International Journal of Scientific Research and Engineering Development

Compliant, or obedient behavior, is a feature that will be required in intelligent systems operating in chaotic and/ or fast-moving environments where humans are not present. Recent machine learning breakthroughs along with the rise of generative models (e.g., VAE, GAN and diffusion model) suggest that decision-making autonomy can be achieved qualitatively by relying on better data representations, uncertainty estimates and scenario generation. This paper offers an in-depth perspective on generative models for autonomy. We review the theoretical connections between generative modelling and decision making, with an emphasis on probabilistic reasoning and policy optimization. Leveraging curated datasets and simulated environments, we use the combined architecture with decision system to deploy and evaluate several generative frameworks on their ability of accuracy of inferences, adaptation dependence, computational complexity and scalability. We compare the two approaches and identify their merits and drawbacks, which may present valuable references for implementation in real-world intelligent systems. Finally, the results emphasize the usefulness of generative models in achieving hierarchical, flexible and effective autonomy for ML decision-making systems. The presented work draws on the void from what generative modelling offers to autonomous system design, and prospects towards enhanced decision autonomy in the future generation of AI technologies.

Jaya Asthana, Gaurav Goel, Shalini Raghuvanshi,"Generative Models for Autonomous Decision Making: A Machine Learning Perspective" International Journal of Scientific Research and Engineering Development, V9(3): Page(20-33) May-June 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.