Skip to content Skip to sidebar Skip to footer

Source: World Economic Forum as of 08-05-2020

  • Artificial Intelligence (AI) technologies like Natural Language Processing (NLP) can help researchers tackle COVID-19 by crunching huge amounts of data that would be impossible for humans to process.
  • Machines can find, evaluate and summarise the tens of thousands of research papers on the new coronavirus, to which thousands are added every week as scientists race to find a vaccine or treatment.
  • AI can also help us track the spread of COVID-19 and other diseases, by detecting early warning signals such as clusters of symptoms in a new place.

In the summer of 2015, I was diagnosed with breast cancer. Over the course of my treatment, I read dozens of research papers to assess my options. I am a specialist in Artificial Intelligence (AI), and was able to use my training in concepts like statistical significance to understand this medical research. As I worked my way through the papers, I realised what a difference AI technologies from my field could make to doctors, scientists and patients, for example in the form of sophisticated tools to search a global cancer database.

AI has the power to transform medical research, and not just cancer research. Right now, it can help us address a different and extremely urgent medical challenge: fighting COVID-19.

Machines can assist us in analysing the masses of research published on the new coronavirus, and extracting the most relevant information. They can crunch data much faster than humans, and detect patterns that we may be slow to see. This means they can also help us quickly understand data on the spread of COVID-19 and other diseases, and intervene early.


Coronavirus research has exploded in response to the COVID-19 pandemic. There are now more than 60,000 papers.
Image: CORD-19: The COVID-19 Open Research Dataset

Here are two ways researchers and institutions around the world, including my team and myself, are tapping the power of machines to overcome the COVID-19 crisis.

Fast answers to urgent questions

There are about 67,000 scholarly papers related to the new coronavirus, and the number is growing every day. Only in the space of the past week or so, thousands of papers were added. Every scientist working on a cure or vaccine must understand this prior research in order to work as fast as possible, and avoid errors and duplications. However, going through all these papers would take a person years, even if they did nothing but read. Machines could change this, and help us get an instant overview.

In March, a partnership of several institutions including the White House created the COVID-19 Open Research Dataset (CORD-19) as a free resource for the global AI community. It consists of the tens of thousands of coronavirus-related papers published so far, and is constantly updated with new publications. AI specialists around the world, including my team and myself, have set to work to develop tools to search this dataset and extract the most significant information, using machine learning. This could allow medical researchers to keep up with the fast-evolving field.

Machine learning, an essential part of Artificial Intelligence, consists of machines being trained on huge data samples, after which they can independently analyse data sets and detect patterns, trends, relationships and such. Machine learning is already being used to predict the speed and pattern of COVID-19 spread, and to enable the research for a vaccine, but has rarely been used to process medical research papers in a large scale before. It could prove game-changing when it comes to COVID-19 research.

My own team’s proposed tool for the COVID-19 open dataset uses an AI technology called Natural Language Processing (NLP). This involves training machines to analyze a user question in full sentence, then to read the tens of thousands of scholarly articles in the database, rank them and generate answer snippets and summaries.

A scientist may for example type a simple question into a search engine tied to the database, such as “Is COVID-19 seasonal?”. The search engine then finds and ranks all related papers. A programme called a “neural question-answering engine” goes through the top-ranked papers and identifies passages that answer the question. Another programme called a “neural summarization engine” creates a summary of all these passages, and presents them to the user in the form of an abstract with the most important and up-to-date information on the seasonality of COVID-19.

“Neural” means that these algorithms are based on neural networks, similar to the ones in our brain, with information travelling along many different connections. This allows machines to learn by considering many examples of something, then independently evaluating a new piece of information. The process is also known as deep learning.

Using these technologies, our team was able to build a system that can answer questions on the spot, based on the ever-growing database of scientific publications. We are releasing the source code of our system to the public, so that other developers can use it for further work.

Detecting the threat

Every day, a piece of information emerges somewhere in the world that could probably help us prevent the next pandemic. It could be something seemingly small, like a cluster of strange syndromes in an unexpected place. The challenge is to detect this crucial information in a sea of other, irrelevant observations. This task is at the core of epidemic intelligence – spotting and assessing epidemics, ideally in their earliest stage. It’s another field where a number of machine-learning projects are promising to make a huge difference.

Epidemic Intelligence from Open Sources (EIOS) is an initiative run by the World Health Organization’s Health Emergency Information and Risk Assessment Department. It gathers and analyses data from sources such as health monitoring programmes and different health databases. The first article reporting on a pneumonia cluster in Wuhan in December was picked up by this unit.

We recently started working with the EIOS team to provide question-answering and summarization technologies similar to the ones we developed for the CORD-19 database. In this case, our system needs to sift not through tens of thousands, but hundreds of millions of pieces of online material, to find answers to questions. Health ministries and centres for disease control in all member states of the United Nations, as well as medical and healthcare professionals will use these tools for epidemic research.

These projects are advancing quickly. However, in order to really harness the power of AI for public health and medical research, we need to have more collaboration across nations and between scientists, as well as open access to medical databases while protecting patient privacy.

Using AI for human health

The fight against COVID-19 has brought untold misery and destruction to the world, but it has also shown what we can achieve if we work together. The open and collaborative nature of Artificial Intelligence research and development across borders show us what such joint progress might look like in the future.

My experience in the field of Artificial Intelligence served me well during the treatment of my own cancer. It helped me understand the research that ultimately helped save my life. Developing a global cancer database, and unlocking its potential through machine learning, remains one of my principal goals.

Given what has been achieved by doctors and researchers in the COVID-19 crisis, working in extreme conditions and under huge pressure, I know that we can all rise to the challenge of addressing the biggest threats to our health, and using technology to benefit us all. Artificial Intelligence and machine learning can assist diagnosis, speed up the development of treatments, vaccines, drugs and preventive measures for millions of patients, and improve human life now and in the future.

Show CommentsClose Comments

Leave a comment

News ORS © 2020. All Rights Reserved.