Mila > COVI

COVI

A risk prediction app against Covid-19

Project Description

COVI is a research project which resulted in the development of an AI-enabled health mobile application to empower citizens in their fight against the COVID-19 virus. 

Made in Canada, the app was created to help citizens make better informed decisions about their actions to reduce risk, while preserving individual privacy. The COVI app draws on epidemiology, behavioral psychology and artificial intelligence to propose an innovative approach to contact tracing mobile technology. 

The design and development phase of the project has been finalized. Our codes and models are available in open source for academic consumption and accessible to states should they wish to deploy an AI-enable health app inspired by our approach. 

Project Overview

Benefits

The main high-value benefits of an AI-enabled app are:

  • Informs users to self-isolate or get tested if they have been exposed to someone who later tests positive to Covid-19. By synching with provincial test results databases, this can all be done automatically, thus reducing the burden on the health system. 
  • Predicts daily personal risk factors for each user, based on demographics, behaviors, health conditions, symptoms and contacts. This goes beyond the traditional notification apps thanks to the AI tool that is able to consider a wide range of variables to assess the probability that a specific individual is carrying the virus, ahead of any visible symptoms.
  • Translates public health guidelines into daily personal recommendations. As Public Health authorities adapt guidelines in various phases of confinement and deconfinement, it is challenging for citizens to make sense of “what-does-this-mean-for-me-today.” The app is designed to bridge this gap by translating the personal risk factor into concrete recommendations, e.g. taking public transport, going to work, socializing with friends.
  • Reduces the overall number of cases. Thanks to AI predictions, users are notified to adjust their behaviours and reduce the number of interactions as soon as their risk score increases. This can be days ahead of any symptoms and/or of being informed of a COVID positive person in their network. 
  • Provides early warning signs of outbreaks. Thanks to the aggregate risk data computed, the app can support population level public health by flagging at-risk communities or regions ahead of the actual number of case numbers starting to increase. This would provide early warning signs that an outbreak is going to happen and lead to very targeted interventions such as doing community screening in specific regions with higher risk or local level lockdowns.
  • Helps defines new epidemiological models. The data shared by the app users enables an artificial intelligence tool to identify new patterns and specific parameters regarding how the virus spreads (e.g. distance, time, sex, age). This allows us collectively to get smarter about Covid-19 and enables public health authorities to make better informed decisions based on evidence.
  • Predicts the impact on the contagion curve. Public Health interventions can be simulated to assess the probable impact on viral spread. Given its knowledge about the virus at a regional level, the AI model can predict what will happen, stress-testing and optimizing of deconfinement strategies.

The COVI app was designed as a privacy-conscious app that meets the highest standards of data privacy and protection. The system was designed to never directly identify a user and make all the information exchanged encrypted. The system would minimize the collection of data and facilitate its deletion. Data would be stored on users’ phones with their consent, and any data more than 30 days old gets automatically deleted. Users would be offered the option of consenting to sharing data with COVI to build the model and improve its risk calculations. No data would be kept longer than 3 months, and consent can be withdrawn at any time. The project complies with Canadian privacy laws and will follow all applicable recommendations put forward in the Joint Statement by Federal, Provincial and Territorial Privacy Commissioners on May 7, 2020.

Resources

COVI mobile application code
Read
Risk model training code for Covid-19 tracing application.
Read
Covid-19 spread simulator with human mobility and intervention modeling.
Read
Here, we present an overview of the rationale, design, ethical considerations and privacy strategy of ‘COVI,’ a Covid-19 public peer-to-peer contact tracing and risk awareness mobile application developed in Canada.
Read

Partners

The COVI project is led by world-renowned AI researcher Yoshua Bengio at Mila and brings together a coalition of Canadian researchers and experts, combining research from public health, epidemiology, privacy, machine learning (ML), ethics and psychology.

Developed by

Partners

Mila Team Members

Core Academic Member
Scientific Director, Mila & IVADO, Full Professor, Samsung AI Professor, Université de Montréal, Canada CIFAR AI Chair
Vice-President, Policy, Society and Global Affairs, Leadership Team
Core Academic Member
Full Professor, Université de Montréal, Canada CIFAR AI Chair
Core Academic Member
Associate Professor, HEC Montréal, Canada CIFAR AI Chair
Student Ph.D., École Polytechnique de Montréal, Université de Montréal
Student Ph.D.
Intern, Intern MSc
Student Ph.D. - University of Oxford, Alan Turing Institute, Alumni
Affiliate Member
Assistant Professor, University of Toronto
Student Ph.D., Université de Montréal
President and CEO - Mila, Leadership Team, Observer, Board of Directors
Student Ph.D.
See all professors