Let's read the text
Required reading time: 4 minutes
It was an algorithm that was the first in the world to recognize the danger of a pandemic. An outbreak risk software, BlueDots, registered the emerging risk of COVID-19 in Hubei Province, China, on December 30, 2019. The founder of the Canadian start-up, Kamran Khan, learned already from the SARS epidemic in 2003 that data can help to be faster than the virus.
Countless data scientists and AI experts are now in a similar learning process. Many have been involved in countless voluntary initiatives in recent months: Google counts 19 million search results for the query "COVID Hackathon" alone. All of them pursued the goal of using data and intelligent technologies to cope with the pandemic.
But what has AI got to counter a virus?
To answer this question, I looked at the 26 presentations of this year's virtual spring conference of the Stanford Institute for Human-Centered Artificial Intelligence (HAI). Experts from the Harvard Medical Institute, the Chan Zuckerberg Initiative or the non-profit ML Community Kaggle discussed on 1 April 2020 "How AI can be used to combat COVID". It is easy to describe in one word the biggest challenge for all speakers in the fight against COVID: data.
It is worth mentioning that I owe access to the event to the virus: Stanford had to digitalize the conference because of the pandemic and was able to increase the number of participants from 800 to 10,000. All presentations are also available on YouTube available.
Data Tsunami: In the center of attention
The challenge: Anthony Goldbloom, Founder and CEO of Community Kaggle, reported an extremely high volume of new scientific texts suddenly published around the world. Many researchers have devoted more attention to the pandemic, publishing an average of more than 1,300 scientific reports per week since January 2020 (source: Google Scholar). This amount of information was difficult for people to grasp.
The solution: Natural Language Processing (NLP) helped to cope with the flood of data. The Kaggle network, for example, developed an automated literature review process: newly published papers were scanned and summarized. A user could now ask a question and the system showed all the appropriate answers.
Conclusion: AI can capture large amounts of data faster and filter out certain information quickly. This benefits doctors and researchers all over the world, for example.
Data ebb: Blurred traces
The challenge: John Brownstein, professor at Harvard Medical School, stressed how difficult it was to get a realistic picture of the spread of the virus due to a lack of official sources. For example, to track the outbreak in China, he and his team used social media (WeChat) and press data. But not even data mining, machine learning and web scraping were enough to capture the necessary key points. An army of people was necessary.
The solution: The expert established a network of universities to jointly collect, curate and make information available to the public. He also used data from intelligent products: The digitally networked Thermia clinical thermometer provided input. The intelligent chatbot Buoy, adapted to Covid (1,000 users per day), recorded symptoms and was then able to diagnose an infection with a high degree of certainty. The tool "Flue near you" - inspired by the film Contagion - records symptoms of people in the USA and thus makes the appearance and spread of the disease visible. But Brownstein did not stop at this point: He combined the collected data with demographic and regional information (e.g. school closures at the municipal level). This enabled him to better understand the effect of social distancing measures, for example.
Conclusion: AI can help to generate new data sources and supplement existing data sets. Among other things, this can help to determine the course of the pandemic and analyse the effect of countermeasures.
Data treasures: Reinterpreting known knowledge
The challenge: Stefano Rensi, Research Engineer at Stanford University, wanted to find out whether there were existing drugs that could be used in the fight against COVID.
The solution: He used existing literature on approved drugs and NLP to combine different data points: information on proteins, chemicals and genes. This resulted in the tool "Docs to Graph" which created a map that determined the relationships between the individual components. The expert was thus able to discover a promising combination that is currently being tested on mice in Japan.
Conclusion: AI can help to curate and reinterpret existing knowledge. Thus, medical solutions can be found and applied more quickly.
Four basic research questions can be derived from the approach of all experts:
- Which accessible data already exist? Is there any existing specialist literature, such as scientific publications or information on medicines?
- Can different data sources be combined to create new data sets, like social media and press releases? Or regional weather data and case numbers?
- What other networks or groups can you work withto achieve more quickly or to obtain data from different regions? For example other universities, schools or non-profit AI communities.
- Which digital or AI-based tools or products are already on the market and can contribute problem-specific information? In the case of Covid-19, for example, digital clinical thermometers, medical chatbots or language assistants.
The mentioned case studies only give a small insight into the concentrated knowledge that was shared in the virtual conference. I found the noticeable commitment of all scientists remarkable, as well as the personal involvement. The fact that COVID took the lives of a few researchers who were friends of mine only spurred the community of developers and specialists on even more.
Ultimately, data and artificial intelligence cannot defeat a virus such as COVID-19, but they can make its triumphal march more difficult or even drastically shorten it.
Photo by Erik Mclean, Unsplash