Original article can be found here (source): Artificial Intelligence on Medium
Can the current technology (Artificial Intelligence) stop Coronavirus COVID-19 Outbreak and help in diagnosis?
Don’t forget to answer the questions at the end of this article!
In 2009, Google released a scientific paper with the name “Detecting influenza epidemics using search engine query data” and developed an AI tool to look deeply into people search patterns queries by tracking their search behavior activities on their search engine hoping it might be able with their massive available data structure “big data hubris” to build an AI model to predict flu outbreaks even faster than traditional health authorities such as CDC. Google announced “Google Flu” model but the model has failed after couple or month missing prediction of swine flu (H1N1 influenza) pandemic and give false prediction after all, not only that but Google model failed to read between the lines and was wrong for more than 90% of results since 2011 despite the 95% accuracy of predicting the spread of flu which was results based on training a model on past data, it’s easier to predict what happened in the past because all data are available somehow to feed it into the model with more connected dots to draw the final image. Goggle released another scientific paper in 2011 with the name “Assessing Google Flu Trends Performance in the United States during the 2009 Influenza Virus A (H1N1) Pandemic” trying to optimize their Google flu trends-GFT model but the model has also failed to predict pH1N1 and many pandemics after sporting sharp changes in people online search behavior that made the whole model confused to predict what will happen in future and started to give wrong random predictions based on how their algorithms are making it out with generated live stream data, diverse dots with no correlated function between them that couldn’t be connected to form the final play on the stage and give nearly a precise predictions enough for mass adoption. In the end, the model can’t take our searches into granted to predict future pandemics, we are not experts and I myself do a lot of queries that are not even understandable to me and will get finest search results but it’s not enough for GFT to track our search behavior and use it in predicting any upcoming pandemics such as our current Coronavirus COVID-19. in 2013, Google dropped down this project and provided the results to the public for further research and developments done by many universities such as Harvard, Houston, and Northeastern universities that compared between GFT and baseline model built on using CDC datasets, a huge gap in both performances triggered how GFT was not the chosen model to do the predictions.
Nowadays, we are facing similar problems. Many data science companies are working on analyzing COVID-19 data for further prediction, such as the Canadian company “DarwinAI: that has been developing a deep learning model (basically AI model) using image processing technique and Convolutional neural networks CNN (called COVID-Net) trained and tested on chest x-ray photos to find COVID-19 hidden patterns in chest x-ray scans (similar has been done on breast chest cancer) since this pandemic target lungs and weaken the pneumonia system and there should be a new advance lungs symptoms and sever inflammation newly emerged toward uniquely fingerprints that have never seen before by doctors, the first warning was by Chinese doctor that warned the authorities after seeing unusual advanced developments. this model is pretty efficient since it uses a semi-supervised approach (supervised learning with labeled data + unsupervised learning with unlabeled data where the model can learn by itself without teaching it). However, it is not the best way to diagnose an early stage of the virus that will take 14–27 days to show symptoms on the infected host which makes it efficient only on late stages and getting harder to control the spread due to the current measures. Another AI models and diagnostic tools used to assist in developing the vaccine but still in the early stage and require a massive dataset and global pieces of information to be unlocked to the public in order to acquire a bigger image for better developments, in all cases it will take a lot of time to put an effective vaccine in the hands of health care systems, but (N) months better than (N+M) months where M>0!. BlueDot and Metabiota are AI companies to help monitor outbreaks, their business model to spot changes around the world and report warning alarms to protocol customers such as governments, NGOs, Hospitals and so on so forth. BlueDot alerted clients, WHO and Chinese governments regarding SARS-COVID2 (aka Coronavirus-.COVID 19) on 31-dec-2019, Metaniota with automated AI service called HealthMap and BlueDot astonishing models both used in helping to acknowledge early warnings that should have more value in saving more lives with a prompt response and immediate actions taken by authorities that supposed to has already well-prepared precautions and backup plans for worse case scenarios. Also, BlueDot and Metabiota used natural language processing (NLP) algorithms to go through all news and official healthcare reports in different languages around the world and flag the most important trend and give a certain prediction on air travel data and the risk of traveling. on the otherwise company named by Startifyd scans posts from social media such as Facebook and Twitter and cross-reference them with official health organizations WHO and CDC or other for validating information and conduct certain prediction and analysis accordingly, but it’s not a novel approach to fight disinformation.
So, the world now has tools that can acceptably detect later stages of infection individually and early-stage outbreak globally, but how about (1) later stages outbreak globally and (2) early-stage infection individually?. There are many scientific studies with various and different results about potential symptoms, development period of time, the duration of staying in the air or on different surfaces, proportional value of transferring the virus to others, sever cases “L” and “S” levels and more statistics based on local data provided to enrich AI models, strategic approaches to reverse engineer the virus mutation to the original source and digging into how COVID-19 evolve with time and the probability to shift-shape RNA to a new structure that could infect the recovered patients again according to one study shows over than 40 mutations happened since the outbreak. (1) and (2) are pretty connected where revealing one of both will assist the other. the current problem is as described, There are not enough answers or disjunctive union of available results to follow in diagnosing (2) that will lead to the fine-tuning (1) model and optimize the way of controlling the outbreak and vice versa. Otherwise, we are living now a global quarantine because there is no state of art models available for the public to help achieve the inflection point and flatten the curve besides limited healthcare resources, lack of knowledge about the virus behavior + symptoms + mutations, low preparation rate in all countries and delays by not taking it serious enough or for other political leverage and reasons.
The current problem as well defined as governments are not sharing patients’ data with each other due to different privacy regulations and laws within each country that bound their act to citizens (with exceptions!), companies are not sharing data with each other from a business perspective and to survive the current global recession. citizens don’t want to share their private records with companies, governments, or any other third party agencies for Data Scientists, researchers, and developers to build a state of art models trained on various data experiences with more features extracted from aggregating databases and fine-tune the model following optimization process to get better accuracy and diagnoses in return, also health care workers will not be able to use the recent technology and AI models to assist them in their diagnoses, at the end automated AI and doctors need each other to achieve satisfying results and it requires combining both forces. AI models better with sensitivity to spot hidden patterns in X-ray photos, while humans are better in specificity to have the final say of the context of the whole provided case.
It is a critical situation, where the feel of ignorance is out there and current solutions with AI can’t predict and solve (2) and (1) on a large scale.
The three Pendamic Triangle Thorey PTT says should sacrify one of the three key-value points “Economy”, “Public Saftey” or “Privacy” for a short period of time to be able to overcome the current situation with less damage on all scale. Asian Countries like China, Taiwan, Noth Korea, and Hong Kong violated the privacy of their citizens on a different scale and measures to temporarily contain the outbreak and flatten the curve, I said temporarily because some studies predict a second wave of infection and maybe more who knows. The UK followed the mitigation process of reserving Economy and privacy before updating its plan to suppression to follow EU countries to preferring Public safety and Privacy over the Economy. sacrificing Economy will increase the rate of unemployment and bankruptcy, sacrificing Public safety will increase the rate of death and sacrificing Privacy will violate the current laws and regulations (GDPR and PwC). Fine tunning between those 3 keys will be a tough challenge in the long term. In Germany, Telecom and Telefonica (O2) follow the Differential privacy technique and share anonymized data with institutes such as Robert Kock (RKI) for further analysis of population movements without reveling the actual identities of users or they can use something called federated learning used by google or apple to train the model on local data and fetch the model back to the source and aggregate it with other models to have the master model, similar to a puzzle game where each piece is a model trained on users local data and the master model is the final result of stitching. Many projects using Blockchain technology to build a protocol for unlocking and exchange data between entities in a secure environment taken into consideration the privacy concerns especially in the healthcare ecosystem, however, these cutting edge projects still in early-stage facing many challenges related to scalability, transparency, interoperability and need more research and time to prove its value for mass adoption.
There is no available data to the public to optimize current solutions and build better technology, Not enough global transparency, a lot of disinformation spread online on the internet, media and governments will play things up and down or hide information, a lot of mysteries questions about COVID-19 behavior and symptoms with the current exponential growth curve and many more that might at particular point lead to changes in political models to follow other successful experiences such as Asian countries such as China that used social credit system that mines all kind of data from its citizens for classification and to applies curfew on infected people, also building apps to explore patients (potential patients) on a map near them in order to take precautions.
I talked about countries fully equipped, have a high technology to fight this and should be well prepared but it seems like there is a lot of medical storage at their end and barely manages internal issues, how about the countries that don’t have a strong economy, fragile healthcare systems, and no human rights?
There are many refugee camps that don’t have the luxury to commit to the “wash hands” process since they wash their kids once every week if water was available!.
Q: Will other countries follow the Chinese model?
Q: Will countries are more transparent about the current situation or when may a violation of citizens’ privacy or tracking location take place?
Q: Will citizens trust to give their privacy (medical records and locations) in exchange for developing better technology and AI models that will help healthcare workers and accelerate the process of containment? or for if they knew that their data might save others’ life?
Q: Will governments or third parties take the situation granted to misuse private data?
Q: Will the WHO and CDC put a contingency plan for how to deal with COVID-19 if it hits poor countries or refugee camps around the world and it will sooner or later?
Q: Will international communities push to help refugees camps around the world, first the for sake of controlling the outbreak! and second to save human life by avoiding any catastrophic disasters
References that inspired me to write