AI only when they fail to work. I like machine and deep learning a lot, I would use artificial intelligence everywhere if it were up to me, but it has to be done prudently.
Where do you get the data, on which you test the algorithms, from?
From hospitals in Birmingham and Massachussetts; it is the data of patients with depression and other mental problems, sometimes after suicide attempts. We also collaborate with universities in Great Britain – mainly with the London School of Economics; we have access to their database.
Can you predict crises?
It is a difficult question. A pilot study is supposed to give the answer. We will soon begin the study in Birmingham hospitals. The problem is that, even though our algorithms on historical data give good results, it can turn out to be dramatically different in practice. When it comes to mental health, there often are no clear indicators, no clear data, about the symptoms… a patient fills out a form and a mental health professional, and the diagnosis of depression is given. These criteria are quite subjective.
Suicide attempts are a concrete criterion…
But to know earlier that a patient will attempt suicide in the future, we needed personal data of this individual. Where do you get it? Most of the data is anonymous.
Maybe from Telefonica? I understand that it was the original idea – so that data which they gather, can be used for public good.
In the future, maybe. We are in talks with government institutions in Great Britain to be able to link the personal data with the hospital data in such a way, as to guarantee anonymity. It is not easy from the legal and technical perspective. Well, artificial intelligence needs much data and we currently have no access to it.
I like machine and deep learning a lot, I would use artificial intelligence everywhere if I could, but you need to do it prudently
Which of the applications has already been implemented?
Remix operates in Spain – a program for young people which monitors their mood. We are currently rolling out a new application, Evermind – it is targeted at corporations and is used to tackle chronic stress and burnout.
Does Remix work well? Does it fulfill its purpose?
It works, but what does that mean, exactly? It improves the frame of mind – a bit. We had three thousand users daily at times so we gathered considerable data on their mood. We looked at not only what they write but how they write – how often they use applications, in what way, which activities they enjoy and which they do not. We trained many models on this information. We did recommendation systems, we correlated the way of using the application with personality. We also prepared an agent which recommends various mood-improving activities. These things worked. Still, the bigger problem with Remix is that people play with it for a couple of days, and then they leave it be. How to convince them to use it longer?
Especially today when so many phone applications are stealing our attention…
I see you are disappointed. I am trying to be honest. If you spoke to my boss, he would probably advertise these products. I am currently more skeptical. Especially that the matter of data protection is extraordinarily important to me.
I would like to say everything works wonderfully, but it is not true. Not many applications in the mental health field work really well: on one hand, it gives the statistical results better than a placebo, on the other it is sufficiently interesting for users to regularly use them for a longer period of time. Our team is working hard to make our products this way.
*Bartek Skorulski works as a Data Scientist in a research department of Alpha. He is a Team Lead responsible for the personalization of therapeutic applications supplied by cell phones. As a Data Scientist, he previously worked at Schibsted (he helped develop the Messenger for Leboncoin portals, Willhaben, Milanuncions) SCRM-Lidl (he led the team responsible, among other things, for machine learning in Lidl+ application) and King (a company known for Candy Crush, Bubble Witch and others). He has a Ph.D. in mathematics, specializing in dynamical systems, he worked at universities in Chile and the USA. Presently, he does courses on machine learning, deep learning and data science at the University of Barcelona and at the Barcelona Technology School.
Depression epidemic
According to WHO data from December 2017, over 322 million people across the world suffer from depression, in Europe alone – 29 million, in Poland – almost 2 million, that is over five percent of the population. The number of the ill is increasing – between 2005 and 2015, it increased by more than 18 percent. Women get depressed more than men (5.1 percent of women and 3.6 percent of men). Depression often leads to suicide. In 2019, 800 thousand people died by suicide. It is the second cause of death for people between 15 and 29 years of age.
The economic costs of depression are enormous – according to WHO estimates for 2013, the illness has cost Europe 170 billion euros, and Poland – over a billion zloty (according to the data of the Safety and Health Management Institute). Due to inability to work as a result of depression, 290 thousand sick leaves were granted, totalling together over 15 years.
In Poland, 800 thousand people are pharmacologically treated for depression, but many of the ill do not undergo therapy. It is estimated that in some countries it is even 85 percent of the ill. The reason is lack of access to specialists, drugs and therapy, but also the fact that depression, as well as other mental illnesses are still a shameful taboo. It turns out that, according to WHO expertise, depression will be the most common, the most widespread, of all illnesses that affect us.
Przeczytaj polską wersję tego tekstu TUTAJ