Generative Al: A Survey of Recent Advances and Applications

Abstract

One of the latest subfields of artificial intelligence that has become one of the most disruptive ones is generative artificial intelligence (Generative AI), which can generate very realistic and contextually relevant content that is multimodal, i.e., it can be generated in the form of text, images, music, and video. In contrast to the classic AI-based solutions, whose core task was to divide and analyze data, generative AI is aimed at producing completely new content, which in many cases can hardly be distinguished from human content. This is going to be a thorough survey of how this idea of generative AI has been progressing, beginning with the theoretical principles of symbolic AI and early work on machine learning, to propose relatively modern techniques of deep learning systems. Within the framework of the paper, the basic generative models, including generative adversarial networks (GANs), variational autoencoders (VAEs), Transformer-based models, and Diffusion Models, are discussed thoroughly with the emphasis on the specificity and peculiarities of their structures as well as their advantages and drawbacks. We go deeper into the newest technologies of the training methodology and scalability factors, multimodal learning, as well as into the open-source technologies, which have provided a chance to democratize access to the most powerful tools. The survey spans a wide variety of applications, including natural language processing (NLP), computer vision, creative arts, games, healthcare, and scientific research. It also provides a weathervane indicating how geographically widespread the field of generative AI applications is. The mentioned critical ethical considerations are bias, fairness, intellectual property issues, deepfakes, and misinformation, privacy concerns, and environmental footprint. Major limitations like data needs, computational complexity, assessment challenges, and text hallucination are addressed together with the future expectations of cross-disciplinary incorporation, robotics usage, model explanations, regulating methodologies, and democratization endeavors. The present survey is targeted at helping researchers, practitioners, and policymakers obtain a complete picture of the generative AI ecosystem and its capabilities of transforming human-computer collaboration.