Does Packback’s AI-Powered Writing Tutor Cause Students’ Work to Be Flagged as AI-generated?

Short answer, absolutely not!

Packback utilizes safe & academic Instructional AI that provides actionable feedback for students to make adjustments and corrections themselves. Packback will never generate content for students.

Similar to any other web-based writing application (Microsoft Word, Google Docs, etc) we may make suggestions to change one word (ex. change “1” to “one”, fix a spelling error, or recommend a capital letter or hyphen), or recommend students add or remove punctuation to break up a long sentence or combine sentences.  In all instances we are not re-writing what they’ve written, only suggesting grammatical improvements to what they’ve already written.

How Does Packback Use Generative AI?

Our AI is an engine that combines more than 50 AI models and rules that have been matured and trained over a 6-year period. While many AI tools in education put sophisticated “wrappers” around ChatGPT, Packback offers a proprietary Instructional AI engine with a combination of models and rules, built on pedagogical principles such as Blooms Taxonomy and Principal Design Theory.

We use minimal generative AI workflows to power our chatbot (which will never generate content for students), and those features can be toggled on and off at an institutional level.

How Does Packback Detect Plagiarism and AI-generated Content?

We incorporate proactive flagging for students before they submit work on both standard plagiarism detection and AI-generated text, allowing them to learn from their mistakes in real time and make the necessary corrections to avoid repercussions.

If a student does utilize a generative AI system outside of Packback and copy/paste that content into the platform, we will both warn and coach them before submission.

How Accurate is Packback’s AI-generated Text Detection?

Many commonly used gen AI text detection solutions claim they are 99% accurate. While this can sound impressive in marketing, the reality is that to achieve these claims, these providers are tuning their models to accept a very high “false positive rate”.

What research on these models shows is that at 99% “accuracy” they’re accepting roughly a 1% false positive rate. In real terms, this means that 1 out of every 100 student submissions is being incorrectly tagged as using generative AI.  In an academic setting this means in a class of 50 with 2 writing assignments per semester, at least 1 of those submissions you’re incorrectly accusing a student of plagiarism.

We don’t believe this level of “accuracy” is acceptable. At Packback, we took the opposite approach and tuned our model to a .005% false positive rate. In academic terms, this means only 1 out of every 20,000 submissions runs the risk of being incorrectly identified.

How Many Sources Does Packback Check for Plagiarism Against?

Packback’s source database includes:

• >100M Documents pulled from the external web including open-access journals, websites, and documents, and web-based public education and reference materials.

• >50M Student Submissions uploaded to Packback
  
• Paywalled Academic Journals planned to be added to the Packback source database in early 2025, with initial planned coverage including ~2,800 popular journals.