How AI Can Potentially be Used in the Battle Against Pandemics
The News: The ongoing coronavirus COVID-19 pandemic has received saturation coverage not only in the mass media, but in specialized scientific and research channels. In the public interest, many news outlets have put their COVID-19 coverage outside their paywalls.
One of the most useful of these free news sources for the technical community has been the MIT Technology Review. If interested, here is a link to their free coverage of the coronavirus COVID-19 outbreak. One of the recent articles on this site is a comprehensive discussion of research into how AI can potentially be used in the battle against pandemics, and how AI is, artificial artificial intelligence (AI) is helping to stem the outbreak’s tide. Just as important, the article highlights the limitations of current AI tools, approaches, and implementations in dealing with the current pandemic.
How AI Can Potentially be Used in the Battle Against Pandemics
Analyst Take: AI is playing many roles in the world’s battle against the COVID-19 pandemic. But AI is certainly not a panacea and its role in helping stem the tide of infections and mortality should not be overstated.
Pandemics have afflicted the human race for as long as our species has walked the Earth. As sure as the sun rises every morning and sets every evening, these devastating viral outbreaks will return and wreak havoc.
But that doesn’t mean the human race is defenseless in the battle against contagious disease. Indeed, we have added a powerful new weapon — AI — in this struggle, and it’s proving its worth in the present coronavirus COVID-19 pandemic. While there are no doubt many benefits AI can provide, AI also has its limits, which is what I wanted to discuss here. With that in mind, let’s go a little deeper.
AI Can Be Used as an Early Warning System
One immediate benefit of AI is that it can be used as an early warning system. AI enables epidemiologists both to spot emerging outbreaks and to predict how they might spread from region to region and perhaps even from one demographic cohort to others.
For example, vendor BlueDot uses an AI-based solution to monitor outbreaks of infectious diseases around the world. In late December 2019, more than a week before the World Health Organization officially flagged the COVID-19 outbreak, BlueDot alerted governments, hospitals, and businesses to an unusual spike in pneumonia cases in Wuhan, China. The outbreak was also identified early by AI-based tools HealthMap (at Boston Children’s Hospital) and Metabiota in San Francisco.
However, AI-automated early warning systems may find themselves racing against online social channels for the distinction of being first to detect a new outbreak in the offing. The MIT Technology Review article that I cited earlier in this article reported that human teams spotted the current Coronavirus/COVID-19 outbreak on the same day as these AI-powered research tools. That’s not surprising, considering outbreaks tend to have highly localized initial stages, in which at least one close-up observer raises an alarm. We are seeing that play out today, as physicians and healthcare workers the world overtake to social media channels, private or otherwise, to share concerns, thoughts, observations, and we are also seeing that as citizens of affected areas share their stories. Social media channels are powerful conduits of information
Bottom line, AI can be used as an early warning system, but let’s not overlook, or underestimate in any way, the power of human to human contact.
The Potential for AI as Infection-Path Predictor
I think there is great potential for AI being used as an infection-path predictor, predicting how COVID-19 or any other outbreak is likely to spread, and, just as importantly, how tactics such as “social distancing” might curtail or even lessen its severity.
In theory, it might be possible to run unsupervised learning algorithms that simulate all possible evolution paths, experiment digitally with how well potential vaccines perform in each scenario, and even determine whether and how the viruses develop resistance through mutations. But this approach is a bit far-fetched to offer near-term hope in the current pandemic. That’s due to the need for rapid advances in the science, modeling, and computing capabilities that would be needed to pull it off.
Another practical obstacle is the need to find sufficient amounts of behavioral, social, clinical, airline, and other data sources of sufficient quality to build and train accurate enough machine-learning models of an outbreak’s likely evolution path. The companies that detected the current COVID-19 outbreak were using NLP algorithms to look for relevant reports coming from news outlets and official health care channels in different languages around the world. However, especially in fast moving viral outbreaks, those sources may be too vague, inconsistent, and biased by political, cultural, and other factors to offer the proverbial “single version of the truth.”
In addition, the chances of pooling this data from diverse global sources in the middle of a fast-moving pandemic are not great, and the difficulties of harmonizing and cleansing it all are so great that the effort would take longer than the pandemic itself to come to fruition.
It’s also next to impossible to find reliable data on “social distancing” variables of a behavioral nature, such as the incidence of handshaking, the frequency with which people wear surgical masks and gloves in public, the average size of public gatherings, and so on. As one of the researchers in the MIT article states: “We…don’t really know what behaviors people are adopting—who is working from home, who is self-quarantining, who is or isn’t washing hands—or what effect it might be having. If you want to predict what’s going to happen next, you need an accurate picture of what’s happening right now.”
Where behavioral factors come in, there’s the need for high degree of predictive precision to drive proactive alerting of the relevant countries, regions, and authorities. The efficacy of such tactics as quarantines, school closures, and vaccination of at-risk demographics depends on having early enough intelligence so that outbreaks can be squelched before they spin out of control.
But having early warning is not enough in a fast-moving public emergency. All the AI-driven insights in the world are powerless in a situation such as what we’re facing in the United States and a federal government that has been slow to respond, less than transparent, and difficult to trust to have the best interests of the public first and foremost.
No matter how powerful its tools and accurate its data, AI can’t immunize us against a political establishment that refuses to take effective timely action.
Using AI as Diagnostic Instrument
AI is being used to examine medical images for early signs of many diseases that human doctors might miss. In recent weeks, preprint research papers have begun to appear online in which machine learning has been shown to diagnose COVID-19 from CT scans of lung tissue.
However, this approach might not be effective as an early diagnostic, considering that physical signs of the disease may show up in scans only after infection, making it not very useful as an early diagnostic. Also, the paucity of training data on a disease so new makes it difficult to assess the predictive accuracy of the approaches in the research literature, especially where it concerns identify subtle patterns in medical images.
Techniques such as few-shot learning and transfer learning might be used to train AI models to look for COVID-19 in the absence of much training data, but those approaches remain largely unproven for the current outbreak.
Exploring AI as a Research Discovery Tool
AI can accelerate access to a vast, constantly changing corpus of research literature, data, and analytical tools pertaining to outbreaks, their spread, and effective treatments.
This recent MIT Technology Review article discusses a new open database—known as CORD-19 (COVID-19 Open Research Dataset)—which contains over 29,000 coronavirus research papers. Researchers from several organizations released the Covid-19 Open Research Dataset, which includes papers from peer-reviewed journals as well as preprints from websites such as bioRxiv and medRxiv. The research covers SARS-CoV-2 (the scientific name for the coronavirus), Covid-19 (the scientific name for the disease), and the coronavirus group. It was compiled under the request of the White House Office of Science and Technology Policy (OSTP).
The database, now available on AI2’s Semantic Scholar website, leverages AI to speed searches through academic literature. It incorporates natural-language processing models such as ELMo and BERT to map out the similarities between papers and create personalized feeds based on researchers’ interests. The OSTP also launched an open call for AI researchers to develop new techniques for text and data mining that will help the medical community comb through the mass of information faster.
Though no one can dispute the value of this database or the need for more powerful AI tools to search it, it’s clear that most insights that researchers might gain from them over the next 4-8 weeks will apply more to the next pandemic—which could be decades in the future—but probably won’t come in time to be much use in combatting the current outbreak. And most research studies included in the database now are probably derived from studying previous outbreaks, limiting their usefulness in devising strategies for dealing with today’s unfolding emergency.
AI as Treatment Tool
AI can facilitate medical researchers’ investigations into pharmaceutical and other treatments to arrest COVID-19’s progress and possibly find a cure.
Though there’s no proven treatment yet, the World Health Organization has identified more than 70 drugs or “therapeutic” combinations thereof that are potentially worth trying. It’s highly likely that AI-based tools such as this experimental DeepMind offering are being used to explore how protein structures and interactions how the virus functions and how, conceivably, it can be neutralized. In addition, generative design algorithms use AI to produce millions of candidate biological or molecular structures and sift through them to highlight those that are worth looking at more closely for their possible efficacy.
However, this approach may be too little too late, because it can take months before a promising candidate emerges from the pack.
The Takeaway on the Potential of AI in Battling Pandemics
Clearly, AI is stepping up to the many challenges of dealing with current coronavirus COVID-19 outbreak. It has proven to be an invaluable tool for detecting the pandemic’s onset; predicting when, where, how, and at what speed it’s likely to spread and evolve; diagnosis its incidence and severity; and discovering effective cures and treatments.
However, AI is largely standing on the sidelines when it comes to helping people, groups, businesses, and government agencies to cope with the outbreak. Though simulation tools like this, developed by the Washington Post allude to the possibility of “flattening the curve” of the pandemic’s spread through “social distancing,” the underlying technology doesn’t seem amenable to packaging into a personal digital assistant of the sort that would be needed to help each of us avoid exposing ourselves to the virus in the normal course of living our lives.
Even if it were possible to have our own personal recommenders that steer us away from behaviors that might expose us to coronavirus COVID-19, many people would find these tools so intrusive and nagging as to be practically unusable.
Futurum Research provides industry research and analysis. These columns are for educational purposes only and should not be considered in any way investment advice.