Surveillance, AI, and saving lives top agenda at coronavirus conference
The Stanford Institute of Human-Centered AI (HAI) hosted a conference to discuss applications of AI that governments, technologists, and public health officials are using to save human lives during the global coronavirus pandemic. Based on current mitigation efforts, the U.S. — which has more cases than any other country on Earth — could see 100,000 to 240,000 deaths over the coming months, according to a White House estimate released earlier this week.
Also prevalent on the minds of experts throughout the day-long digital event Wednesday was the level of surveillance that’s necessary, or tolerable, for tracking people with confirmed cases of COVID-19.
“I think on the technology side we can have pre-thought, emergency assignments that enable the administration in any pandemic situation; when the legislature agrees, they can have infrastructures such as those at tech companies at their disposal, because if a governor is allowed to legally command Sheraton to share its hotel beds as hospitals, I believe the governor should be allowed to command a tech company to share their tracking infrastructure for public health for a limited period of time,” said Dr. Nigam Shah, whose work at Stanford focuses on machine learning models that predict things like available hospital beds.
Stanford University professor and Center for Innovation in Global Health director Dr. Michele Barry gave a snapshot of how countries around the world have responded to COVID-19, sometimes in ways that may be considered invasive in the United States.
“I think what we’re all worried about in some of the more authoritarian countries … is that it’s often hard to unroll strict laws that are made in the time of public health emergencies. It’s very easy to roll them out, but we’re all looking with bated breath about what’s going to happen in the future,” Barry said.
Topics on the agenda at the conference ranged from AI’s role in the hunt for a vaccine to misinformation to bioweapons to applying AI to millions of tweets to understand the psychological toll of social distancing and rampant unemployment in the past month. Currently, the majority of U.S. citizens, and more than one in five people around the world, are practicing social distancing.
Public health and government officials are actively considering surveillance, quick tech builds, and ways to accelerate research or gain predictive insights. Last week, WHO executive director Dr. Michael Ryan said surveillance is part of what’s required for life to return to normal in a world without a vaccine. Countries hit hard by coronavirus in the past few months have adopted smartphone apps or other means of tracking the movement of people diagnosed with COVID-19, such as wearable wristbands.
Stanford HAI codirector Dr. Fei-Fei Li described one solution for tracking. It’s an AI-powered, in-home COVID-19 monitoring and tracking system designed for seniors who live alone, and for the people who care for them but who have to minimize contact to reduce the risk of infection. The system is built with cameras and smart sensors that track biometric information like temperature, telltale body movements, and sleep patterns. It sends the data to a secure central server, where AI models search for clinically relevant patterns. Then, the system would alert caregivers to any pertinent results — perhaps via an app.
Tina White, a Ph.D. candidate at Stanford University, said oppressive surveillance she read about in February concerning China’s coronavirus app inspired her to action. The COVID Watch app for iOS and Android smartphones uses the Bluetooth protocol for measuring close contact between individuals. Privacy advocates prefer Bluetooth because it’s a decentralized way to share information locally on smartphone devices and could be more effective than GPS at tracking movement indoors, where people are more likely to transmit the virus. When a person receives a positive test, they share their contact number in the app, and anyone they came in close proximity to receives a notification.
The COVID Watch app open-sources its code to give people in other countries the choice to make their own version. Researchers from eight European countries also recently open-sourced PEPP-PT, the Pan-European Privacy-Preserving Proximity Tracing, which also uses Bluetooth.
“Our approach was designed so you don’t have to collect identifying information about an individual in order to use it. So our data set is kind of special because there’s no actual identifying information, it’s just these anonymous contact event numbers which are then deleted after a couple weeks and are only interpretable by the individual phones,” White said. “So it’s only an intervention and not something that generates a data set, which is kind of what we were aiming for — something that’s so private that we can’t even interpret it ourselves. It’s only interpretable by the phones that were locally involved in the contact events.”
The COVID Watch team is working with the makers of TraceTogether, who on March 20 in Singapore launched the first national app for using Bluetooth to track proximity to confirmed COVID-19 cases. The team is also working with creators of MIT’s PrivateKit: Safe Paths and Community Epidemiology in Action (CoEpi).
Stanford professor and Center for Policy, Outcomes and Prevention director Jason Wang held up Taiwan as proof that you can track COVID-19 cases without autocracy. He said the key to some of the lowest mortality and infection rates in the world is low-tech solutions like preparedness and understanding when evidence on the ground demands quick, decisive action. Officials in Taiwan are using heavy fines and phone tracking, as well as the Communicable Disease Controls Act, a law approved by lawmakers in 2003 after the breakout of SARS. In Taiwan, people can use maps supplied by government officials to see the local face mask supply. Infrared scanners in buildings, and text messages sent to people in the vicinity of confirmed cases, are also in use.
Efforts by academics and researchers scrambling to redesign or repurpose existing solutions to fight COVID-19 was also a major topic of discussion.
Ryan Tibshirani leads the Delphi Group at Carnegie Mellon University, which is one of two CDC centers for excellence in the U.S. that create models for tracking the flu. Tibshirani said his group produced some of the most accurate flu prediction models in previous CDC Flu simulation competitions, but he added that this year, the group is adapting. Delphi Group plans to make coronavirus-tracking models at the county level in the United States available in the coming weeks.
“There are models that take things like [influenza-like illness] (ILI) prevalence that can then predict hospitalizations, predict critical hospitalizations, predict deaths, and these models are very well established for influenza. The question is, if we can train them fast enough … by ‘fast’ I mean, do we have enough training data to make useful forecasts for coronavirus?” he said.
Symptom surveys or screeners are not the same thing as a positive confirmation, but Tibshirani said symptom tracking can still be very useful. Determining whether a person has an ILI is important information right now since the majority of people with coronavirus show signs of fever, sore throat, and cough.
More surveillance that gathers ILI data, he said, could help fight the pandemic.
“A national or international surveillance system at this point which is capturing that would be extremely useful,” he said. “Data privacy is of course one of the foremost issues people discuss today, and this is an unfortunate time for this to have been such a big issue, because there are real issues of privacy surrounding sharing of medical data. That being said, what we’ve seen in our group — and I think this is true across, probably, anybody who’s working on COVID modeling and forecasting — is that there’s been more movement and more accomplished in the last two weeks than I’ve seen in years in terms of data sharing.”
Following a series of presentations on apps for tracking COVID-19, a moderator said a common question among participants was why data from big tech companies wasn’t being used to power more solutions.
The White House has had conversations with tech companies like Facebook and Google about data sharing, according to the Washington Post.
The Harvard University and Boston Children’s Hospital epidemiologist John Brownstein described numerous projects for real-time COVID-19 mapping that range from conversational bots that ask questions about COVID-19 symptoms to social media data mining. Research published in Science last month used location and movement data from Baidu to prove the effectiveness of social distancing. Boston Children’s Hospital is working with the CDC to understand the impact of social distancing and areas. There’s also Covid Near You, an adaptation of Flu Near You that asks people how they’re feeling every day to map people exhibiting COVID-19 symptoms. Since its launch last week, COVID Near You has been used more than 250,000 times, and was built with engineering time donated by Google, Facebook, and Airbnb.
Cooperation from tech companies is “great as I’ve ever seen at any point in time,” Brownstein said, but he added that they remain under the same privacy and data pressure they’ve faced in the past several years.
Brownstein called the idea of a national surveillance tracking app unlikely without federal government action.
“It would be hard to do some of the same types of app-based approaches that we saw in [South] Korea, Israel, and other places without really major federal support, and I think that’s where there’s potentially a little bit of a disconnect between what’s happening at the federal government level and what they’re trying to push in terms of an app-based contact tracing approach,” Brownstein said.
Tibshirani said he’s also working with Google on an ILI survey that’s collected about a million responses in the past week or so from people about how they’re feeling.
“I think it requires a giant like Google or another kind of really big tech company, and I think there are many others of these in the works to adopt one survey or mode of surveillance and just make it available to everybody. I think that would be really beneficial to the entire community and so we’re working on making that happen,” he said.
Researchers focused on epidemiological models described a sense of urgency to tweak and adapt models before communities in the southern hemisphere see a rise in cases as they enter fall and winter seasons. Scholars with opinions on all sides of the surveillance spectrum also described some optimism about cooperation happening between academic journals, businesses, and researchers in response to COVID-19.