The systematic review investigates between 2020 and 2025 meta-cognitive mechanisms that enable Artificial General Intelligence systems to achieve adaptive reasoning capabilities. Our research team performed a controlled literature search according to PRISMA 2020 guidelines by searching Scopus Web of Science IEEE Xplore and arXiv which produced 2845 records but only 103 studies fulfilled the requirements for final analysis. The selected works were classified into five themes: self-monitoring architectures, uncertainty modelling, adaptive control systems, reflective reasoning frameworks, and evaluation alignment studies. The quantitative synthesis shows that research publishing activity increased by 3.2 times during the review period with reinforcement learning and neuro-symbolic architectures being the two most common types of publications. The reviewed methods show token-efficiency improvements between 30 and 78% together with accuracy gains between 5 and 14% across the selected benchmark tests. The comparative analysis demonstrates that interpretability and scalability and computational cost and integration complexity create systematic trade-offs. The review shows that standardized benchmarking together with uncertainty calibration and scalability issues and human cognitive offloading create ongoing difficulties. It further outlines priority research directions focusing on reproducible evaluation protocols, scalable monitoring mechanisms, and human-in-the-loop reasoning systems. Overall, the findings indicate that meta-cognitive control mechanisms are emerging as an important design principle for enabling efficient, reliable, and adaptive reasoning in AGI-oriented architectures, informing the design of AGI systems, evaluation pipelines, and deployment strategies in reliability-critical, resource-constrained settings.
Bohra et al. (Sun,) studied this question.