For doctors and clinicians to be able to diagnose depression, they need to ask specific questions, related to past mental illness, for example, or lifestyle, mood, etc..
With computers getting better and the advancements of AI, researchers from MIT experimented with machine learning to build a neural network to recognize this human mental health disorder.
The AI works by listening to conversation, to detect signs of in speech and text.
In a paper presented at the Interspeech Conference, the researchers detail their neural network model that can be unleashed on raw text and audio data from interviews to discover speech patterns indicative of depression.
"The first hints we have that a person is happy, excited, sad, or has some serious cognitive condition, such as depression, is through their speech," said the first author Tuka Alhanai, a researcher in the Computer Science and Artificial Intelligence Laboratory.
The model is being referred to as "context-free" because the AI analyzes how things are being communicated, rather than what is being communicated.
The AI is advanced enough that given a new subject, "it can accurately predict if the individual is depressed, without needing any other information about the questions and answers," according to the researchers.
To test their AI, the MIT researchers experimented with 142 people being screened for depression by making them answer a series of questions asked by a human-controlled virtual agent. The AI had no prior knowledge of the questions, and the people were free to answer the questions in any way they wanted.
Here, the AI was able to recognize depression just from linguistic cues.
In the initial tests that exposed the AI to audio recordings, the AI had a success rate of 77 percent. The AI has outperformed other models which relied more heavily on the "question and answer" structure.
A key insight from the research was during the experiments, the model needed much more data to predict depression from audio than it did with text. With text, the model can accurately detects depression using an average of seven question-answer sequences. Whereas with audio, the model needed around 30 sequences.
Due to the flexibility of the model, James Glass, a co-researcher on the project, thinks that the method has the potential. It can be developed as a tool to detect the signs of depression in natural conversation, such as a mobile app that monitors its user's text and voice for mental distress, and send alerts to doctors if necessary.
.
This advancement of passive automated monitoring of human communication sounds eerily scary.
If the project is to be true to realization, it may impact the society where a person's mental health could be subjected for an evaluation by a machine.
What's more, Therapists who see patients tend to believe their years of training and experience to diagnose patients have made them better than computers, because after all, the human mental health can only be judged by another fellow human, not by an emotionless machine.
For the foreseeable future, the team is looking forward to expand the model's capabilities by including additional data from patients with other cognitive disorders.
Alhanai added that the model's potential is "not so much detecting depression, but it's a similar concept of evaluating, from an everyday signal in speech, if someone has cognitive impairment or not."