The State of AI in Education at UAF, Fall 2025

This teaching tip focuses on two areas of generative artificial intelligence’s impacts on higher education: The state of automatic detection of text generated from AI (AIGT), and three AI-based tools which can help in generating learning materials and activities, and make grading and feedback easier for instructors.

Tool Based AI Detection

Despite vendor claims of high rates of accuracy in detecting AIGT, several recent peer reviewed journal articles paint a picture of mixed results for detection of AIGT in student submitted work. From dozens of studies published between 2023 and 2025, the authors of Accuracy and Reliability of AI-Generated Text Detection Tools: A Literature Review summarized findings in their tables 3 and 4 [1]. There are several trends that can be spotted in this collection of research:

Takeaways

  • There are no tools that can reliably and consistently detect writing that has been created with AI generated text (AIGT).
  • Low and no-stakes assessments provide opportunities for practice and self-evaluation and boost success rates in courses.
  • UAF has made some powerful AI tools available to support teaching and learning for all courses. ​
  • Text generated from newer Large Language Models (LLM) is more difficult for detection tools to identify than from older, earlier models.
  • Free-to-use detectors are less accurate than tools accessed through subscription based methods, but even paid services provide unreliable detection.
  • AIGT is more readily detected in longer passages of text than short responses.
  • The practice of paraphrasing AIGT, purposefully introducing errors and using prompts for specific styles of output, severely degrades the accuracy of detection results.

UAF has configured its version of the Canvas Learning Management System (LMS) to be equipped with Turnitin’s plagiarism and AI detector. Turnitin was one of the tools tested in various studies covered by the Gotoman et al. literature review and has the same shortcomings listed above. While it can be used to flag potential uses of AIGT, UAF’s Office of Rights, Compliance and Accountability notes [2] that tools such as Turnitin will not be used as a sole means of determining if text was AI generated. Instructors should use their specific domain knowledge, familiarity with individual student writing, and tell tale signs of AIGT such as hallucinations or artifacts remaining from the prompting process. Instructors should also realize what they are subjecting their students to if they decide to employ the use of Turnitin for portions of their course.

Student work is retained in a Turnitin archive and students must agree to a lengthy and highly legalistic “End-User License Agreement“. Instructors should take the time to read and understand what students face by agreeing to that document and thinking about whether this tool use, in the end, benefits the education process and the student.

These factors have contributed to leading institutions like MIT and Vanderbilt prohibiting the use of AIGT detector tools such as Turnitin at their respective universities. The group Parents for Leveraging Ethical AI Standards in Education maintains a more comprehensive list of universities that have ceased to use AI detectors at their institutions [3].

The advice that some of these other institutions provide their faculty is consistent with what UAF’s Center for Teaching and Learning has already shared with faculty over the last few years:

  1. Craft a policy on AI use that is consistent with your course learning objectives.
  2. Communicate that policy to students.
  3. Create a learning environment where AI is another tool to be used in ethical means.
  4. Encourage critical and reflective thought in your learning activities and assessments.

The use of AI in academia isn’t all about detection and punishment. The Sloan School of Management at MIT offers this advice to faculty:

Ask your students to write a ‘process statement’ in which they briefly explain how they completed the assignment, including if and how they used any AI tools. They can describe what parts were AI-assisted, how they verified information, and how they made final decisions. This can reduce the chance that students will over-rely on AI without reflection [4].

Combined with proper citation and reflection, such a process statement can move students to a more ethical and responsible use of AI.

Rapidly Develop Training Exercises with Lecture Notes

One of the burdens of instruction is the creation of practice exercises with rich feedback. The benefits of this labor is that students gain from participating in formative assessment. Research backs this up.

Kirsten Davis, a Construction Management professor at Boise State University revamped an introductory course in her department that had been taught to incoming freshmen for over a decade. Her major change was introducing low-stakes quizzes in a course that previously had only a few summative exams. Her main goal was to provide students with self knowledge about their level of preparedness.

In her article, Using Low-Stakes Quizzing for Student Self-Evaluation of Readiness for Exams, Davis summarizes the effect of adding low-stakes assessments, “the quizzing appears to have drastically reduced the failure rate on the exams and reduced the failure rate of the course [5].”

This is wonderful but should be recognized at the time of her article in 2013 as a significant investment of time. It would be optimal if all instructors had the resources to do this for every course, for every module and for every semester. The obvious barrier to this is that creating a large pool of relevant questions with helpful feedback can consume large amounts of valuable faculty time.

Some courses will use textbook publisher materials or other openly available materials, but one of the leading positive use cases of using the power of AIGT in teaching is the creation of low or no stakes assessments directly from lecture notes or recorded audio/video.

Consider the following workflow I used to create a dynamic crossword puzzle that could be placed into a UAF Canvas course.

  1. I prompted ChatGPT to provide me with 50 terms and definitions used in small-foundation concrete construction. I saved that as a text file. The text file could have come from lecture notes or another already existing resource.
  2. I sent this file to my UAF / H5P account as a “smart import”. 
  3. H5P used AI to summarize and present me with the concepts for approval.
  4. I identified which kinds of assessments to create. I opted for crossword puzzles, flash cards, and a traditional multiple choice quiz.
  5. I selected the crossword puzzle and inserted it into a Canvas External Tool Assignment in my course.

In the span of some five minutes, I created a reusable learning activity that would help support student learning. 

The Center for Teaching and Learning recently hosted an H5P workshop, which was presented by one of the H5P engineers. Smart AI Import was covered. UAF does have an institutional license for H5P and this AI-powered “smart import” feature. 

An example H5P crossword activity built with the assistance of AI.

Create AI Powered, Customized “Agents” which Can Converse with Students

One of the very promising tools that we’ve been exploring at the Center for Teaching and Learning is called Noodle Factory. This tool allows instructors to set up AI chatbots that are primed with particular knowledge. Noodle Factory allows you to upload a variety of documents or point to website resources to define a scope of knowledge that will guide the agent’s responses to students. For instance, one could take historical works about Maria Skłodowska-Curie and use that as a knowledge bank upon which conversations with students could be built.

Various facets of the student-AI interaction can be customized by instructors including whether the AI-agent provides simple answers or provides more clues and acts in a socratic manner.

Instructors can view all of the student-agent conversations and see the quality of responses from both the student and the agent. The tool also provides students with the ability to flag responses which they don’t think are correct. This is an excellent way to first, correct mistakes in output, but also to foster AI literacy and deep command of a subject. We are actively exploring this tool with a few departments right now. If you think your class would benefit from this kind of interaction, please reach out to us.

Scale Instructor Grading Time and Feedback with Gradescope

If you teach a discipline with assessments that students usually show their work and converge upon the same answer (like Mathematics, Physics, Chemistry, Biology) there is a tool that can help faculty effectively grade using less time. Gradescope, another institutionally licensed tool available to our Canvas courses, uses AI and pattern recognition to categorize student responses and even identifies sub components of answers. Instead of evaluating the same algebraic answer dozens of times, Gradescope empowers the faculty member to grade and provide feedback for the different categories of answers that students provide. This does two things: It speeds up the grading process for the instructor, and provides consistent scores and feedback to all the students. This adaptive process allows instructors to see patterns in student understanding and frees up time that would be spent grading the same answers over and over again to be spent on actual instruction and individual feedback.

UAF instructors are most welcome to contact an instructional designer to see if Gradescope can help with assessments in their courses.

Conclusion

The widespread availability and use of AIGT has made assessing student work more difficult for instructors. It may be time to rethink some of the forms of assessment that we use, and determine if what is being measured really aligns with what competencies we are trying to instill in our students. A shift in time and focus will most likely be necessary to accomplish the same goals that we took for granted just a few years ago. Fortunately, the same AIGT tools that raise concern with assessments also bring opportunity in terms of making course design and development more efficient and scalable.

References

[1] Gotoman, J. E. J., Luna, H. L. T., Sangria, J. C. S., Santiago Jr, C. S., & Barbuco, D. D. (2025). Accuracy and reliability of AI-generated text detection tools: a literature review. American Journal of IR 4.0 and Beyond, 4(1), 1-9.

[2] (n.d.). Academic Misconduct Policy. Office of Rights, Compliance and Accountability, University of Alaska Fairbanks. Retrieved on 4 September 2025 from https://www.uaf.edu/orca/student-conduct/academic-misconduct.php

[3] PLEASE. (September, 2024). Schools that Banned AI Detectors. Parents for Leveraging Ethical AI Standards in Education. Retrieved 29 August 2025 from https://www.pleasedu.org/resources/schools-that-banned-ai-detectors

[4] MIT Sloan Technology Services (n.d.). AI Detectors Don’t Work. Here’s What to Do Instead. Retrieved 29 August 2025 from https://mitsloanedtech.mit.edu/ai/teach/ai-detectors-dont-work/
[5] Davis, K. A. (2013). Using low-and no-stakes quizzing for student self-evaluation of readiness for exams. International Journal of Construction Education and Research, 9(4), 256-271.

Dan Lasota

Dan LaSota

Instructional Designer
Certified QM Peer Reviewer
Certified QM Training Facilitator

dlasota@alaska.edu