Alzheimer’s disease (AD) is a progressive cognitive and neurological disorder that affects millions of people worldwide. Retinal imaging has emerged as a cheap, non-invasive biomarker for AD, using which artificial intelligence (AI) models can be trained to streamline AD screening and diagnosis. Given the sensitivity of AD diagnosis, the lack of interpretability of conventional AI models, such as convolutional neural networks (CNNs), is an important barrier to their practical application. In this context, the Kolmogorov-Arnold network (KAN) architecture provides an intrinsically interpretable alternative to traditional deep learning models. Using KANs, we propose an interpretable AI model for AD detection that achieves an accuracy of 87%, outperforming state-of-the-art models by margins of more than 6%.