Pregledni rad
https://doi.org/10.31337/oz.81.1.5
Challenges of Explainable Artificial Intelligence in Healthcare Decision-Making Processes
Luka Poslon
orcid.org/0000-0002-7389-7694
; Hrvatsko katoličko sveučilište, Laboratorij za etiku digitalnih tehnologija u zdravstvu (Digit–HeaL), Zagreb, Hrvatska
*
* Dopisni autor.
Sažetak
In this paper the author draws our attention to the potential and the challenges of
the application of artificial intelligence in healthcare. Artificial intelligence offers advancements
such as better diagnostics and personalized treatments, but also raises
ethical challenges related to explainability, bias, and trust. A key challenge is the
“black box” phenomenon, where artificial intelligence predictions lack transparency
and cannot be easily explained. To address this, explainable artificial intelligence
aims to make predictions explainable and foster trust in use of artificial intelligence
in healthcare. However, this approach faces its own hurdles, such as balancing between
accuracy and explainability, the lack of standardized tools for measuring ex-planation quality, and under–researched relationship between bias and explanations
in automated systems. Understanding these limitations is vital for the responsible
use of artificial intelligence in high–risk healthcare settings. Special attention is given
to the overconfidence bias, where artificial intelligence systems or their users may
overestimate the reliability of outputs. The paper proposes an original explainable
artificial intelligence method designed to detect and mitigate the effects of overconfidence
bias, thereby reducing decision–making errors. By identifying such risks and
addressing them through tailored an explainable artificial intelligence approach, the
paper contributes to the development of trustworthy artificial intelligence systems in
healthcare.
Ključne riječi
artificial intelligence; explainable artificial intelligence; healthcare; medicine; decision–making; ethics; bias
Hrčak ID:
341006
URI
Datum izdavanja:
7.1.2026.
Posjeta: 870 *