[Paper Review] A Survey on ChatGPT: AI-Generated Contents, Challenges, and Solutions

Yuntao Wang, Yanghe Pan+3 more|arXiv (Cornell University)|May 25, 2023

Artificial Intelligence in Healthcare and Education16 citations

TL;DR

This survey examines AI-generated content (AIGC) with a focus on ChatGPT, detailing working principles, security/privacy threats, IP protection, watermarking, and future research directions.

ABSTRACT

With the widespread use of large artificial intelligence (AI) models such as ChatGPT, AI-generated content (AIGC) has garnered increasing attention and is leading a paradigm shift in content creation and knowledge representation. AIGC uses generative large AI algorithms to assist or replace humans in creating massive, high-quality, and human-like content at a faster pace and lower cost, based on user-provided prompts. Despite the recent significant progress in AIGC, security, privacy, ethical, and legal challenges still need to be addressed. This paper presents an in-depth survey of working principles, security and privacy threats, state-of-the-art solutions, and future challenges of the AIGC paradigm. Specifically, we first explore the enabling technologies, general architecture of AIGC, and discuss its working modes and key characteristics. Then, we investigate the taxonomy of security and privacy threats to AIGC and highlight the ethical and societal implications of GPT and AIGC technologies. Furthermore, we review the state-of-the-art AIGC watermarking approaches for regulatable AIGC paradigms regarding the AIGC model and its produced content. Finally, we identify future challenges and open research directions related to AIGC.

Motivation & Objective

• Explain the enabling technologies and general architecture of AIGC and how ChatGPT fits into the AIGC ecosystem.
• Characterize security, privacy, trust, and ethics threats across the AIGC lifecycle.
• Review intellectual property protection, watermarking approaches, and regulation of AIGC content and models.
• Discuss future challenges and open directions for green, explainable, and trustworthy AIGC systems.

Proposed method

• Present a three-layer general architecture of AIGC-as-a-Service (infrastructure, engine, service layers).
• Provide a taxonomy of security/privacy threats across data, models, and usage, including countermeasures.
• Survey IP protection and watermarking techniques for AIGC models and outputs.
• Compare AIGC with PGC/UGC and discuss knowledge representation/usage shifts enabled by large AI models.
• Summarize enabling technologies (generative algorithms, pretrained large models, multimodality, RLHF) and working modes (assisted vs autonomous).

Experimental results

Research questions

• RQ1 What are the core working principles and architecture of AIGC and how does ChatGPT exemplify them?
• RQ2 What security, privacy, trust, and ethical threats arise in the AIGC lifecycle and what defenses exist or are feasible?
• RQ3 How can IP protection be achieved for AIGC models and produced content, and what are the related regulatory challenges?
• RQ4 What future research directions can ensure green, explainable, and regulatable AIGC services?

Key findings

• AIGC enables fast, high-quality, multimodal content generation at scale, driven by big data, big models, and big computing power.
• Security, privacy, trust, and ethical concerns are pervasive across the AIGC lifecycle, requiring diverse defense mechanisms.
• Watermarking-based IP protection for both AIGC models and outputs is a focus, with ongoing discussion of threats and countermeasures.
• AIGC architecture comprises a three-layer service model and supports ToB and ToC deployments with considerations for privacy and provenance.
• Two primary working modes exist—assisted and autonomous content creation—with implications for speed, cost, and content quality.
• The survey positions AIGC within evolving knowledge representation paradigms, highlighting shifts from databases and search to large AI model-based retrieval and generation.