By Theodoros Evgeniou, David R. Hardoon and Anton Ovchinnikov
SINGAPORE — Data driven firms, from “big tech” to financial services and even healthcare, make personalised recommendations using predictions from models that uses machine learning and artificial intelligence trained on customer data these firms amassed.
We use the latest AI tools for recommending movies and showing ads, but not for saving lives.
The Covid-19 outbreaks follow a pattern: first, exponential growth, then blanket lock-down to “flatten the curve” and finally, the “how to reopen” question once the famous curves peak. This pattern replicated across countries in almost identical ways.
As a famous quote, often attributed to Einstein, says, “Insanity is doing the same thing over and over again and expecting different results”. The repetition of Covid-19 patterns indicates structural weaknesses in our global health system and points to opportunities for managing Covid-19 (and future pandemics) better.
We explore one such opportunity, based on leveraging data, machine learning and artificial intelligence technologies, that transformed businesses over the last 20 years.
Personalised Covid-19 policies
How could this work for managing pandemics? Start by training a model to classify all individuals into either a high or a low clinical risk group (for example, whether an ICU bed would be required if they get infected)? Unlike current policies, we can then use the predictions and information from the model to confine and protect those with high risk, while deconfining those with lowest risk even if they may be infected – to also run the economy. We must also focus testing to people in touch with the vulnerable, fast – far less tests needed for that purpose than to test everyone.
We now know that only 1% of the Covid-19 cases experience severe symptoms. A perfect prediction would allow to fully free the remaining 99% — as long these are not in touch with the “vulnerable-one-percent” while contagious. We also know a lot about who is vulnerable to Covid-19.
Simulations that combine standard epidemic modelling with standard data science and AI principles indicate that even a moderate quality identification of high and low risk vulnerability people can allow significantly faster and safe de-confinement — without the challenges of medical or antibodies tests we now face.
This is also exactly the opposite of what current thinking suggests: instead of testing-and-confining with hindsight, we should predict-and-protect using data. Unlike medical tests which are scarce, expensive, and slow to deploy, this approach is digital, fast and easy to scale. It also leads to very different Covid-19 polices from those currently used, likely allowing for faster, and safe deconfinement.
Keeping ethical and behavioral considerations in mind, tracking technologies, using smartphone apps, or transportation data should be used to isolate those infected from those at risk, not from those who may not even notice if they get infected.
Scarce PPE, tests, and other resources should also be allocated in this personalised and targeted manner, rather than being spread around the general population thus reducing availability for those at risk – a key factor for many Covid-19 deaths as we slowly learn.
Technological roadblocks need to be resolved with appropriate policies. We can already use simple rules to identify the vulnerable, for example people with known preconditions known to increase severity risk, but careful identification requires careful models and good data.
For now, we need to do our best given the available data. But one lesson for the world to draw from this humanitarian disaster is that gong forward we cannot afford missing data to kill people. It is counter-intuitive to apply less sophistication in helping one breathe than to helping one shop.
National governments will need to agree on data standards to gather medical and other data for their citizens, and protocols for determining when data could be shared.
For example, a declaration by the WHO or UN that an outbreak qualified as a pandemic could serve as a trigger to temporarily and safely suspend normal privacy laws to allow the sharing of anonymised data.
During such times, many people might be willing to exceptionally and temporarily provide their data, through appropriate and secure channels, for training models that can guide policy decisions with major life and economic consequences.
Theodoros Evgeniou: Professor of Decision Sciences and Technology Management at INSEAD.
Anton Ovchinnikov: Distinguished Professor of Management Analytics at Smith School of Business, Canada, and a Visiting Professor at INSEAD.