Back in September, Forbes reported that an estimated 16.2 million people experience some form of depression in the United States. In the same vein, they have suggested that the diagnosis and treatment of such conditions are not quite as up to par as it could be.
Whatever the case is, in this respect, it appears clear that Machine Learning can change the process of diagnosing a patient. To understand how this could be true, it may help you to begin with the statistic that 37% of adults who were later found to be experiencing depression, had not ever received treatment for it before. In connection with this, Forbes also suggested that this indicates the need for diagnosing depression in those who are not seeking help for it.
When it comes to AI and this process, on August 29th, MIT news published a piece on research at their university, which found that neural networks can help to identify depression in patients.
In a general sense, what this particular article indicates is that neural networks can conclude that someone has a form of depression based on certain speech patterns that they exhibit. Before jumping into what this entails, it is important to mention that this type of research has typically been based on pre-programmed questions that are given to people then fed to the AI system later. In other words, it has usually, heavily been based on practices rooted in supervised learning.
In MIT’s case, it may be said that they are changing the game by more actively involving the AI in the process of achieving a diagnosis.
While at this time, it appears that they have not used their AI in real-time conversation, it is being used with recorded interviews. Using both text and audio in these cases, MIT reports that this network is able to “accurately predict if the individual is depressed.” If you’re asking yourself how such accuracy is possible at this point, then you would not at all be wrong in doing so.
The easy answer is that this research is a step toward improving this process, while still basing its work on supervised learning principles. As the network tries this method out in practice, again and again, it continues to get better at providing an accurate diagnosis. Thus, because of this, we do not really seem to have reached a point in this field in which we can let AIs learn how to be doctors in any sense, without significant human intervention.
In future posts, we are going to look further into this type of research as it develops, with the added aim of better understanding whether structuring AI systems based on unsupervised learning principles will ever be a trustable solution.
As with anything in the Artificial Intelligence industry, at this point, nothing is certain and nothing is completely accurate.
References:
Primary Source: http://news.mit.edu/2018/neural-network-model-detect-depression-conversations-0830
Further Reading: https://blogs.nvidia.com/blog/2018/02/13/ai-more-effective-depression-treatments/
