Experts discuss AI’s potential to combat COVID-19, noting caveats

April 2, 2020, 2:08 p.m.

Experts in the fields of data science, international security, medicine, biotech and the humanities discussed the potential for artificial intelligence (AI) to tackle the challenges in governance, medical research and security posed by the COVID-19 pandemic at a virtual conference on Wednesday.

Discussion ranged from applications of machine learning to epidemiology and vaccine development, to government surveillance powered by information technology. Hosted by the Stanford Institute for Human-Centered Artificial Intelligence (HAI), the conference was marked by its ambition to bring in a broad spectrum of expertise to tackle the pandemic. 

“It’s a fully interdisciplinary approach so that the ethical, social, economic and psychological impacts and in orientation to thinking about artificial intelligence,” said Rob Reich M.A., Ph.D. ’98, professor of political science and a faculty associate director of HAI, who co-moderated the conference with Russ Altman, Ph.D., M.D. ’90, professor of bioengineering and a faculty associate director of HAI.

Around 3,000 viewers watched the conference, which was live-streamed on YouTube, asking more than 300 questions on a Q&A board by the conference’s end.

Session I: Overview and Policy Response

Rep. Ami Bera (D-Calif.), former chief medical officer of Sacramento County, began the conference by emphasizing the importance of a coordinated policy response to the COVID-19 pandemic. So far 38 states and several counties and cities have issued their own stay-at-home orders, but a federal order has yet to be put in place.

“We’ve been pushing the administration to issue that order as opposed to leaving it to the 50 states,” Bera said.

Asked when shelter-in-place orders could be rolled back, Bera sought to emphasize the need for serologic testing, which examines antibodies in a patient’s bloodstream to assess past exposure to the virus and potential immunity.

“Serologic testing will tell us how many people in a community have already been infected by covid-19, recover, and have some level of immunity,” Bera said. 

Researchers from Stanford Medicine, along with several private companies are launching serologic tests, but it is unclear when they will come into widespread usage.

The importance of accurate data and realistic models was the focus of a talk by Nigam Shah, a professor of biomedical informatics. 

“Without good data, all these models will have exponentially bad outputs,” Shah said.

Shah also described tension between U.S. privacy law and obtaining access to data in a timely fashion.

“Getting the right data has been a challenge and I will say publicly that U.S. law does not help,” Shah said. To alleviate the issue, Shah suggested, “We should put in place networks that are already pre-authorized to share data at very short notice.”

Seema Yasmin, director of Stanford Health Communication Initiative and a clinical assistant professor of medicine, explained the difficulty of communicating the results of data analysis to the public.

“I’ve reported a lot on models, because models make great headlines,” Yasmin said. “I worry about how the public is fed this information without the context of how models are built and the number of assumptions that are put into them.” 

Nonetheless, “journalism is part of the public health ecosystem,” Yasmin said, calling for data literacy in the public. 

Session II: Social Impacts & Biosecurity

Echoing Yasmin’s call for accountable journalism, Kate Starbird ‘97, an associate professor of computer science at the University of Washington, urged governments and media outlets to build trust with their audience to combat an “infodemic” caused by an “abundance of information.”

“COVID-19 is a perfect storm,” Starbird said. “Crisis communicators should build policy and recommendations using the best science available at the time.”

Experts also cautioned against the damage done by abuse of state surveillance and propaganda to civil rights.

Renée DiResta, technical research manager at the Stanford Internet Observatory, said the COVID-19 crisis made the Chinese Communist Party’s propaganda much more visible in Western social network sites. Calling the pandemic a “PR challenge” for the party, DiResta said English-language Chinese state media, whose audience is growing in the Western internet, has been shaping worldwide perception of China and the government’s response to COVID-19.

Increased access and attention to biotech may pose additional risks to global security. The current crisis will transform the “security landscape” in international society, according to Megan Palmer, a senior research scholar at the Stanford Center for International Security and Cooperation (CISAC). Biotech advances, for instance, may enable the reconstruction of viruses, which can be strategically deployed to undermine regional security, Palmer said.

Session III: Tracking the Epidemic

Jason Wang, associate professor of pediatrics, praised Taiwan’s use of data analytics in mitigating COVID-19. Taiwan has a centralized national health command center that pulls data from various healthcare systems as well as the media. “Taiwan immediately started checking incoming flight passengers for symptoms” and registering each individual in their command center, according to Wang. 

The government also deployed a number of other tech solutions including text messages to individuals who have visited at-risk locations, an app for mask supplies in local stores and fever detection during entry to many buildings, according to Wang.

But this centralized approach of extensive data collection may not be possible in the U.S due to privacy concerns, according to John Brownstein, a pediatrics professor at Harvard Medical School. 

“It would be hard to do the same app-based intervention as Korea and Israel and other places without really major federal support,” Brownstein said.

Researchers are working on solutions for accurate data collection without privacy infringement.

Tina White, a mechanical engineering Ph.D. student, described a mobile app she and other researchers developed to use crowdsourced data while protecting individual privacy. The app does not “need to collect any identifying information about a person” because it uses local Bluetooth connections  — as opposed to GPS — to determine which individuals a person has come into contact with, according to White. 

Reich raised a concern about the inability of standard data collection techniques to track asymptomatic individuals with COVID-19 who are unlikely to be tested.

Lucy Li, a data scientist at the Chan Zuckerberg Initiative’s BioHub, has worked on a method to alleviate this problem using information from viral mutations. 

“If we know how quickly the virus mutates, we can measure how many missing transmissions occurred between the tested cases,” Li said.

Tracking the epidemic also requires testing at scale, a problem that has plagued the US since the inception of the virus.

Xavier Amatriain, cofounder of tech startup Curai, sought to stress that scaling healthcare requires the use of tools like AI and automation.

He said his team has pivoted towards “incorporating coronavirus features” in their AI-powered diagnostic tools. Curai has now developed the first “machine-learning COVID-aware diagnostics model” using expert-curated training data, according to Amatriain.

Session IV: Treatments & Vaccines

“We are in a war, and we need wartime strategies,” said Ron Li, a clinical assistant professor of medicine, hospital medicine and biomedical informatics, who introduced AI-driven solutions meant to help combat the pandemic. His team’s machine-learning models can triage patients who require more intensive medical care and streamline workflows of the health delivery system, Ron Li said.

Research introduced by Kristen Beck, a lead bioinformatician at IBM, uses AI to study how bacteria mutates. By automating the process of tracking, associating, and sharing the data, the researchers were able to “accelerate the study of microbial life at scale,” Beck said.

But experts also cautioned against overhyping the capability of technologies like AI and machine learning. Responding to an audience question asking when her project on AI-driven elderly care can be implemented, Fei-Fei Li acknowledged that her team’s research primarily serves academic purposes.

However, biotech research may be able to ramp up into entrepreneurial opportunities. With the right development of technology, the machine-learning community can devise technical solutions to address specific demands from society, Fei-Fei Li said.

In a previous version of this article, the photo caption named the individuals depicted in an incorrect order. Also, a previous version of this article incorrectly attributed a quote from Rob Reich to John Etchemendy, and incorrectly identified Etchemendy instead of Reich as the moderator of the session. The article has also been corrected to reflect that Reich is a faculty associate director of HAI, not a faculty assistant director. The Daily regrets these errors.

This article has also been updated to reflect that Reich moderated the conference with Russ Altman.

Contact Arjun Ramani at aramani3 ‘at’ and Won-Gi Jung at jwongi ‘at’ ‘at’

Arjun Ramani ’21 is a senior staff writer for the Data section and was The Daily’s Data Director for Volumes 257+258. He hails from the Hoosier Heartland of West Lafayette, Indiana and is studying economics and computer science. Contact him at aramani ‘at’

Login or create an account