How NLP can help transform mental & behavioral health services

In the intro to his new book “The Patient will See you Now: The Future of Medicine is in Your Hands”, author Eric Topol describes how technology is poised to democratize medicine by drawing the analogy of how the printing press liberated knowledge from the control of an elite class. I love this analogy, not least because I believe the written word is likely to play a central role in reshaping mental and behavioral health services. Here’s why:

1. In the world of therapy, cognitive behavioral therapy (CBT) is king. It took therapy out of the dark ages of endless talking and analysis of dreams to being problem focused, time-limited and evidence-based. CBT rests on the assumption that we feel bad not because of events but because of how we think about those events. And language is the best proxy we have for our thoughts. CBT helps people understand how specific distortions in their thinking are undermining them. So language is central to this therapy.
2. Natural language processing (NLP) algorithms are getting more and more powerful, especially in a deep learning context. It is not going to be long before NLP will be capable of deciphering the lies you are undermining yourself with, based on the biases embedded in the things you say. For example, having repetitive thoughts along the lines of “I didn’t get that promotion and I feel like crap” eventually builds a meta picture of a person’s underlying “core belief” or bias, in this case, that they are only worthwhile while they are achieving.
3. While this is currently a complex task, it’s significantly simplified by what Beck (1979) called the Theory of Cognitive Specificity. This is the idea that certain problems are related to specific cognitions, which therefore feature similar distortions. For example, people with anxiety tend to have a lot of thoughts that include fortune telling (I’m going to fail). People with depressive thoughts tend to magnify the negative and use all-or-nothing thinking (I’m a total failure). Anger is usually associated with a sense of entitlement expressed in a should statement (he shouldn’t have said that) and so on. This specificity dramatically reduces the necessary complexity of an algorithm in the context of illness.
4. The current model of treatment is not working very well and even if no disparities of access existed, there will never be enough expertise to go around. We have to start outsourcing some of the easier tasks so that humans spend their time and expertise on the difficult cases. There is no reason why we can’t task shift the business of identifying self-defeating and maladaptive patterns in behavior and thinking (vis-a-vis language) to an NLP algorithm that is silently working on our behalf from our mobile device. After all, a regular therapist only gets to see you once a week and is limited to the data that you chose to present. Your mobile device — assuming a sufficiently accurate algorithm — could have of thousands of daily data points from which to identify patterns.
5. I understand that the therapeutic relationship is curative for many people but these are actually complementary roles. Let your device do what it does best and be your personal analyst by gathering data to reveal your maladaptive behaviors, and let the humans do the human work of helping you change them.

Originally published at https://www.linkedin.com on January 15, 2015.