Two media update this system used since 2018 in the country, having potentially pushed employees to return to work despite a medical situation that did not allow them.
Spain has set itself an objective to fight against health insurance fraud, targeting in particular employees who provide sick leave when their medical situation allows them to return to work.
Since 2018, the country has thus had an algorithmic system to detect such cases. A system which, for five years, has remained largely unknown despite the wish for transparency stated by the country, and its leading position in the use of “reasoned” artificial intelligence.
A collaboration between the journalistic investigation unit Lighthouse Reports and the Spanish media The confidential has made it possible to update these files which have remained secret until now.
Opaque system
Concretely, the artificial intelligence used by the Spanish government aims to decree whether or not a person on sick leave is able to return to work. A complex decision, which involves sensitive personal and medical data, and an algorithm probably decried by the experts interviewed by the two media.
“The algorithm determines which work stoppage files are examined first by Social Security (…). It is a sophisticated computer program that defines the order in which each employee must consult a doctor to find out s “he can continue to receive the subsidy (dedicated to workers on sick leave, NLDR). The system identifies who must return to work and, if the discharge has not yet been processed, it flags the file as potential fraud”, describe The confidential.
The documents and information collected by The confidential et Lighthouse Reports “reveal an opaque and unprofitable system, which makes decisions whose stakes can concern millions of citizens, and which can potentially push patients who are not ready to return to work”, indicates Lighthouse Reports.
The system consists of two steps, both at the heart of an application used by Spanish Social Security referring doctors, which analyzes patient-related information. At the end of these two stages, the medical file receives a score ranging from 0 to 1, the number 1 representing the capacity for the patient to return to work.
“Poor” and “unbalanced” algorithms
Problem: Experts interviewed by The confidential et Lighthouse Reports alert on the real efficiency of the system. Like Ana Valdivia, researcher at King’s College (London) and teacher of artificial intelligence at the Oxford Internet Institute, who describes “poor” and “unbalanced” algorithms.
An anonymous doctor, affiliated with the Spanish Social Security and with about fifteen years of experience, affirms that “it is an opaque program, of which those who work there every day are not able to explain what it is about. ‘acts’. Same story for the six other doctors and Social Security inspectors interviewed.
As for the data used – medical and personal – to operate this system, if The confidential et Lighthouse Reports were able to obtain a fairly precise list, the ministry did not provide information on the weight of each piece of data in the final calculation.
Finally, The confidential highlights two of the main shortcomings of the situation: the processing of sensitive medical data, which must be handled with the legal precaution in force, and algorithmic biases, which can put certain categories of people at fault. Two questions that have remained unanswered by the Spanish Ministry of Social Security.
