ai writing detection tool

ai writing detection tool

AI Writing Detection Tool

1. Introduction

Creating an AI writing detection tool is valuable in the educational sector in order to make the development of the next generations more efficient. This tool can be used by teachers, educational institutions, and AI writing developers among others. Users can ascertain whether submitted documents are valid original work or if AI-generated content was employed. The tool can also serve as a means to investigate techniques used by students, surfacing the most critical transcripts requiring the foremost attention. It also allows teachers to assess whose writing exceeded real human capacities, whereby once a certain threshold has been exceeded, it signifies required consultation. Additionally, upon completing this project, code and other methods tested may serve as a preliminary inspiration to assist future researchers solve similar discrimination tasks.

The goal of this project is to identify state-of-the-art models and methods that serve as a basis to create an AI writing detection tool. The tool is intended to provide teachers with a valuable assistant in detecting AI-generated writing. This task places emphasis particularly on creating models in the categories of text generation while being practical. Upon investigation, the best-suited method, in terms of ease of applicability, efficiency, effectiveness, and overall practicality, was the one based on a recurrent neural network (RNN). This research aims to elucidate how effective a basic RNN model is at detecting AI-generated writing and, in one of the subtasks, who will detect it first. In the main task, the output assigns input writing into two categories: the first of which represents generated AI writing and the second category consists of all other writing. Several RNN models of varying order were tested.

2. Benefits of AI Writing Detection Tool

Undoubtedly, the use of writing detection technologies for the purpose of verifying the authenticity of a written text does not cover all types of content generated by deep learning: for instance, audio, video, numerical data, etc. In addition, technologies that are now highly advanced and capable of creating different types of media will emerge just as they did for writing. Nevertheless, the value of the outcomes demonstrated in this article is still considerable, given that written language remains the primary medium for formal communication: research works, project reports, technical evaluations, past evaluations, heads-ups, or any other type of formally written communication in natural language publicly supposed to propagate credible content.

In today’s world, it is essential to safeguard the integrity, authenticity, and ethical compliance in various communication mediums. Written content available in digital public repositories was not subject to systematic and automatic scrutiny. The AI writing detection tool exhibits a reliable analysis evaluating the likelihood of the dissemination of artificially generated content. The AI writing detection tool is intended to provide a quick rating to determine the authenticity of written content across all social mediums. This tool will guide the legitimate use of artificial intelligence in professional communication, something that will come to be an integral part of multiple human activities and interactions.

3. Features of AI Writing Detection Tool

This work describes an AI writing detection tool that has been developed and intended to be used to assist the research community in reviewing papers before formally submitting them for review to a scientific conference or to a scientific journal. Papers will be anonymized and checked with the JEMS application to detect text fragments that have excessive similarity to text fragments of papers retrieved from diverse sources. Additionally, to excessive similarity detection, JEMS also pinpoints text matches within different sections of papers and text that is hidden within image artifacts, for example, from pasted images or from adjusted figures.

Authors and other stakeholders of publishing research are often put to the test of detecting which content has excessive overlaps with other works for different reasons. In the case of authors, excessive overlaps are cause for concern when submitting papers, while stakeholders of publishing research also need to verify submitted papers. In recent years, authors have had to deal with the incredible level of pressure to document new or contradictory work that has emerged only within a specific period of time. This situation prompts us to be vigilant for plagiarism, but even then, due to different styles of writing, detecting excessive overlaps in a manual way is quite tedious. Along with the ever-increasing quantity of available papers, this issue is becoming harder to manage.

4. How to Use AI Writing Detection Tool

When you first go to your dashboard, you’ll see a summary of recent activity within your classroom. If a student has submitted a paper, it will be tagged “Submitted” and you’ll see how many “Document Views” or “Inspection Attempts” have been performed. You may also see the status of the document as it processes within the queue here. Every document submitted to the AI Writing Detection Tool goes through an office inspection process. The dashed line underneath each document represents the processing stage of a document. This process involves examining the text for matches against a comprehensive database of the world’s journal articles, periodicals, webpages, and other students’ papers. After a document has been submitted, it typically takes as much as 30 seconds for the initial office inspection process to complete. Upon successful completion of the processing stage, the document will be scanned for matching text for plagiarism and analyzed for likeness in language and style to other author content.

When you first log on, you’ll be greeted by a friendly onboard. It’ll ask you to create a new classroom and give it a name. You’ll then be asked to invite students. It’s easy to create a new classroom. Just type the name of your classroom in the box, and then click on “Create”. Next, you’ll need to invite your students to join your classroom. You can do this by typing in their name and email address one by one… or you may invite many students at once with this handy feature. Just upload your CSV or Excel file with your student information, and you’re done! After you have created and invited students to your classroom, it’ll be ready to use. As students submit documents, these documents will be displayed on the “Recent Activity” page. They will also be available to view on the classroom dashboard.

5. Conclusion

Lastly, we showed the inadequacy of traditional closed data for evaluating tools on AI in the light of binary-oriented measures. We hope the improved metrics we put forward can help future evaluations of AI writing detection methods. We believe this area needs to be much more investigated, especially when AI and adversarial learning (AI-resistant text transformation) will make AI-generated papers become closer to actual authored documents from the perspective of text. As a result, the number of rejections of manuscripts of good quality but humanly flawed goes up due to AI incorrectly identifying them as detrimental documents.

We presented an AI writing detection tool, DeTAS, which is based on BERT and is capable of clearly detecting AI writings with a performance (0.89) in line with classical plagiarism tools. However, DeTAS considers a distinct set of papers as AI, which has facilitated a fairer comparison by taking document length into account. We outlined the results of DeTAS in detecting AI-generated papers. Thanks to the AI system detecting non-detected paraphrased sentences by other systems, we have DeTAS accurately detecting the only two known completely manually designed AI-generated papers by Liebowitz.