ADVERTISEMENTIn 2018, the Academy Award for Best Documentary Feature Film went to “Icarus,” directed by Bryan Fogel.
The film, which originally set out to investigate doping in cycling, concludeed up uncovering one of the hugegest sports scandals in recent history: a state-sponsored doping programme carried out by Russian authorities for the 2014 Sochi Winter Olympics.
The revelations came from Dr Grigory Rodchenkov, head of Russia’s national anti-doping laboratory during the Olympic Games. He revealed how anti-doping officials and ininformigence agents had systematically replaced urine samples containing performance-enhancing drugs with clean samples collected months earlier.
Dozens of athletes were part of the programme. None were caught, and in 2014, Russia topped the podium with 13 gold medals. It was stripped of several medals when the doping scheme was created public.
But why was this possible in the first place? How does doping detection actually work? And can artificial ininformigence (AI) create it more effective?
To answer these questions, Euronews Tech Talks interviewed Wolfgang Maass, a professor of business informatics and computer science at Saarland University in Germany and Francesco Botrè, Director of the World Anti-Doping Agency (WADA), an accredited anti-doping laboratory in Rome.
How does doping detection work?
Botrè explained that doping detection relies on two approaches: markers of exposure (direct detection) and markers of effect (indirect detection).
“Markers of exposure mean finding the bullet; markers of effect mean finding the wound or the scar, and stateing ‘this could only be done by a knife,'” he stated.
In practical terms, this means that anti-doping officials either detect the banned substance itself or identify biological alters consistent with performance-enhancing drugs.
Detection can involve either a single test or longitudinal monitoring, depconcludeing on whether the athlete is tested once or multiple times.
For longitudinal analysis, athletes’ data are stored in the Athlete Biological Passport (ABP), which supports officials to monitor unusual biological variables over time.
Can AI support detect doping?
“What AI in principle can do is to find patterns that are hard to detect by humans,” Maass stated. That’s where his team’s AI models come in.
Maass launched experimenting with anti-doping machine learning systems around 2016 and has since collaborated with WADA on various projects.
His technology aims to address two major challenges in doping detection: identifying EPO (erythropoietin), a hormone that stimulates the production of blood cells, and preventing sample-swapping incidents such as the one that occurred in the Sochi Winter Olympic Games.
To enhance EPO detection, Maass’ AI model analyses not just isolated data, but the athlete’s entire metabolic pathway. This allows the software to compare data with the athlete’s unique biochemical profile, assessing whether anomalies suggest the utilize of EPO or are simply the result of some personal natural characteristics.
Similarly, to detect sample swapping, the AI system compares the athlete’s current test data with their historical biological profile to detect inconsistencies over time.
According to Maass, these technological tools are not only effective but also quicker and cheaper than the methods and machines currently employed in laboratories.
Botrè also displayed great enthusiasm for the development of AI. He stated AI tools could support interpret the combined effects of multiple doping substances in an athlete by analysing large amount of data and identifying patterns.
According to him, this could provide insights into how performance-enhancing drugs mix, a topic that cannot be studied through traditional research becautilize it is considered unethical to inject or deliberately administer multiple banned substances to a person for scientific purposes.
The real challenge: Data and funding
When questioned about the hurdles in developing AI programmes for doping detection, Maass pointed to two key issues: access to data and funding.
As Botrè noted, it is difficult to do studies on doping while respecting ethical and health security criteria, so accessing the necessary data to train an AI model is very complicated.
Both Maass and Botrè believe athletes who utilize performance-enhancing drugs are likely already applying AI to test to beat existing doping detection systems.
“If we have a new method, and we have to publish it, we have to wait for the feedback of the scientific community, and we have to wait for their approval by WADA. In the meantime, they are doping. And if they find a way to cheat this method, well, they do not publish it, there is no Journal of Doping Science,” Botrè stated.
“Doping is a dirty matter, but we have to fight it with clean hands,” he added.
Leave a Reply