Artificial intelligence for social distancing


By Florence Véronneau-Veilleux

Image Source : Pixabay


In an effort to control the current Covid-19 pandemic, several companies and research institutes have developed applications using artificial intelligence to help maintain social distancing. How do these applications work?


The end of Covid-19 lockdown led to the implementation of various measures of social distancing and prevention. Different mobile applications have been developed in the last few months to help citizens maintain these distancing rules. The Quebec Artificial Intelligence Institute (MILA) proposed an application called COVI. The algorithm behind this application would calculate the probability of being infected by Covid-19. To be more user-friendly, its users would receive recommendations adapted to the value of their probability: get tested, aks for delivery at home, avoid going out in public places, etc.



A data-driven application


Data are needed in order to make proper recommendations. Some of these data may come from the user’s phone Bluetooth. Phones can recognize each other when they are close using Bluetooth. Thus, when a person receives a positive diagnosis, a warning would be sent to those who have encountered him/her. The COVI application, however, planned to use more data, such as the health status of users, their characteristics (age, sex, …) in addition to their interaction with others through Bluetooth.


These data are used as input for the artificial intelligence algorithm within the COVI application. The idea behind such algorithms is to take input data and then find the links leading these data to an output conclusion. These links are very difficult or even impossible to identify by a human being when there are several input data.


Let's take a simplified example: a 70-year-old asthma patient with little outside contact. The algorithm will link age and asthma to being vulnerable to Covid-19 and will associate little social contact with a lower probability of having contracted the virus. By weighting these factors, the output conclusion is likely to be to limit outdoor activities without needing to be tested for Covid-19. These links are found in the algorithm through different equations.


Source Image : Florence Véronneau-Veilleux



Prior learning required


But how did this algorithm find these links and establish the relative importance of the different factors in the first place? It must be trained with data whose conclusions are already known. The algorithm can then test different links (several thousand hypotheses!) until it finds the right ones, those that link the input data to the right output conclusion. Thus, when the algorithm encounters, during its training, several data from patients over 70 years old with the conclusion to avoid outdoor activities, the algorithm "learns" to link these two elements. The MILA team had planned that some users could provide their data voluntarily for the training of the algorithm.


Ethical questions were raised regarding the protection of personal data in the event of COVI application implementation. The implementation of the COVI application will therefore not go forward. Other applications for Covid-19 have been developed but have not yet been implemented in Quebec. This remains a matter to be followed!


24 vues
  • Facebook
  • LinkedIn
  • Instagram
  • Twitter

©2020 Fatéma Dodat. Tous droits réservés.