Apple and Google’s common coronavirus contact tracing solution for smartphones has continued to attract a lot of attention and debate over the past week, and understandably so. It’s an unprecedented partnership between the world’s dominant smartphone operating system makers, but people are worried about privacy and the notion that tracking tools deployed in the name of coronavirus will outlive the crisis. Debate over Apple and Google’s contact tracing solution seems to have opened up an old argument between people who see a tech solution for every problem and those who say tech can’t solve all our problems, and can even cause new ones. These debates certainly carry over to the kind of AI being deployed right now and the important question of when a company should ship or shelve a coronavirus solution.
A lot of AI solutions are being rushed out in an attempt to save lives and speed up the day when we’ll return to something resembling normal life, and you’ve been able to read about many of these in our coverage. But these AI systems aren’t all winners.
In February, a robotics company sent its service robot to Times Square in New York so passersby could answer questions to help them determine whether they had COVID-19, but the experience relied on a shared touchscreen. Given how bad the pandemic is in New York City today, that seems pretty irresponsible.
AI is also being used in more productive ways to understand how coronavirus and social isolation are impacting people’s psychological health and well-being. Some AI, like a flu model from Delphi Group at Carnegie Mellon University, is being repurposed to forecast coronavirus models for the United States. An MIT model out this week suggests the effectiveness of social distancing and the potential for an “explosion” in cases if those measures were relaxed today.
Of course, also present in this environment is opportunism from startups anxious to remain relevant, raise funding, or attract publicity at a time when much economic activity is at a virtual standstill.
Innovation in a crisis can lead to outcomes that better human lives but can also distract from priorities like widespread testing and protecting health care workers and the most vulnerable among us.
Some initiatives seem almost outlandish in their ambition, or, however promising, unlikely to be taken up in a timely manner. One project straddling that chasm is Cough Against COVID.
Cough Against COVID, which launched this week, is a project by Mumbai-based nonprofit Wadhwani AI in partnership with the Bill and Melinda Gates Foundation and Stanford University. The people behind it collect audio recordings of coughs by people who have confirmed cases of COVID-19. Online submissions for people quarantined at home must be accompanied by a photo of a diagnosis from a doctor. To spur additional research, all data sets collected will be made available in an anonymized, open-access data set. In addition to a data collection website, Johns Hopkins University doctors are collecting data directly from patients at a hospital in India.
The hope is that voice recording data can power AI for screening apps being made by public health officials and create an additional diagnosis signal that doesn’t exist today. The project was inspired in part by Global Good’s work around tuberculosis identification using sound in Madagascar, and the work of Massachusetts’ FluSense, which uses cough sounds for health forecasting.
Jigar Doshi is a senior researcher at Wadhwani AI. Before moving back to India three months ago, Doshi headed machine learning efforts at computer vision startup CrowdAI, a company that worked with Facebook AI Research on multiple projects to assess damage after a natural disaster and help governmental or humanitarian organizations determine need.
Doshi admits he doesn’t know if Cough Against COVID will work, or how much data they’ll need to make a robust and accurate model, because COVID-19 is a novel disease. But an additional way to detect it could be helpful in parts of the world where hospitals, health professionals, or diagnostic testing are in short supply.
“It’s sort of a moonshot idea where it may work, and if it works it would really help. We don’t know if it will work, but the only way to find out at this point is to collect the data [and] do our best modeling,” he told VentureBeat. “This is all centered on limited testing ability, especially as we move away from Western countries.”
When we asked what he’d say to people who dismiss Cough Against COVID as a kind of techno-solutionism, Doshi said “This is one of those things where generally some degree of skepticism toward technical people from their high horse, castle, privilege — whatever you want to call it — coming down to help, generally is good.”
Doshi continued to highlight the importance of working with medical professionals to keep things grounded and said the project is only asking COVID-19 patients with mild cases for five minutes of their time to find out if it’s possible.
Charles Onu founded Ubenwa, a company using AI to detect birth asphyxia in the sound of crying newborn babies. He sees a lot of merit in work like Cough Against COVID and called it a valid and intriguing venture for diagnosing a respiratory disease. Onu said he sees promise from research published in June 2019 that demonstrated the ability to recognize and distinguish between the sounds of respiratory diseases like bronchitis, asthma, and pneumonia with 80-90% accuracy.
With Ubenwa clinical trials in hospitals on hold due to the crisis, Onu, who is based in Montreal, said the company is in early talks with Canadian government officials to further COVID-19 diagnostics with sound. Onu said he generally agrees with the idea of continuing experimental efforts, particularly as a way to help in areas where testing and resources are limited.
“One side is making it possible in Canada or the U.S., but also in my village in Nigeria and many places where they may have to go on a very long trip to take a test, so this could definitely close that gap,” Onu said.
Like Doshi, Onu thinks companies and developers deploying AI solutions right now should discuss their approach with medical experts.
“I really hope that at the end of the day, people do whatever [they] like, but at the deploying, you have a gating mechanism with the public health system to make sure that they’re not spitting out fancy things that don’t solve the problem,” he told VentureBeat.
These are unprecedented times, and what’s needed from one moment to the next can change. For example, a month ago public health officials told people they don’t need to wear face masks unless they’re sick or taking care of someone who’s sick. Now the Centers for Disease Control (CDC) and others suggest people wear them whenever they’re outdoors or near others.
So when should you ship or shelve a coronavirus-related AI solution? Some of the principles seem similar to those for ethics: Speak with stakeholders and consider societal well-being and potential impact. The decision should also depend on whether the tech can deliver immediate results, but what’s considered best might change depending on testing and health care resources.
Some solutions and the companies peddling them, as a cryptographer advising the U.K. government about contact tracing apps put it, may serve best by just staying out of the way.
Thanks for reading,
Senior AI Staff Writer