ADM Systems in the COVID-19 Pandemic: Spain

New report: 'Automated Decision-Making Systems in the COVID-19 Pandemic: A European Perspective' by AlgorithmWatch and Bertelsmann Stiftung – a special issue of the upcoming Automating Society Report 2020, to be published in October.

Country analysis: Spain

By Jose Miguel Calatayud

A controversial self-diagnosis application

On 18 March 2020, the Madrid regional government launched a website including a Covid-19 self-diagnosis application based on a simple algorithm composed of eight yes-or-no questions. Each question was assigned a number of points, and if a user got 30 or more points, the application told them they might be infected with the Covid-19 virus and what they should do next.

The application had been developed by the Madrid government and several private companies, including Google, Telefónica, Carto and Ferrovial. Its stated aim was to free up the emergency phone lines and to provide the authorities with information to manage the crisis situation, including a first assessment of individuals who might need medical assistance and follow-up.

To use the application, people had to submit quite a few personal details, including the national ID number, full name, birth date, full residential address and email address. Then the questionnaire asked about symptoms the person might have.

The privacy policy stated that the collected data could be shared with the national security forces, the judicial system and all the companies acting as suppliers or working with the Madrid government, including those acting as subcontractors. The data was to be stored and processed for statistical aims and for biomedical, scientific or historical research; and it would be “deteleted, anonymised and/or blocked” when “the period of keeping the data finishes, and according to the requirements established in the applicable norm”, without any other specification on when exactly that would be.

This lax privacy policy was criticised on social media and in some media reports, and when on 22 March the Madrid government released the Android and iOS mobile versions, the application no longer asked for the user’s email address and the terms and conditions and the privacy policy had been updated.

Now they specify that the data would be used both to describe the pandemic and to predict how it might evolve, and added that the companies would only get “temporary” access to the data under instructions by the authorities, and that the companies would not be allowed to use the data for their own aims. Users could opt out of giving their phone GPS location, but if they chose to do so, the application would use their residential address for geolocation purposes.

The same results, just without personal data

Shortly afterwards, the Spanish national government released its own version of the web and mobile applications, based on the code of the one developed for the Madrid region. In its privacy policy, this version detailed that the data would be stored for a maximum of two years, and that it would not be shared with any private companies but only with Spanish national and regional administrations, and with international authorities if needed.

Several other regional governments within Spain released their own applications, in some cases also based on the code of the Madrid one. Interestingly, an independent developer released an open source version of the Madrid application, using the exact same algorithm and therefore producing the same output, that people could use without giving any personal data.

Predicting the lifting of lockdown restrictions

On 14 April, the Spanish government announced a new study, run by the National Scientific Research Council (CSIC), that would gather and analyse mobile phone, map servers and social media data to predict different social distancing scenarios and help in the decision of when, where and how to start lifting lockdown restrictions. The data is to be collected by the operators and companies taking part in the project, and the announcement said that no information that could identify individuals would be accessed.

The project was described as using “artificial intelligence tools and data science, and (integrating) big data in real time on human mobility, geo-localised surveys and computational models”, without going into more detail about what all that exactly meant.

The announcement also stated as a long-term goal “to establish the basis for a computational epidemiology network in Spain, as in other countries, and a series of interoperable analytical tools based on epidemiological theories, data science and artificial intelligence, to inform the decision-making process in future situations of epidemiological crisis”. No further information on which precise data would be collected and on how it would be processed could be found.

Temperature screenings for the basketball league

From the 17th to the 30th June, the Spanish professional basketball league held its playoffs in Valencia. The ACB, the sports association that manages the league and is made up of the top-tier 18 clubs, set up an ADM system to check people’s temperature before entering the premises where the matches took place.

Developed by Valencian company Sothis and known as Thermal Vision System, the device measures the person’s tear conduct temperature (which is supposedly very accurate to determine’s someone’s temperature) remotely by combining a “thermographic camera and artificial vision”.

The company says the system gets the temperature in less than a second and with a margin of error of ±0.3º. In the basketball playoffs, the limit for a person to be allowed in was set at 37º. The whole process is automatic and the inner working of the system is not public. Reportedly, the whole exercise is seen as a pilot in Spain and in the future could be used in other public events.

Promising results from decentralised contact tracing app trial

Lastly, Spain announced its national contact tracing app, “Radar COVID”. Available for download in late August, it is based on the Apple/Google decentralised Bluetooth protocol. The app has been trialed on La Gomera, an island in the Canary archipelago, where it has been downloaded some 60.000 times (against an initial objective of 3.000), with interesting results.

Over the one-month experimentation, concluded on July 31, the app has been "twice as effective as human tracers” in the pilot, simulated outbreak, Reuters reports. According to a statement by Carme Artigas, head of the state digital and artificial intelligence unit, “for every virtual positive diagnosis, the app identified an average 6.4 contacts with others (…), compared with an average 3.5 contacts identified by human tracers in the Canary Islands”.

Download the full report as PDF (1 MB).

Automated Decision-Making Systems in the COVID-19 Pandemic: A European Perspective is a special issue of the report Automating Society 2020 by AlgorithmWatch and Bertelsmann Stiftung, to be published in October. Subscribe to the our newsletter to be alerted when the report is out.