There’s been much talk of the various AI virus prediction software platforms that predicted the spread of the Covid-19 pandemic. One platform that’s received a lot of attention is Bluedot, a Toronto based startup using artificial intelligence, machine learning and big data to locate, track and predict the movements of infectious diseases.
Although this startup was by no means the only one to flag-up the outbreak in Wuhan (as documented by MIT), the questions revolving around their prediction software begs the question of how such information should be acted upon in future cases.
Many have rightly noted that Bluedot’s virus prediction software could have allowed Wuhan and the surrounding cities to better prepare themselves. But with this assumption comes significant ethical questions of how much data such platforms should be allowed to access.
A Global Early Warning System
Using NLP (natural language processing) algorithms, Bluedot’ s AI virus prediction software gathers data on more than 150 diseases and syndromes across the globe, recalibrating its search every 15 minutes and running 24 hours a day.
Impressively (and crucially), this data is not limited solely to information from the WHO and CDC, but also spans wider areas such as commercial flight tracking, climate data from satellites and huge swathes of reports from international journalists: it’s claimed that, daily, over 100,000 articles in 65 different languages are processed.
Through manually classifying the data, Bluedot’s specialists created a taxonomy in which keywords are scanned, allowing the machine learning and NLP algorithms to run their course. The result is that minimal cases are actually flagged up, meaning that when they are, human expertise is required to analyse the anomalies.
With the information it gathers, Bluedot sends regular updates and alerts to its clients: mostly healthcare systems and services, governments and businesses. These alerts consist of brief synopses of anomalous disease or virus outbreaks that its software has discovered, along with a breakdown of the risks they pose.
Covid-19 was by no means the first time Bluedot has been one step ahead: predictions on the movements of Zika and Ebola virus being notable cases. So when in late December, 2019, their software came across a cluster of 27 pneumonia cases at a seafood and live animals market in Wuhan, experts were called in to analyse the anomaly.
We’re all painfully aware of how events panned out from that moment. But one of the biggest questions raised has centred around what role machine learning and artificial intelligence could play when another outbreak occurs.
How to act upon the signal?
Founder and CEO, Dr Kamran Khan has been quick to point out that their product can by no means offer a full solution to outbreaks such as Covid-19. It’s about utilising the programme to find the signal in the noise, or the needle in the haystack, as Khan puts it. Once we have this needle, it’s crucial that we know exactly when and how we should act on it.
Perhaps the most painful issue with such data is that there is no neatly drawn out path the international community should follow. The next steps invariably have to be taken by official governing bodies: how each government chooses to implement new measures and to coordinate these internationally will vary in each instance.
So whilst Bluedot successfully flagged up the outbreak, its alert served only as a heads-up. This is not to take away from the immense utility of Bluedot’s software. But it does lead us to ask how useful such technologies can really be beyond acting as alerts.
It’s worth noting that once any virus spreads exponentially, the difficulty in reading the large swathes of data covering it augments simultaneously. This is mostly due to the inconsistencies that inevitably occur as the coverage of the data spreads: many news channels will offer alternative, or even contradictory articles which challenge the software’s ability to pinpoint reliable information.
In the case of Covid-19, there has also been much confusion over what the symptoms actually are and how they are spread, meaning that the AI’s task switches from just finding the needles in the haystack, to determining what actually are verifiable needles.
Inextricably linked to this problem are the vested interests of each government and the information they want to portray. Whilst it’s easy to point the finger at China’s overt censorship, the issues faced in the international community’s relaying of reliable data are multifaceted and highly complex, to say the least. This is before we even consider the controversy regarding data protection, and whether or not health-tech startups should have access to huge amounts of medical records, as suggested by Darren Schulte, CEO of Apixio.
What we do know is that as the pandemic progresses, much of the data surrounding it will inevitably become less reliable. As made patently clear by Kahn, virus prediction software can find the signal, but ultimately it is down to humans to decide how to act on it. Having seen how apt Bluedot’s technology was at flagging up Covid-19, we surely have to ask ourselves, how should governments and the international community act on such data when it does arise?