Can AI support our mental health?

AI technology offers incredible opportunities for improving mental health treatment. For far too long, mental health was a stigmatized topic that people avoided discussing for fear of being judged. Thankfully, more people are seeking care, which means the demand for solutions has increased! Additionally, numerous breakthroughs in machine learning and its applications lead me to think that something amazing will happen in this field.

Deep learning (one sub field of Machine Learning and AI) can support the diagnosis and understanding of mental conditions, including autism spectrum disorder, depression, and even neurodegenerative diseases.

David Luxton writes about this topic in-depth in his ground-breaking book, Artificial Intelligence in Behavioral and Mental Health Care According to Luxton, AI has is able to provide a greater understanding by leveraging visual and speech-based temporal data for behavioral health.

The recent progress in machine learning (typically computer vision but not limited to this) is extremely useful to applied sciences, such as biology and neurology. We’ll we don’t expect to see “pure-AI” solution for health, there are several benefits it can provide that I’ll outline below.

For example, Stanford’s Dr Guido Pusiolo and Dr Andre Esteva have researched one such data type for Fragile X Syndrome (FXS) autism spectrum disorder is a patient’s eye-gaze in response to an interview with a physician. Let me walk you through the steps:

• You build a multi-modality data collection system consisting of a camera viewing a physician’s face, and an eye-tracker following a patient’s gaze
• Time-series data can be constructed that indicates the location of a patient’s gaze at any point during an interview.
• If you review the findings from a large number of interviews, this data can be used to train recurrent neural network (RNNs) to discern between patients that exhibit autism spectrum disorder from different developmental conditions Similar methodologies have been employed for other conditions such as ADHD

Neuro-linguistic programming can already classify the variation in the human voice when talking, whispering, sighing, laughing, yawning, or crying And applied clinically, vocal patterns in patients have been used to diagnose clinical depression as well as neurodegenerative diseases such as Parkinson’s Disease These approaches have historically been based on small datasets in which raw speech data is first hand-featurized and then used to train traditional ML models. However, as consumer and clinical devices collect large voice datasets, deep learning could enable end-to-end learning of diagnostic labels directly from raw speech data.

Recently, medical chatbots have emerged as computational agents capable of simple, healthcare-related dialogue with humans, via text or speech. Woebot, a sophisticated AI assistant for depression created by Dr Alison Darcy is a example. Those chatbots tend to follow decision-tree conversational structure, in which the user first answers a question, and is asked a follow-up question based on their answer. This process continues until sufficient data has been collected for the goal at hand. Here, deep learning (specifically, language-model trained RNNs) enhances chatbots by improving the understanding of the human responses, which can be semantically and syntactically broad, as well as improving the quality of the machine-generated responses to the humans Additionally, studies show that cognitive-behavioral therapy, administered through a mobile chatbot, can reduce depression and serve as an adjunct therapy.

Originally Answer by Alexandre Robicquet, Co-Founder of Crossing Minds, on Quora.