Beyond Bats and Electric Sheep

Manila. March 2020. Philippine Strongman President Rodrigo Duterte just declared a
nationwide lockdown due to the rapid spread of COVID-19 across the country. The echo
chamber on social media platforms such as Facebook and Twitter would indicate this action as a
decisive one. In reality, however, it’s a little too slow, a little too late. In a month, COVID-19 had
already spread to every single region in the Philippines, and cases continued to soar. In just a
little over a year, this spread has resulted in more than 1.5 million cases, and more than 25,000
deaths. The Philippines has consistently been ranked to be in dire straits. Despite having the
longest lockdown in Asia, not only has the Philippines suffered significant economic losses, but
COVID-19 has also wreaked havoc to many sectors of the country with national healthcare at the
brink of collapse from its core. As pervasive as COVID-19 is, it seems as if it cannot be stopped,
merely slowed down. Yet, there’s a beacon of light that’s massively accelerating in crafting
solutions to combat this angel of death. At no other time are artificial intelligence (AI) and data
science being developed at such a rapid pace resulting in breakthroughs in healthcare and
medicine. AI and data science have been employed towards vaccine creation, drug discovery,
disease spread monitoring and management, contact tracing, and mass testing efforts, among
others.

 

In the Philippines, the Philippine Red Cross (PRC) utilizes AI and data science in
countless ways. In collaboration with Y-Combinator startup Dashlabs.ai, the PRC is able to
monitor, provide, and predict real-time hospital information, including that of bed availability. At
the onset of the pandemic, hospitals were quickly overwhelmed. Consequently, many were not
able to find beds immediately, and some of those who waited ended up dying. To resolve this
extremely concerning issue, a dashboard was developed providing essential real-time
information — location, contact details, and regular and ICU bed capacity — on hospitals in the
country. Using data science and AI, the platform also provides capacity predictions based on
historical data so the PRC can know where to direct availability requests.

 

While predicting hospital capacities and bed requests can largely be considered reactive,
COVID-19 response must more importantly be proactive. Enter clinical epidemiologists
empowered by AI. To predict the spread of COVID-19 in the Philippines, researchers from the
University of the Philippines (UP) and the Philippine General Hospital (PGH), used different
epidemiological models for different cities (UP COVID-19 Response Team, 2020) within the
National Capital Region (NCR). These models included a population density-based variation of
the SIR model (susceptible-infected-recovered), an SEIR model
(susceptible-exposed-infected-recovered), and another SIR model created by Darwin Bandoy and
UP Los Baños. To aid in their calculations, the researchers used R0 or the basic reproduction
number of an infection in the aforementioned models. By running simulations, the researchers
involved were able to create suitable predictions of varying types. On their own COVID-19
dashboard, they present projections for optimistic, actual, and pessimistic figures of the number
of new cases based on a 7-day moving average (UPLB Biomathematics Initiative, 2020).

 

Another facet of COVID-19 response where AI is used is that of data integrity. The
Electronic Case Investigation Form (e-CIF) system developed by the PRC allows arriving
passengers who enter their personal and travel details in advance to easily go through the testing
process. They are able to monitor their specimen, get their test result, and receive a clearance
certificate. In the e-CIF, a photo of an identification document is required. While some passenger
information may be taken from the text fields, passport photos remain vital, as they still serve as
the most common and trusted source of information for passengers. To extract the necessary
information, passports uploaded to the e-CIF go through AI object detection and Optical
Character Recognition (OCR). OCR converts images containing text into text objects. This
process is done for millions of passports, significantly fast-tracking authenticated patient data
processing. We were able to build a fast and easy way to link these with the mandatory swab
tests for passengers without any special equipment. With the increased data accuracy from an
originally manual encoding process marred with errors, every passenger got their correct result in
no time (Chua, 2020)

 

Aside from data integrity, data science and AI can also be used for analysis of large
datasets. The Philippine Red Cross used this to implement mass testing at rapid rates. As we
know, there are a number of tests that can be used to detect the presence of COVID-19 in an
individual — nasopharyngeal swab reverse transcription-polymerase chain reaction tests
(RT-PCR), antigen tests and many more. Using data science, the PRC determined that the results
provided by Saliva RT-PCR are not significantly different compared to that of a nasopharyngeal
swab test! Due to this proof, the country’s Department of Health has authorized the use of the
cheaper Saliva RT-PCR tests as a viable replacement for the more expensive RT-PCR swab test,
without sacrificing quality and accuracy (Hernando-Malipot, 2021). We’ve since seen an uptick
of people getting tested, as testing is now more economical!

 

At the very center of testing is ensuring that COVID-19 test results are as accurate as
possible. Machine learning works by feeding an input — called training data — processing them
through some code, then the machine giving a decisive output to a certain degree of accuracy.
Once again, the PRC and Dashlabs.ai collaborated to design and implement a machine learning
model to quickly and accurately recommend test results to pathologists. Nerd hat on. The model
consists of an artificial neural network with multiple layers. About 48,000 pre-labeled test results
were used for training the model and 12,000 for testing. The model’s first layer consisted of a
135-neuron dense layer, with relu as its activation function. A sigmoid function was also used as
the network’s final layer which ultimately provided a quantitative measurement of the presence
or absence of the virus for a particular test (Dashlabs.ai, 2021). Got it? Got it. The model
achieved an accuracy of 99.03% for samples from the test set, a remarkable feat for the
developers and the entire country!

 

There are even more efforts by the Philippines to adopt AI in COVID-19 response. The
Department of Science and Technology (DOST) developed the “CHERISH App” which uses
artificial intelligence and machine learning to provide an alternative diagnostic for the presence
of COVID-19. In accordance with the DOST’s thrust of “AI for a Better Normal”, this app
analyzes chest x-rays to determine the possibility of a person contracting the virus via
pneumonia. Foundationally, this app echoes one of the fundamental usages of machine learning:
anomaly detection. In a nutshell, machine learning models can be trained to distinguish between
images with varying characteristics; in this context, it would be images of healthy chest x-rays
versus those showing pneumonia. In relation to the potential success of the CHERISH app,
DOST-Philippine Council for Health Research and Development (PCHRD) Executive Director
Dr. Jaime C. Montoya said that “Once fully developed, this technology will be able to assist
medical professionals by offering an alternative, scalable, and efficient mechanism that would
augment the current workflow of our hospitals and testing center to screen for possible
COVID-19 infection among suspected patients. This could pave the way for a total[sic] different
future of healthcare” (Luci-Atienza, 2021).

 

The Philippines’ Department of Health also employs AI in COVID-19 response. When
people have inquiries about COVID-19, and are unable to speak with medical professionals
immediately, alternatives must be provided. In collaboration with Facebook and Aiah.ai, the
DOH Kontra COVID Bot was launched in April 2020. With “Kontra” being the Filipino word for
“against”, this chatbot aims to address any concerns about COVID-19, whether it be regarding
protection against the virus, clarifications about quarantine guidelines, or if one is at risk of
contracting the disease (Aiah.ai, 2020).

 

Other countries heavily use machine learning as well to solve other COVID-related
problems. BenevolentAI, a UK biotechnology company does AI-assisted drug discovery.
BenevolentAI uses their AWS-powered Knowledge Graph that “incorporates the entire corpus of
relevant, publicly available data and is continuously enriched by in-house experimental results”
(BenevolentAI, n.d.). Together with their models, they are then able to gain clearer insights of
the nature of a disease, essential in finding effective drugs in COVID-19 treatment. From the
results, they found that Baricitinib, a drug often used for treating rheumatoid arthritis, possesses
anti-viral properties that can also be used in treating COVID-19. Since their discovery,
Baracitinib has been “approved by the FDA for emergency use and proven to reduce mortality of
hospitalised patients by 38%.” Across the pond, researchers from Johns Hopkins University
published a COVID-19 severity prediction model. Depending on certain information such as
race, age, BMI, and comorbidities, the model can successfully predict how much a patient’s
condition worsens after a certain time period (Garibaldi et al., 2020). Consequently, this enables
better preparation by all stakeholders.

 

Did you know, however, that AI was used to confirm the existence of COVID-19?
Remarkably, COVID-19 was actually first detected as a pneumonia-causing disease in Hubei,
China through AI software by Bluedot. Using artificial intelligence, their software is able to
process vast amounts of data through their intricate surveillance system. By combining various
real-time sources of information such as news, flight patterns and schedules, population
movements and many more, Bluedot was able to accurately predict COVID-19’s outbreak risk.
Through charting how COVID-19 spreads around the world and which countries will be hit the
hardest, the company was able to advise appropriate authorities on the risk involved and what
measures they can take to minimize the spread of COVID-19 in their country, state, or
community.

 

There is no doubt that data science and AI have played an invaluable role in COVID-19
response both here in the Philippines and around the world. The recent rapid developments in
data science and AI, however, were partly borne out of necessity. Why don’t we make these
developments, and their pace, the norm? Despite their fairly nascent nature, data science and AI
seem to be the keys to revolutionizing lives worldwide.

 

The 1968 dystopian novel asks, Do Androids Dream of Electric Sheep? We say, without
the need to wait for dystopia: We need to quickly build AI that can dream beyond electric sheep!

 

Bibliography

Aiah.ai. (2020, April 23). In partnership with the Department of Health (Philippines) , #Aiah
launches KIRA #KontraCovidBot to help disseminate credible news and information about #COVID19PH. [Image attached] [Post]. Facebook.
https://www.facebook.com/AiahAutomation/posts/in-partnership-with-the-department-ofhealth-philippines-aiah-launches-kira-kont/561694421396381/

AI-Driven Drug Discovery | BenevolentAI. (n.d.). Retrieved July 10, 2021, from
https://www.benevolent.com/

Bluedot. (2020, June 03). CNN: How AI is tracking COVID’s spread. Retrieved July 11, 2021,
from https://www.youtube.com/watch?v=ZyGdwRYmS6A

Chua, S. (2020, November 03). In-Depth Analysis of Identification Documents Through Vision
AI. Retrieved July 9, 2021, from
https://pub.towardsai.net/in-depth-analysis-of-identification-documents-through-vision-ai
-2c85651e80b0

Contracovid. (2020, July 23). Retrieved July 10, 2021, from https://aiah.ai/contracovid/

Dashlabs.ai. (2021, May). Medtech/Pathologist Onboarding (Machine). Retrieved July 12, 2021,
from https://help.dashlabs.ai/onboarding/medtech-patho-onboarding

Discovering COVID-19 treatment | BenevolentAI. (n.d.). Retrieved July 10, 2021, from
https://www.benevolent.com/covid-19

Garibaldi, B., Zeger, S., Gupta, A., Robinson, M., Rouhizadeh, M., Perin, J., . . . Bandeen-Roche, K. (2020, September 23). Johns Hopkins Researchers Publish COVID-19 ‘Prediction
Model’. Retrieved July 10, 2021, from
https://www.hopkinsmedicine.org/news/newsroom/news-releases/johns-hopkins-research
ers-publish-covid-19-prediction-model

Hernando-Malipot, M. (2021, June 25). Saliva test of PH Red Cross now included in all official DOH laboratory results. Retrieved July 11, 2021, from
https://mb.com.ph/2021/06/25/saliva-test-of-ph-red-cross-now-included-in-all-official-do
h-laboratory-results/

Luci-Atienza, C. (2021, June 24). ‘AI for a better normal’: PH’s COVID-19 Artificial Intelligence technologies unveiled. Retrieved July 9, 2021, from
https://mb.com.ph/2021/06/24/ai-for-a-better-normal-phs-covid-19-artificial-intelligencetechnologies-unveiled/

UP COVID-19 Pandemic Response Team. (2020, April 07). Preparing for a Post-ECQ Scenario: Analysis and Recommendations. Retrieved July 11, 2021, from
https://up.edu.ph/preparing-for-a-post-ecq-scenario-analysis-and-recommendations/

UPLB Biomathematics Initiative. (2020). PROJECTIONS UNDER THE NEW NORMAL.
Retrieved from https://endcov.ph/projections