As science gets more advanced, we realise how interconnected our world is. The smallest things can have the biggest impacts. For instance, researchers are today looking at how the sand that every year is blown from Africa through the Atlantic, is actually impacting the number of hurricanes that end up hitting the East coast of the US. Turns out, the small grains of sand ‘pierce’ through the air currents, which are growing hurricanes, absorbing the moisture and consequently cutting off the ‘fuel’ that would allow the hurricane to grow.
That starts to give you an appreciation of the big picture. Things that are happening in one part of the world, don’t just stay in that part of the world.
In an ever more globalised world, this is even more true for diseases, to the point where the question of a global disease is not “if”, but “when”. Aside from the evident life and health casualties, a pandemic has also an elevated economic impact, costing on average between $5 and $8 trillion to the global economy.
More specifically, and according to the World Economic Forum, fighting COVID-19 could cost 500 times as much as pandemic prevention measures. Some experts estimated that COVID-19 pandemic could end up costing between $8.1 and $15.8 trillion globally.
Right now, data is simply not being effectively collected, analysed or deployed to drive decisions to stop outbreaks on their tracks and this year’s pandemic is a live example of that. However, if this has taught us something is that, we are in the age of an information revolution where Big data & Artificial Intelligence (AI) can help us do more informed decisions every day. Shifting from reaction to prevention, reducing the impacts of shocks and thus improve the resilience of our systems.
There is growing optimism that newer approaches, including mobile-phone location tracking and data mining of search engines and social media, can help deliver a faster, more refined picture of where diseases are unfolding and might head to next.
Leaning on AI to Prevent the Next Outbreak
According to some of the most recent reports, AI had detected this coronavirus at its very early stages. The company BlueDot, which uses machine learning to monitor the spread of contagious diseases around the world, had alerted about the rapid increase in pulmonary disease in Wuhan late last year.
More specifically, BlueDot gathers data on over 150 diseases and syndromes around the world searching every 15 minutes, 24 hours a day, including official data from organisations like the Center for Disease Control or the World Health Organization but also data from less official sources like worldwide travel patterns, environmental and animal data, social media sensing. It then classifies this data into a taxonomy and applies machine learning to identify relevant flagged cases for further analysis.
While these are still early stages for AI, if enough trust was given to such a model, it could have helped authorities prepare, alert and take the necessary measures which could have perhaps prevented the outbreak in the first place. It is not unwarranted to think that going forward more attention may be given to these signals.
One can logically expect for these systems to only improve with time, more so since every single day there is more and more publicly available data that can be taken into account to more accurately pinpoint the beginnings of what could derive into an outbreak, or even to sense conditions or potential hotspots of what could lead to a brewing disease. Analysing such vast pools of data can be possible due to the amount and the speed at which AI systems can go through the data to detect patterns.
Although anticipating the appearance of a virus will carry some margin of error, especially at the beginning, since the AI is as good as the quality and volume of data it is fed, over time it could still detect conditions instantly and more accurately than the experts who initially fed it.
This might be only the beginning of an era where applying predictive modelling to millions of data points can help detect the danger before it appears. For example, one way of doing it could be using probabilistic forecasting, data mining and scenario planning, then improving the models over time with deep reinforcement learning so that the system eventually becomes autonomous.
Shifting from Reactive to Predictive
While the ability to predict the future with certainty is not something that can be said lightly, this uncertainty could be reduced through predictive analysis. Predicting the course of an epidemic, even after it has started, can still help authorities plan better to contain its spread.
However, the true potential will come from being able to accurately identify areas with high degree of probability for a potential viral disease to be born, which has the ability to easily grow into an outbreak, therefore predicting or anticipating the event.
By taking into account inherent or created conditions of the specific area, travel patterns, food habits, micro and macro environmental parameters and generally using global datasets on diseases to detect weak signals and then mapping risk-prone areas we could then use AI to predict the danger.
The crucial element here is the broadness, quality and detail of the data since the model needs to make predictions at a global scale. Therefore, to accurately forecast diseases across the world and not just in a few locations the data needs to be expansive.
As opposed to a few years ago, the ability of the latest software to listen to a much wider range of sources and signals is significantly higher and should only increase in the coming years.
Some of the challenges of the predictive models
Although exciting, this technological advance is mostly fuelled by very variable and frequently inaccurate data, which is currently generating doubts on the veracity of the data sources but also, if the model does offer predicative scenarios, it could become less accurate as arising cases are prevented.
What’s more, while trying to predict potential outcomes in other industries may be somewhat simpler, accurately looking into future epidemics or outbreaks could be intrinsically linked to our current biological and medical knowledge. Even now, after the contagion occurred there has been confusion over symptoms and the way the virus passes between people. Therefore, trying to predict where a disease may spread from hundreds of sites is a far more complex task then one would imagine.
One of the main concerns of feeding the AI with inconsistent data is the criticality of the decisions to be made, specifically when applied to health. It will therefore be essential for the data to triangulate between quality, reliability and agility and find a balance between the trade-offs between agility and reliability depending on the context at hand. As for the confidence or reliability of the data itself, risk can be minimised if more data is made open to the public for analysis, which then calls for a hard conversation on privacy.
Although it wouldn’t necessarily solve privacy issues, some distributed ledgers (e.g. Blockchain) could solve some of the other issues by bringing more decentralization, transparency and guarantee of data integrity.
AI has shown much promise in augmenting human capabilities. Certainly, the combination of the current and future available data with our intelligence and the power of AI will create more innovation and get us more prepared for future risks. As with every crisis, we have become stronger, more innovative and creative, hopefully leaving societies more prepared to face such threats.
COVID-19 attracted a lot of attention to the financial impact, but even the cost of common infectious diseases can be enormous.
Certainly now, investors and nation leaders are more aware than before of the importance of having not only robust health systems and solutions to threats but also predictive insights to prevent those threats where possible, or at least see them coming. And since fighting pandemics like COVID-19 could cost 500 times as much as pandemic prevention measures, it is not unreasonable to think that the era of predictive analytics is just starting. More so, considering the accelerating number of catastrophes occurring and which are still expected to happen (particularly given the current global warming crisis we are facing), the necessity to move fast is becoming pressing.
Leveraging Big Data & AI to deliver predictive analytics will definitely improve decision making not only in healthcare but across all sectors and industries.