What if we had better models for prescribing solutions to catastrophic events when they occur, or even before they struck? As a society, we could address a severe food shortage or calculate the supplies needed to treat victims of an oncoming natural disaster. It would certainly improve the ability to sustain life as we know it.
Reflecting back on the ongoing and prolonged COVID-19 crisis, it’s clear that predictive modeling capabilities would have come in handy, and prescriptive analytics may have helped decision makers around the world. In retrospect, many in the public health and medical communities knew the world was due for an imminent infectious disease. Some even saw it coming in the fall of 2019, before the first official case of the disease was disclosed in late December in Wuhan, China.
But few saw the severity of the pneumonia of unknown etiology known as the novel coronavirus. One important reason: The models were not only limited in predicting the vast numbers of people who would eventually be stricken with or succumb to COVID-19, but they were also constrained in prescribing potential mitigation strategies to prevent the global pandemic that eventually ensued.
Reaching Model Limits
What’s wrong with the models? Traditional modeling approaches are inherently limited. Moreover, COVID-19’s epidemiology is complex, which has limited public health officials the world over in their efforts to stem the spread of this highly infectious disease. Where did we go wrong?
- For starters, the lack of accurate, near-real-time data made prediction, if not prescription, difficult.
- With a disease that knows no borders, it was difficult to predict how it would spread and mutate globally. Complicating matters were variances in cultures, public health systems and underlying economics. These entities interacted in unpredictable ways, confounding even the best data scientists.
- COVID-19’s spread wasn’t well-known at its inception and is still a matter of conjecture and constant reassessment. (Does it spread through casual contact with paper and product packaging, for instance?)
- Also vexing was its prevalence in the population, which was a challenge due to the large number of asymptomatic carriers and limited testing in the early phases.
- Lastly was the question of immunity: How many people are immune because they contracted the disease and have the antibodies? If they are fortunate enough to be able to ward off COVID-19 again, what was the strength of their immunity?
Taken together, these factors made it difficult to parameterize the models accurately.
Creating Better and More Prescriptive Models
To mitigate uncertainty and potential model fallibility, it’s crucial to have a high level of data quality (predicted on timeliness and precision). But better data alone is insufficient for building more prescriptive models. We also need smarter algorithms to fully take advantage of the relationships within the data, and to identify which levers have the greatest ability to derive more informed predictions and correlative responses.
Cognizant’s Evolutionary AITM offering is tailor-made for this purpose. It uses evolutionary computing and other AI techniques to create prescriptive models by learning and adapting to available data.
With Evolutionary AI, a schedule of suggested policies could be generated by trying various decision strategies against a forecast model built on past data on the spread of a disease. This approach minimizes assumptions regarding an as-yet unknown virus. For instance, common sense says that the proximity of people would cause infections to rise. Early on in COVID-19’s onset, the rate and manner of infections were not well understood, and different assumptions resulted in vastly different prescriptions. Evolutionary AI can assist in determining which prescriptions may work through trial and error with the forecast model.
Our data scientists applied our patented evolutionary computation tool, Learning Evolutionary AI Framework (LEAF), to the latest public data from Oxford University on COVID-19, which provides insights on infection rates and mitigation strategies that have been used the world over.
Using the prescriptive model, we can obtain a better idea of which mitigations may help the most to achieve specific objectives, and where more focus is needed to optimize the outcomes. And, these models improve over time as new data comes in. In one example, our model generated a creative strategy of opening and closing work in intervals in a unique combination with other recommended policies in order to achieve the desired balance of reducing the spread of the disease while minimizing economic impact. (For a more detailed technical description of the system and to see the demo, visit us at http://evolution.ml/esp/npi.)
Beyond Machine Learning and Infectious Disease
Our approach propels AI from providing incremental improvements to delivering creativity and the ability to prescribe strategies based on a range of objectives. Advanced Evolutionary AI techniques can improve predictions, drive impactful outcomes and facilitate effective experimentation by providing prescriptive guidance, across a variety of business scenarios.
While traditional models predict what might happen, they don’t offer counsel on what should be done about it. By using computational modeling and Evolutionary AI, we can leverage data to set goals, make decisions about those goals and recommend a solution to move forward.
The learnings in this case in using Evolutionary AI are being applied by decision makers in many industries. Take a business decision maker in retail. Using Evolutionary AI, she can set a range of goals, like revenue, margin and client satisfaction, and adjust the priority of these objectives to arrive at suggested strategies. This is a better way to make decisions, informed by data and advised by AI, to find the best options.
Subscribe to our newsletter and get expert insights straight to your inbox.×
SUBSCRIBE TO OUR NEWSLETTER✖
Thanks for your interest in Digitally Cognizant.
To complete the subscription process, please click the link in the email we just sent you.