Evaluating the efficacy of hybrid deep learning models in distinguishing AI-generated text

Conference item

My research investigates the use of cutting-edge hybrid deep learning models to accurately di!erentiate between AI-generated text and human writing. I applied a robust methodology, utilising a carefully selected dataset comprising AI and human texts from various sources, each tagged with instructions. Advanced natural language processing techniques facilitated the analysis of textual features. Combining sophisticated neural networks, the custom model enabled it to detect nuanced di!erences between AI and human content. The results show a remarkable ability to distinguish whether the text belongs to a human or Arti"cial Intelligence (AI) technologies like GPT-3.5, GPT-4, PaLM 2 and LLaMa 2. Exhaustive metrics underscore the precision of these methods in pinpointing what or who created the text. It is bene"cial in an era where AI’s writing abilities increasingly resemble those of humans. These advancements in AI technology provide signi"cant advantages, such as enhanced e#ciency in content creation, the ability to generate diverse perspectives, and the capability to handle large-scale data analysis that surpasses human speed and accuracy. This evolution in AI writing tools is transforming industries, ranging from journalism to marketing, by o!ering scalable content generation and data interpretation solutions. Moreover, it’s fostering a more data-driven approach in decision-making processes across various sectors. This technological leap propels us towards new frontiers in AI authenticity. The results promise signi"cant applications from academia to media, emphasising the importance of ensuring content integrity. They underline the necessity for AI development to align with ethical standards for transparent creation and use of synthetic content. This study highlights the dual nature of AI text generation—its potential and risks—and calls for a commitment to responsible innovation as our reliance on natural language systems grows. The implications go beyond immediate applications, prompting re-evaluation of our interactions with and regulations for evolving AI technologies.

Authors: Oketunji, AF

• Publication status: Accepted
• Peer review status: Peer reviewed
• Publisher: Elsevier
• Journal: IFAC-PapersOnLine
• Publication date: 2023-11-23
• Acceptance date: 2023-11-22
• Event title: 17th Workshop on Discrete Event Systems (WODES 2024)
• Event location: Rio de Janeiro, Brazil
• Event website: https://wodes2024.eventos.ufrj.br/
• Event start date: 2024-04-29
• Event end date: 2024-05-01
• EISSN: 2405-8963
• Language: English