Fighting COVID-19 with Data: Saving Lives with Artificial Intelligence

Original article was published on Artificial Intelligence on Medium

Fighting COVID-19 with Data: Saving Lives with Artificial Intelligence

To 21st-century observers, the COVID-19 pandemic may seem totally new, a modern-day plague with no parallels in history. But that is not the truth, and pandemics have happened in the past. What does make this situation new is the vast amounts of data and advanced technology humans now have at their disposal. In the past these scientific advances would have seemed like magic, but in the modern age advanced pharmaceuticals, targeted antiviral drugs, plasma-based therapies and widespread virus testing are so common they are often taken for granted. But at the end it could be data that will turn the tide, transforming a global pandemic with the potential to take tens of millions of lives into a more manageable disease with a higher survival rate. Here are some key things to know about the COVID-19 pandemic, the rise of artificial intelligence and why everyone should have hope.

Pandemics as natural cycle. There is nothing new about pandemics. In fact, the rate of pandemic illness is distressingly easy to predict, and experts have been warning that the world was overdue for another one. From the black plague in medieval times to the Spanish flu a century ago, pandemics have ravaged populations for millennia, leaving societies devastated and struggling to survive. But while there are eerie parallels between the Spanish flu pandemic of 1918 and the current COVID-19 virus, it is important to understand the differences as well as the similarities.

In the Spanish flu pandemic, for instance, as many as one third of people around the world are thought to have been infected, with deaths as high as 50 million, eclipsing the war dead from both world wars combined. And with a mortality rate of roughly 2.5%, the Spanish flu was hundreds of times as deadly as the typical flu virus it would morph to become.

Those parallels are certainly striking — a novel virus with no population-based immunity, a series of missteps by some government officials, a high mortality rate and the race for a cure. But there are some key differences between 1918 and 2020, most notably in the medical advances that have taken place in the intervening 100 years.

Less than five months after the novel coronavirus first emerged, there are already dozens of vaccine candidates in the pipeline, and several of them have already entered clinical trials. This is not an accident; many of those vaccine candidates are based on early coronavirus outbreaks like SARS and MERS, close cousins of the current COVID-19 strain.

Artificial intelligence and COVID-19. What is new in the current pandemic is the existence of artificial intelligence (AI) — an emerging science and technology that did not exist a decade or two ago. AI was still in its infancy when SARS struck Asia and MERS cut a swath through the Middle East. Now the technology is more mature and far more capable. It infiltrated many everyday domains from ad placements and what news we see in our social media feeds to suggesting the movie we should watch tonight. While the full extent to which AI is being used for general medical diagnosis is unknown, AI has already been employed for some coronavirus diagnosis. For example, Alibaba, the Chinese tech company, developed an algorithm that can differentiate the virus from ordinary viral pneumonia in CT scans within 96% accuracy and in seconds (see: https://www.alibabacloud.com/solutions/ct-image-analytics). But, surprisingly, mentions of AI in the context of a large-scale public health response to COVID-19 has been mostly absent from mainstream discourse.

Could AI help in the fight against COVID-19? The short answer is yes, and the return on investment would be measured in lives saved. Consider this example: one of the hardest things to predict has been which coronavirus-confirmed case prior to the development of symptoms will require hospitalization, and possibly the use of a ventilator, and who can safely quarantine themselves at home. On an individual basis, this kind of prediction is almost impossible.

AI has already proven that it is better at these kinds of community-based predictions than human judgment. The strength of AI algorithms is in the ability to quickly comb through massive amounts of data points, identify new parents and connections and adjust predictions accordingly. With the right data (more on that a little further down), a well-trained and calibrated AI algorithm could produce more rapid and more accurate predictions than human judgment alone. And with limited capacity of hospital and ventilators speed and accuracy matters. Based on the results of those projections, healthcare personnel can monitor those identified as being at higher risk. Mobile application predictions could quickly adjust as users self-report changes in symptoms (e.g., via a chatbot) and trigger an intervention (e.g., alerting a health care provider), if needed.

It may seem like it is already too late for an AI-based intervention, especially since the virus seems to have peaked in many parts of the world and cities are looking into relaxing social distancing protocols. It is worth noting, however, that the first wave of the Spanish flu was not the most deadly. In fact, it was the second wave of illness that was the real killer, and public health experts are worried about a similar scenario with COVID-19. Mitigating the impacts of a potential second wave is exactly why we should explore the potential of AI-based interventions.

The Devil is in the Details. All this sounds promising, but the devil is always in the details. AI algorithms may be highly capable and adaptive, but they rely on data, a lot of it very nuanced individual-level data. for example, in order to develop an AI algorithm to predict the need of a ventilator would require the following individual-level data domains: individual characteristics (e.g., age, gender), health history (e.g., medical diagnoses codes and treatments), current health symptoms (including COVID-19 testing, hospitalization and need of ventilator), and geographic information. The quality of that data could make or break the potential of a COVID-19 response. The data, as with the pandemic, are dispersed across cities, countries and continents. There is no single government entity that has jurisdiction over the data needed, nor the public trust to collect and develop fair AI algorithms.

While navigating data privacy laws are more than daunting, global philanthropy may be the most suitable for the task at hand and a number of data for social impact initiatives are already in place, some spearheaded by tech giants and others funded by 21st-century philanthropists. COVID-19 provides these initiatives the biggest stage on earth for a proof of concept and for unparallel social impact.