At regional and national levels, policy makers urgently need the best scientific information in order to make informed policy decisions. When should schools be re-opened? What proportion of the population is wearing masks in public? Useful data can come from various sources (scientific articles and preprints, ‘grey literature’ reports, social media, and more) but the quantity is huge, and the quality varies widely. AI technologies can help to filter the most relevant results, but the amounts of information are often still overwhelming, and require human intelligence to further extract the most essential facts for making policy decisions.
Over two recent hackathons, the Swiss VersusVirus (4-5 April 2020) and the Europe-wide EUvsVirus (24-26 April 2020), a team of citizens and scientists from research institutions in Switzerland (UZH and ETH), France, Italy, Spain, and the UK, has hacked to find better solutions to this problem.
The aim of the CrowdvsCovid challenge was to provide policy makers with relevant and actionable information on a range of Covid-related issues, as quickly and reliably as possible. The idea is to combine the speed of machines with the intelligence of the crowd by pipelining automated filtering (AI algorithms) with crowdsourced validation and refinement of the resulting data, to ensure the retention of only the most relevant results for health experts, epidemiology researchers, and policy makers.
To achieve this, the team has been “beta testing” the Citizen Science Project Builder, a DIY crowdsourcing platform developed by the Citizen Science Center Zurich in collaboration with the Citizen Cyberlab in Geneva (stay tuned for more news on the platform soon!)
In the space of a weekend, the team implemented in the platform two demos.
DEMO 1 – POLICY INTERVENTIONS
The first demo focuses on reviewing scientific articles (pdf files) and extracting key information based on specific policy concerns. For example, as European countries open up schools, policy makers want to know what the experiences have been in countries that have already opened schools or did not fully lock down, as well as experiences from other public health crises that may be relevant to this issue.
The projects examines relevant articles selected using machine learning algorithms, that are presented to the crowd for validation and data enrichment.
DEMO 2 – SOCIAL DISTANCING AND MASKS
The second demo uses social media data from twitter, focusing in particular on images. Especially during a lockdown, where standard survey techniques become impossible, analysis of social media provides useful insights, for example by analysing the types of masks that people are wearing in public places, and the compliance to measures such as social distance.
In the project, the crowd validates and classifies the images associated to tweets selected with machine learning algorithms developed by Politecnico di Milano.
The results of the crowdsourcing demos, which use a relatively small data sets of about a hundred publications and 1000 social media images, where used to develop a full pipeline model, going from the raw data input to policy advice output.
If you want to more about the CrowdVsCovid work, check their Devpost page.