How Artificial Intelligence Knew About COVID-19 Before We Did
Yes, you read the title correctly. An Artificial Intelligence Software designed to trace and locate the spread of viruses was alerted about the infectious diseases, COVID-19, nine days before the World Health Organization (WHO) released its statement alerting the public of the emergence of the coronavirus.
How It Went Down
It was December 30th, just a couple of hours past the hours of Midnight when an artificial intelligence platform known as BlueDot picked up “unusual pneumonia” cases occurring in Wuhan, China.
Little did the Canadian Tech Startup know that they had just discovered traces of the disease which would send the Globe into a frenzy unlike any other.
How Did The AI Detect COVID-19 So Fast?
With any Artificial Intelligence platform, the model needs data to train. Think of it like this: the more data which can be inputted into the model, the more precise outputs are given.
When starting up, BlueDot had a choice of what type of data to input into the model to predict potential emergences of diseases such as COVID-19. The two options which they narrowed down to were:
- Using Social Media Postings to detect concern in Geographic Locations
- Using Airline Ticketing Data + Official Reports to detect unusual activity
BlueDot ended up going with the second option, through which the platform predicted where the virus would spread to. This option examined where infected residents were flying out to, in order to track the spread of the virus.
Such practice has a name called Natural-Language Processing or NLP for short. These NLP algorithms used by BlueDot monitor airline ticketing data, official news outlets, and official health-care reports in different languages around the world, flagging whether they mention high-priority diseases, such as coronavirus and other diseases such as Zika.
Furthermore, these algorithms work in conjunction with unsupervised machine learning. Such is when the NLP algorithms are used to draw inferences from datasets consisting of input data without labeled responses. Meaning that there were no preselected examples/data sets which the model would train off of, just its own patterns which it had recognized from previous virus outbreaks.
What has happened after the First Positive Detection?
It was a breakthrough for the Artificial Intelligence Community when BlueDot was able to first detect COVID-19. However, one must remember that such was possible because there was accurate and precise data within news outlets, meaning the data was much more concrete.
However, now there have been tens of thousands of news articles published, thousands of videos, many of which contain false statements, meaning there is much less reliable data to be processed. There has been confusion over symptoms and how the virus passes between people to confusion in the decisions made by prominent political figures with authority. All of this noise decreases the accuracy of BlueDot’s and other Artificial Intelligence Companies.
Hence, although BlueDot was able to detect the first hint of the novel coronavirus way back in December of 2019, the reliability of its model has decreased, as with any machine learning model when presented with less accurate data to make predictions off of.
Examining the Impact of Different Diseases
SARS: 800 Fatalities, costing $50 Billion
Zika: 90 Countries impacted, costing between $10–$18 Billion
However, there has been nothing like COVID-19
COVID-19: Estimated 260,000+ Deaths, costing upward of $2 Trillion worldwide. (Note: This is the estimated cost as of May 6th, it will be much higher once we get past this pandemic).
How is Artificial Intelligence Helping Stop COVID-19?
As explained above, the number of cases of COVID-19 throughout the world are increasing at a rapid rate, one which no one expected, which consequently is leading to a massive hole in the healthcare industry- there simply aren’t enough resources and time for every person to be tested at a laboratory for the virus. Here is where Artificial Intelligence is trying to help.
In the upcoming weeks, it has become evident that machine learning can be used to diagnose patients with COVID-19 from CT Scans of patients’ lung tissues. However, this model depends on being trained to spot clear signs of the disease. There is where the problem lies at the moment, for physical signs of COVID-19 do not physically appear after about a week of infection, in which case the diagnostic is not very effective due to the timing.
Another difficulty in making sure that Artificial Intelligence provides useful help in the healthcare industry in such times is the problem of data. We discussed that the data prevalent right now is misconstrued due to the hundreds of thousands of people reporting on the virus on a daily basis. Such disallows for any data which may be true to be diluted. In turn, image recognition models can’t improve to detect those early signs of COVID-19.
How Can We Get Reliable Data Right Now?
We have discussed the importance of data to improving Artificial Intelligence models and how helpful/accurate data is scarce right now.
However, there is one immediate way, the predictions from models can improve. It’s controversial.
In order to drastically improve the accuracy of models in the status quo, the trade off to obtain better data: More Private/Personal Information about us has to be shared with companies like BlueDot and the federal government.
Take the United States, for example, if medical records of patients diagnosed with COVID-19 and those who showed symptoms were open to the public for data analysis, artificial intelligence could identify constituents at risk by examining the conditions it picked up to be causing the virus.
Additionally, the infrastructure and technology to read medical records and extract valuable/important information already exists.
However, the problem is that patient data is split up into hundreds of different health services, meaning split into many different databases, effectively making it harder to analyze those records.
Overall, it is truly remarkable that we are at the point in time where computerized algorithms can detect things to such a magnitude. However, there is still ground to cover in terms of improving the models, and more importantly the data which those models receive, making sure that what is reported is factual and not speculation. Furthermore, once we have crossed the data barrier, there must be a two way trust established: between the government (and its people) and the AI. Indeed are there privacy concerns, that’s something which won’t change, but is it worth it if we can stop the next COVID-19?
I will leave you with something to think about.
The Million-dollar Question: Are we willing to give up our medical privacy to stop the next global pandemic?
Thanks for reading my Medium Article! I hope that you learned something interesting and will use what I talked about to your advantage! The future of Artificial Intelligence has a lot of potential and can be very advantageous in many situations, however as with anything, there are drawbacks. Follow me on Medium to stay updated on how the ride goes!