Applying AI to the COVID-19 outbreak – projects that matter via C3.ai DTI and Microsoft

I generally blow off “AI awards” and “exciting industry partnership announcements,” but I’m glad I made an exception for C3.ai’s event. The following real-world AI projects show how tech can be thoughtfully applied to vexing social problems.

Readers who frequent diginomica have figured it out: I’m not a fan of AI-handwaving, nor do I have much time for “AI industry consortiums” that seem to spend a lot of time patting themselves on the back for helping out.

Look, coronavirus won this round. I have every hope we will prevail against this virus, but all our wonderful predictive tech didn’t see this coming. The current state of “AI” was exposed much more than it effectively warned us.

That said, AI in the broadest practical sense, including robotic process automation, has certainly been of value helping companies respond to the pandemic. At times where human-to-human contact is problematic, robots and automation matter.

But not all AI fanfare is tedious. This summer, a launch event from the C3.ai Digital Transformation Institute got my attention. At the media roundtable (see YouTube replay), five award recipients shared potent examples of how they plan to apply AI technology. The projects show how teams are urgently applying the best tech they can get their hands on. I also got an obsessive question answered: How do you press on with AI when COVID-era data sets are lacking?

The award recipients were announced by the C3.ai Digital Transformation Institute (DTI) in June. All told, 26 projects were awarded $5.4 million to “Accelerate Artificial Intelligence Research for COVID-19 Mitigation Across Medicine, Urban Planning, and Public Policy.”

In March, C3.ai DTI, along with the University of Illinois at Urbana-Champaign, the University of California, Berkeley, and in partnership with Microsoft Corp., invited researchers “to take on the challenge of abating COVID-19 and advancing AI-based science and technologies for mitigating future pandemics.”

Applicants were narrowed to the final 26 proposals, which will pursue COVID-19 responses across medicine, urban planning, public policy, and computer science. The interdisciplinary finalists stand out: several of them will focus on the coronavirus impact on racial, economic, and healthcare disparities.

To complete their projects, finalists will have access to the C3 AI Suite platform, Microsoft Azure, and data resources such as the C3.ai COVID-19 Data Lake. Long-time enterprise types will undoubtedly recognize Tom Siebel, an enterprise CRM pioneer who is now Chairman and CEO of C3.ai.

C3.ai DTI COVID-19 project finalists – a few standouts
Some important projects received funding from this initiative. Here’s a sampling from the media call.

Anna Hotton, Epidemiologist and Research Assistant Professor in the Department of Medicine, University of Chicago, kicked things off, by sharing her team’s plan to investigate the impact on Chicago’s urban population. They will model the social determinants of health on COVID-19 transmission and mortality. As Hotton told us:

“The COVID-19 pandemic has highlighted drastic health inequities, and led to the recognition that social determinants of health factors such as poverty, housing instability, incarceration, and access to health care, drive disease transmission and mortality.

“Our research will apply methods from statistics and computer simulation modeling, to better understand how these factors impact COVID-19 outcomes, and evaluate intervention strategies. Through partnerships between the University of Chicago Argonne National Laboratory and local public health agencies, we will use local data to create a model representing the population of Chicago.”

“Using this platform, we will conduct large scale simulation experiments to understand how social inequities contribute to disease transmission, and evaluate interventions under different social distancing testing and risk reduction scenarios.”

COVID-19 Health Impacts

How can the tech help us here? Hotton:

“Computer modeling is ideal for this type of research, because it offers a way to predict future trends and compare interventions in a virtual environment, before deploying them in real life. We hope this work will provide insights to guide local policy for reducing health inequities.”

Next up: award recipient Sanmi Koyejo, Assistant Professor in the Department of Computer Science, University of Illinois at Urbana-Champaign. Koyejo’s team is working on securing federated learning for clinical informatics. Sound fancy? Perhaps, but the goals are eminently practical: manage medical resources across rural areas. As Koyejo said:

“Rural areas are among the most profoundly affected by COVID-19. And in these rural areas, patient outcomes may be worsened due to limited information at hospitals. Our work explores how clinical AI can help medical staff make better decisions about patient treatment by sharing information across large and small hospitals.”

I mentioned the lack of deep COVID-19 data sets. Koyejo’s project addresses a different aspect: data sets that are proprietary, or intentionally siloed.

“One major challenge is that the relevant data is often inaccessible, due to significant privacy and intellectual property concerns… My lab is developing software tools for privacy-preserving clinical AI, using data spread over several hospitals.”

Koyejo hopes that clinical AI can overcome data problems that hinder AI applications:

“For those of you who don’t know what clinical AI is – it’s a software tool designed to improve medical research, diagnosis and healthcare delivery, among others. Currently, such efforts require either using public data, learning from a single site – or complicated agreements to enable multi-site data access.”

His team will tackle this in a new way:

“Our approach avoids many of these issues by taking advantage of ubiquitous computing and the ability of hospitals to train AI models on site. So we’re currently developing tools for AI-supported COVID-19 diagnosis. Our longer-term vision is to enable breakthroughs in real clinical care, by using privacy-inherent computation of large scale medical data across the world.”

The economic impact of coronavirus has also hurt the affordable housing market.

Recipient Karen Chapple, Professor and Chair of City & Regional Planning, UC Berkeley, shared her team’s plans to confront this:

“This country has been experiencing a housing affordability crisis for years now. We’re about to experience another shock, as millions of unemployed workers are unable to make rent or mortgage payments. In this project, UC Berkeley’s urban displacement project, as well as the data-intensive Development Lab, will use the C3.ai COVID-19 Data Lake and cloud computing resources from C3.ai to predict and identify neighborhood residents living in precarious housing circumstances due to the COVID 19 economic and health crisis.”

COVID-19 Housing Impacts

Chapple’s team will mitigate training data needs, by pulling historical data from similar economic downtimes: “We’ll train supervised learning models on historic socio-economic data from the Great Recession. To predict neighborhood vulnerability today, the eviction lab at Princeton will also develop a scraping tool to track evictions in real time. Converting image captures of local eviction filings to text using natural language processing will show how post-COVID patterns of housing scarcity relate to established trends in inequality in the United States. This will allow cities and states to quickly put in place the policies that can stabilize communities into the future.”

My take

These are hardly the only recipients of merit (funded research projects list). Other projects tackled tracking and tracing, or how we can safely re-open public transport while monitoring transmission rates.
In this piece, I shared clues on how these teams will creatively tackle the problem of limited Coronavirus-era data. But I also put this question to the C3.ai DTI leadership team. My question:

Great projects – but I think the biggest obstacle here isn’t bias, but the lack of large/current/relevant data sets that AI tools need for effective modeling. I heard at least one example of using past historical data sets for housing disruption (e.g. past recessions), but overall I question how much use past data sets will provide. Curious how this will be tackled.

Tom Siebel responded:

“One of the projects very closely affiliated with C3.ai Digital Transformation is the COVID Data Lake. We’ve seen many data sources being published in the open domain, from CDC, from NIH from MITRE Corporation, from Johns Hopkins, Allen Institute, World Health Organization, Korea, China, Mexico….”

“And if you do a search on the C3.ai data lake, you’ll see that we have taken the union of all of these publicly available data sources, and aggregated that data into a unified federated image, and established all the connections between those data sets, whether it’s on co-morbidity, location, gender, or what have you.”

It’s now available globally.

“It is available to all the researchers, not only to the Digital Transformation Institute community, but to all the researchers in the world. And it’s available for free… I believe today, it is the world’s largest corpus of COVID related data. And we are expanding it at a very rapid rate.”

My point on bias raised discussions I won’t get into here. One key response from the panelists: it’s not enough to minimize bias from data sets. You must also rigorously look for evidence of bias in the outcomes of your projects.

Anyone hoping for a quick cure to COVID-19 will no doubt find the fledgling nature of these projects disappointing. But if you take the longer view, and think of COVID-19 as the first major pandemic of this century, but not the last, then the worth of these projects is clear.

One of the most disturbing aspects of Coronavirus is how it has heightened our divides, widened our inequalities, and inflicted a disproportionate blow to vulnerable populations.

These projects are encouraging precisely because they reject technology infatuation, in favor of community-based solutions. As an aside, this is also a great example of what you might call “the new marketing,” or just plain better marketing. Don’t spam us with how great your solutions are. Make a difference in the world with that tech, and finding buyers isn’t likely to be a problem. Neither is getting editorial attention.

This is one collection of enterprise award recipients I definitely want to track.

Read the full article here.