Artificial intelligence (AI) holds significant potential for enhancing student learning. This reflection critically examines the promises and limitations of AI for cognitive learning processes and outcomes, drawing on empirical evidence and theoretical insights from research on AI-enhanced education and digital learning technologies. We critically discuss current publication trends in research on AI-enhanced learning and rather than assuming inherent benefits, we emphasize the role of instructional implementation and the need for systematic investigations that build on insights from existing research on the role of technology in instructional effectiveness. Building on this foundation, we introduce the ISAR model, which differentiates four types of AI effects on learning compared to learning conditions without AI, namely inversion, substitution, augmentation, and redefinition. Specifically, AI can substitute existing instructional approaches while maintaining equivalent instructional functionality, augment instruction by providing additional cognitive learning support, or redefine tasks to foster deep learning processes. However, the implementation of AI must avoid potential inversion effects, such as over-reliance leading to reduced cognitive engagement. Additionally, successful AI integration depends on moderating factors, including students' AI literacy and educators' technological and pedagogical skills. Our discussion underscores the need for a systematic and evidence-based approach to AI in education, advocating for rigorous research and informed adoption to maximize its potential while mitigating possible risks.