Leveraging AI to diagnose COVID-19 patients

For COVID-19-related tracking, AI has been leveraged in two critical ways. First, through aggregating information from various sources, such as twitter, facebook, daily CDC website and news reports, data systems could be leveraged to identify and manage Covid hotspots. Secondly, both governments and private companies have partnered together, and in some cases implemented, AI based tools to support surveillance of public spaces through monitoring data.

CDC in partnership with Microsoft developed Clara- an automated chatbox to prioritize screening patients who have cold or flu-like symptoms. It is a "coronavirus self-checker" automated box that helps determine who needs “access to limited medical resources”, and avoid creating a “bottleneck that threatens to overwhelm health systems coping with the crisis. Based on a deep learning model to identify at-risk populations in public health, Clara supports proper and quick medical diagnoses for COVID-19 cases.

But it is critical to note that AI should be used judiciously and in proper context. From 2009 to 2015, Google ran an effort called Google Flu Trends that looked for key words in chats, social media platforms such as twitter to create a prediction score for the prevalence of flu in US. Initially this system was a great success and was able to correctly predict CDC tallies about 2 weeks ahead of time. However, during the period from 2011 to 2013, it started overestimating the prevalence of flu largely due to its inability to distinguish google and twitter search with actual patient symptoms. The result led to a wide overestimation of the flu because the system was not trained on the training set to evolve with people’s search behavior.

Firms and governments relying on machine learning systems and AI to track COVID-19 cases should closely follow current developments in this space. Its is more than critical to consult with counsel as reforms to the legal framework governing these technologies are institutionalized.