Dr. Ayse Ozturk

As educators, how can we adapt our assessment strategies to counter the impact of AI?

I’d like to begin by sharing the current situation following the recent release of ChatGPT Agent mode. 

• Here is a video of a student automating the homework submission using an AI agent.

• Here is a video of an instructor automating the grading of student submissions using an AI agent.

What now?

The outlook for education isn’t promising if we stick with only the traditional methods we have been using so far. Distrust and fear are increasing among students and instructors in academic environments. Rethinking methods of teaching and learning is inevitable. Why we teach what we teach, and how we teach, needs an overhaul. We continue to hear the same problem since AI became an integral part of our daily lives. However, the recipe is unclear and not provided by the AI companies; it is up to us to figure it out.

For the overhauling of the assessments:

I have been considering assessment strategies involving AI since it became a buzzword in our lives three years ago. Some argue that there is no escape, while others contend that we must trust AI detection tools. My main approach has been to encourage students to use AI for creative purposes, such as content creation in marketing, while assessing knowledge through methods that are less AI-prone, such as multiple-choice test questions and oral interview-type assessments during presentations of a scaffolded term project. There appears to be no real escape from AI use in most written assignments, nor are there foolproof control mechanisms. So, my default assumption is that students will be using AI for most of their work. Isn’t this what employers are looking for these days, too? But since we can’t just do with AI or without AI all the time, how do we find the right balance? Let students brainstorm with AI, create results with AI, but how do you reduce the “brain rot” when you allow AI in class?

Here are some AI-proof or AI-mitigated methods for adjusting assignments:

• Use AI for the purpose of generating creative content

This is the approach I am most inclined to use in my classes. I teach marketing, and using AI for content creation proves to be relevant and helpful in my field. My students utilize AI to generate brand names, slogans, logos, PR campaigns, and advertising materials, among other things. This is probably the most straightforward way to incorporate AI into assignments. Some examples are listed here.

• Require handwritten drafts, assignments, or exams during class time.

When students are asked to handwrite the given assignment in class, this old-school method ensures they complete their work without the help of AI.

• Assign oral presentations or live interview sessions

When students are asked to verbally explain their ideas or work in real-time, this ensures that students’ own thinking is reflected. The best approach is in-class presentations. Another option is to ask students to submit videos where they verbally explain (not read from a screen) their assignments.

• Tie the assignment to the student’s lived experience

Asking students to reflect on a time when they encountered a relevant topic (e.g., cross-cultural conflicts in an international business course) can help encourage original student thinking.

• Design scaffolded, multi-step assignments with visible checkpoints

Some assignments can be built by scaffolding over time, in rounds, integrating student input in different ways over the rounds. For example, some instructors are requesting that the document be completed in Google Docs and ask to review the revision history of the submitted document.

• Ask students to act as tutors for AI, teaching the topic at hand, or as reviewers to evaluate AI’s output.

It’s just like reverse engineering. Ask students to teach AI a topic and instruct them to have AI ask questions in return. Boodlebox can be a suitable AI tool for this approach, as instructors can create a chat folder to view student interactions with the chatbot.

• Ask students to use AI to create a “team of rivals” with different perspectives on a topic.

Students can submit a reflection on how multiple perspectives shaped their understanding. They can assign each AI persona a unique perspective on a chosen topic, instruct the AI to have these personas debate the topic with arguments and counterarguments, guide the debate by prompting follow-up questions and encouraging responses between the rival personas, and synthesize and come up with their own decisions on a given topic (inspired by a WSJ article here).

Red-Light Approach: Banning the Use of AI in Class and The Issues

There is also the red-light approach, where an instructor bans the use of AI in assignments. In this case, the only way to check student work is by using the AI detection tools. One can use SafeAssign or Turnitin to check for plagiarism, but the issue with these tools is the false negatives and positives. A University of Maryland study found a 6.8% false-positive rate across 12 AI-detection tools. Schools like UC Berkeley, Georgetown, and Vanderbilt have disabled Turnitin’s AI detection over accuracy concerns. In the case of fully depending on AI-detection tools, honest students can be victimized. There is increasing anxiety and distrust among honest students as they are also being constantly scrutinized for any AI-generated work. Even if they do not use AI in their work, they must prove it, as AI-detection software often misidentifies human-written work as AI-generated. An example from a NYT article is about a college student who was falsely accused of using an AI chatbot to write an assignment, resulting in a zero on a major grade. She proved her innocence using Google Docs’ version history and a 15-page PDF of screenshots. She later resorted to screen-recording her entire writing process to protect herself from future accusations.

Moreover, some AI tools come with “humanizer” features to reduce the chances of detection by plagiarism tools. For example, DeepAgent by Abacus.AI has a feature called the “humanize” function that helps adjust the tone and style of its generated content to sound more human-like. This way, the content isn’t easily flagged as AI-generated by AI detection tools. These are ongoing issues, and I don’t think anyone has yet developed fully AI-proof assessment methods.

If you think you can detect an AI-generated work by catching possible inaccuracies and hallucinations, students can get around this by using some features like “fact checking” on some AI tools. For example, Genspark offers a feature for fact-checking to reduce hallucinations. Or, Gemini offers a feature called “double-check response.” These features likely reduce the likelihood of inaccuracies in the output and make it more difficult to detect AI input.

Figure shows that the “Humanizer” feature on DeepAgent is intended to make the output less mechanic and more like human written.

Conflicted Verdict

Clearly, banning AI is not an option since there are ways to get around AI-generated work. Many of them are probably passing by many of us unnoticed already. This is true for most online content nowadays. We are unsure whether the news articles we read, or the blog posts, are written by AI (well, I claim this one mine).  We are unsure whether the emails we receive are written by AI or not. We start noticing that some social media posts are starting to look increasingly similar. We are not even sure if that soul-lifting paragraph of an article or a novel is written by AI or not, or even that sweet note from a date. So, it seems there’s no way out other than accepting the new place of AI-writing in our lives.

Just last week, DHL executives pushed forward their AI integration at the company with the ultimatum, “love it or hate it, you have to work with it.” Our job as educators is shifting in the same way; whether we love it or hate it, we must accept that AI is here, and we have to deal with it. The challenge is to determine where to apply it to best serve student learning. Finding ways to leverage student learning while mitigating the “brain rot” outcomes falls mostly on educators. Rather than policing AI detectors and banning their use, in most cases, it may be more helpful for instructors to guide students to gain an understanding of AI as a tool to assist them in learning, growing, and achieving their desired career goals. Education will not move forward by looking backward. Whether we embrace it or not, we have to accept that these tools are already flourishing. AI is not optional anymore, fortunately (for all the benefits it brings) and unfortunately (for all the damage it creates).

References:

• https://drayseozturk.org/ai-generated-work-show-case/
• https://www.wsj.com/lifestyle/careers/ai-feedback-decisions-53be7124
• https://www.nytimes.com/2025/05/17/style/ai-chatgpt-turnitin-students-cheating.html
• https://www.webpronews.com/dhl-germany-uses-ai-to-tackle-retiring-workforce-crisis-and-boost-efficiency/