By Sebastian Jon Holmen, Thomas Søbirk Petersen and Jesper Ryberg.
Artificial intelligence (AI) is now widely employed by both private and state actors to aid them in attaining their goals more efficiently. AI is, for example, already being employed to catch tax evaders and to identify persons at risk of developing certain types of cancer, while students all over the world are using AI chatbots to help them study (or to have it do the studying for them). The field of sports is not exempt from this development, with athletes using AI to, among other things, optimize training and prevent injuries. Furthermore, it has been suggested that AI could and should be employed to more effectively catch those athletes that use doping to gain an advantage over their competitors. Our paper considers the ethical viability of this suggestion.
Specifically, we start from the intuitively plausible observation that if doping detection by means of AI would be more efficient than doping detection without the use of AI, then the former is more morally desirable than the latter. Surely, if it is good to catch one athlete that uses doping, then it is even better to catch two. But in our JME paper we argue that on closer inspection, things are in fact a lot more complicated. First, it is not clear that it is better to catch more athletes that use doping – rather, this depends on whether the current doping regulations are morally defensible, and as we argue, it is far from clear that this is the case.
The second main point we make in the paper is that even if we assume that the current doping regulations are in fact ideal, and even if a doping detection system using AI is more accurate (i.e. generates fewer false-positives or false-negatives results relative to systems not employing AI), there is still reason to be sceptical as to whether such a system is morally desirable. This, we suggest, is related to the fact that the moral costs of the different types of errors means that a more effective doping detection system could be considerably more harmful to athletes than a less effective system, even if there is a net reduction of such errors.
Our hope is that the paper shows that the question of whether anti-doping organisations around world should use AI to combat doping is much more complex than has hitherto been supposed. Indeed, the observations made in the paper are relevant not only in relation to the ethics of doping detection, but to discussions of the ethical desirability of employing AI in any situation where its use is defended by claims of enhancing effectiveness. That is, we should remember to ask critical questions about whether the activity that AI promises to make more effective is in fact a morally desirable one. Further, and for most domains more importantly, we should recognize that the error-profile of AI may make it morally undesirable to employ even if it does reduce the total number of errors.
Authors: Thomas Søbirk Petersen, Sebastian Jon Holmen, Jesper Ryberg
Affiliations: Philosophy and Science Studies, Roskilde University, Roskilde, Denmark
Competing interests: None declared