DETECTING AI-GENERATED CONTENT
AI-generated writing is any text that is produced by artificial intelligence (AI) systems, such as chatbots, text generators, or summarizers. AI-generated writing can be useful for many purposes, such as entertainment, education, or research. However, it can also be misused for deception, plagiarism, or manipulation. That’s why it’s important to be able to detect and evaluate AI-generated writing and to use it responsibly and ethically. Below you'll find information designed for this purpose. We hope you find the information contained on this page useful and informative.
Decoding AI-Generated Content: Recognizing Patterns and Pitfalls
To Err is Human: Human text usually contains typos and is extremely variable, whereas AI-generated content vary rarely makes typos and often generates "perfect" sounding text.
Logic Check: AI models sometimes make implausible or incorrect statements.
Survey Says: Large language models are more likely to use predictable words like "the," "it," or "is" rather than more uncommon words.
Testing, Testing: If ChatGPT is asked the same question (or is provided the same exact prompt) from multiple users, it will provide nearly the exact same response to each. This repetition of words, logic or reasoning can be indication of AI generated language.
Watch the Clock: ChatGPT cannot detect events that have occurred recently. This can lead to factually inaccurate information.
Practice Makes Perfect: Researchers from the University of Pennsylvania created a game to test how good humans are at spotting AI-written text—try the game and see how well you can identify computer generated language.
Battle of the Bots – AI vs. AI: Software exists that analyzes the different features of a text, including how fluently something reads, how often certain words appear, and whether there are predictable patterns, to determine the likelihood that it is composed of AI-generated content. Here are a few examples:
Using Turnitin's AI Writing Indicator
From the Provost re: Turnitin's AI Detection Capabilities
Dear Colleagues:
As the generative artificial intelligence (AI) discussion continues worldwide, I want to let you know that Turnitin (TII) is sharing a preview release of its AI writing detection capabilities starting on April 4, 2023. While the AI detection capabilities will be built into the existing TII similarity report in Brightspace, please do not use AI detection applications alone to conclude that there has been plagiarism or a violation of academic integrity.
There will be an announcement in Purdue Today about this new TII capability as well, and the Purdue Global team will be evaluating detection applications and providing guidance. Updates will be shared as soon as new information is available. In the meantime, please review the Purdue Global Generative AI Statement for guidance on appropriate uses of AI generated content.
Please be aware that students will not have access to the AI writing detection indicator and report in Turnitin.
If you have any questions, suggestions, or concerns, please complete this form.
Thank you,
Jon Harbor | Provost
March 31, 2023
Turnitin Videos & Podcasts
Turnitin’s AI Writing Detection: An Introduction for Educators and Administrators
This video provides an overview of Turnitin’s new AI writing detection for educators and administrators.
Understanding False Positives Within Turnitin’s AI Writing Detection Capabilities
This video explains how Turnitin's AI detection model deals with false positives and what constitutes a false positive.
Turnitin Blog Posts
Understanding False Positives within Our AI Writing Detection Capabilities
A Closer Look at Turnitin's AI Detection Capabilities...
Accessing the Report
Turnitin’s (TII) AI detection tool/score is NOT visible to students.
To access the tool simply open the regular TII Originality Report—the AI “score” will be displayed in a small, blue box labeled, AI.
When instructors first access the new AI detection tool there are guidance resources provided that walk them through this new aspect of the TII Originality Report.
Considerations
Turnitin (TII) launched its AI detection tool on April 4th, 2023.
The detection tool is currently trained on ChatGPT and ChatGPT 3.5–it has not been fully trained on ChatGPT 4.
TII’s AI detection tool currently reports 1% false positives.
Just like with other instance of plagiarism, the TII Originality Report should always be read carefully to determine whether a breach of academic integrity has occurred with the use of AI. Please do not use AI detection applications alone to conclude that there has been plagiarism or a violation of academic integrity. For instance, do not judge a submission by a percentage only. Finally, the current Student Code of Conduct should be followed for cases involving the use of AI.
Use as a learning moment.
© 2023 PG AI Task Force | Webmaster: William Ashley Johnson | Contact: wjohnson3@purdueglobal.edu