The Reliability and Limitations of AI Content Detection Tools in Academic Writing" How reliable are AI

 "The Reliability and Limitations of AI Content Detection Tools in Academic Writing"


How reliable are AI detector software such as Originality.ai, GPTZero, ZeroGPT, Scribbr, QuillBot, Grammarly, and JustDone AI? Their produced results are inaccurate in most cases. The scores of AI detection they show in the same content vary from one AI detector tool to another. They always use the words "possibility" or "likely" in the case of any piece of writing, whether it is AI-generated or not. But they fail to ensure the portion of AI-generated content.


If one checks any piece of writing using those AI detector software, he finds different AI-generated ratios of his content, which prove no certainty of their produced results. For example, they detect citations made by one as a part of AI-generated content, which is undoubtedly a false detection. In every research paper, citations remain added but seem to be AI-generated to AI detector software. Human-written sentence patterns seem to be AI-generated to AI detector software.


In the age of AI, many words like interwoven, intertwine, instill, underscore, intricate, interplay, notion, paradigm, domain, trajectory, realm, contextualization, lens, spectrum, encapsulate, redefine, transcend, rooted in, stem from, inclusive, reconfiguration, etc. are highly used in academic writing. This tendency is called AI-induced lexical convergence, according to philologists in the age of AI. AI detector tools detect those words as AI-generated; subsequently, the scores of AI-generated content appear to be high.


However, the reliability of AI detector tools demonstrates very low accuracy, in particular JustDone AI, GPTZero, ZeroGPT, QuillBot, and Grammarly; but Originality.ai and Scribbr are relatively reliable. Some renowned universities like Stanford University, Harvard University, and Princeton University have conducted research on AI detector tools. Those studies have shown that error rates are high (20%–60%). No renowned universities like MIT, Cambridge, Oxford, and others have given recognition to AI-detector scores as valid due to their scientific unverifiability.


©দ্বীন সাঈদীন

Popular posts from this blog

Mental Sickness

E-learning

Imperialism VS Terrorism

At Sixes And Sevens

Bilateral Coherence and Dissonance Relationship between Western and Bangladeshi Culture

Mentalism:07

Shrewdness

Oppression upon Women Folks