We Helped Greece Build an AI System to Make Covid-19 Testing More Efficient. Here's What We Learned.
Two data scientists built a machine learning system named Eva to help Greece safely reopen to nearly 80,000 tourists a day. The system, known as Eva, is nearly twice as efficient at detecting cases as random testing and it can predict spikes in other countries ten days before they show up in official case counts.
Despite government warnings, millions of Americans traveled over the holidays. Schools, universities, and workplaces face difficult decisions as people return in the midst of a coronavirus surge, potentially bringing Covid-19 with them. While vaccines are already being distributed, there’s a long road ahead until we achieve widespread vaccination and herd immunity.
What if there was a better way to determine who is most at risk for Covid-19 and who should be tested and quarantined? We’re data scientists at USC Marshall School of Business and this summer we built an artificial intelligence system to help Greece safely reopen its borders to over 80,000 tourists a day. The system, known as Eva, determined which foreign visitors to admit and who to target for testing. As schools, businesses, and tourist destinations navigate a new wave of coronavirus cases amid the busy holiday season, they should consider how to use data and artificial intelligence to deploy their testing resources efficiently and reduce the spread of the virus.
Tourism is central to Greece’s economy. This spring, the country faced a dilemma: It needed to reopen its borders for the prime summer vacation season but couldn’t test every incoming traveler. We created Eva to safely manage the flow of visitors. Every person entering Greece had to submit basic demographic information along with details about what countries they’d been in recently. Using a machine learning algorithm to predict risk, Eva selected which travelers to test on arrival at the border. After quarantining for 48 hours while awaiting the results, anyone who was positive followed government isolation protocols and contact tracing. Everyone else was free to continue on their visit.
An exercise in efficient testing
This system was a win all around. It allowed the country to reopen to travelers while reducing the risk that anyone with Covid-19 entered. By more accurately targeting testing, it used fewer resources to detect each infection. Random surveillance testing would require almost twice as many tests as Eva to catch the same number of infected travelers at the border. At the peak of the summer, that would have translated to almost 14,400 tests per day – close to the total testing capacity of the entire country. In contrast, Eva used about 8,000 tests to catch the same number of cases, freeing up valuable testing resources for local residents while still filtering out infected visitors.
Eva also provided better data for Greece to use to continually update its travel policies. Because it tests people before they are symptomatic (people who are actively showing signs of illness rarely make it onto the plane), our system produced more timely prevalence rates than confirmed case counts released by national governments. We were able to predict spikes in places like Malta and Spain ten days before they showed up in official statistics – time that was crucial in preventing the flow of infected travelers. Greece used this data to decide when to bar visitors from high-risk countries or require proof of a negative PCR test before they even arrived at the border.
Eva helped Greece reopen its borders this summer without a major spike in coronavirus cases. The country, like much of Europe and the U.S., is currently experiencing a surge and requiring all visitors to have a negative test result before arrival. But during the peak tourist season this summer Covid-19 numbers were relatively flat despite welcoming nearly 80,000 visitors each day. Eva enabled Greece to catch more infections at the border (via smarter deployment of testing) and prevent more infected people from ever arriving (through smarter travel policies). It also helped restore public faith in the ability to reopen the economy while guarding public health.
Offering a middle ground between complete lockdowns and zero restrictions
A smart system like Eva breaks through the false binary that countries must either stay locked down or reopen completely. Many European nations took a much less precise approach than Greece to reopening their borders, allowing visitors from select countries with low official case counts but not requiring any testing when they arrived. Other places, like New Zealand, Canada, and Hawaii, remained closed almost completely, decimating local tourism economies.
In addition to travel and tourism, we believe AI-driven systems like Eva can help schools, workplaces, local governments, and other institutions with limited resources better manage their Covid-19 restrictions and testing policies. If a university, for instance, can only test 1,000 people a week, who should they test? A smart system like Eva is much more efficient than random testing and can surface high-risk characteristics and behaviors. For example, are students in classes with in-person labs more likely to test positive? Is there a spike among athletes now that winter has arrived and practices moved indoors? By tracking what’s happening in real time, the system can help institutions tweak their policies for activities or groups identified as high-risk.
Two takeaways on tracking and target testing
We have two lessons for governments and institutions building data systems to track Covid-19 and target testing. First, success is about much more than just a stellar algorithm. It requires effective operational systems governing where and when testing takes place, how quickly results are available, and how data and information are communicated. The most sophisticated AI is worth little without effective operations in place on the ground.
The other big issue with any data-tracking system is, of course, privacy. Institutions should be very careful about what data they collect, how they protect it, and who has access to it. We built Eva to comply with the EU’s strict GDPR privacy laws; in the U.S., HIPAA and state regulations, like California’s CCPA, would apply. Eva only collects information that is essential for its calculations, avoiding data points like social security numbers that aren’t useful for the algorithm. No names or identifying details are fed into the algorithm itself; the government maintains names separately for use in contact tracing when necessary.
Even though we’ve begun to distribute vaccines, testing and prevention measures to keep Covid-19 in check will be with us for a long time to come. Smart, data-driven systems like Eva are more efficient and less burdensome than random testing, and much safer than not testing at all. This is the first pandemic where humanity has had artificial intelligence and other advanced technology in its arsenal. By using it wisely, we can chart a middle ground that balances safety and economic well-being on the road back to normalcy.
Entrepreneur Leadership Network Contributor