Generative Artificial Intelligence and Assessments

The ease of generating content with AI has blurred the lines of plagiarism (Hutson, 2024). Students can now produce assignments via AI that evade traditional plagiarism checkers.

In fact, studies indicate that current AI detection tools are often unreliable, with accuracy rates varying, and simple rephrasing can evade detection (Sadasivan et al., 2023).

A recent survey found that over half of students would continue to use AI for coursework even if it were banned, which highlights that AI prohibition may be ineffective (NeJame et al., 2023).

AI systems and algorithms can inadvertently introduce bias into assessment. For example, researchers found AI writing detectors were more likely to incorrectly flag essays written by non-native English speakers as “AI generated,” raising serious equity concerns (Liang et al., 2023).

Likewise, AI-based exam proctoring has been criticized for disadvantaging students of color (Yoder-Himes et al., 2022). Ensuring fairness means faculty should be cautious about using “black box” AI judgments as the sole basis for high-stakes decisions. It also requires diverse representation in AI development.

The US Department of Education’s Office of Educational Technology especially emphasizes that any AI tools used in schools “must include algorithmic discrimination protection” to avoid bias, and there should always be human review and override in the loop.

A lack of transparency in how AI models work can undermine trust in educational use. Many AI tools are proprietary, and schools often do not know what data an AI model was trained on or how it is processing student inputs.

This opaqueness makes it hard to explain or justify AI-driven assessment outcomes to students and parents.

If AI is used, faculty should inform students when AI is involved in grading or feedback and how its results will be evaluated. 

Using AI in assessment frequently involves collecting and analysing student data, from academic work to biometrics in online proctoring.

This raises privacy concerns, especially when third-party AI services are involved. Student data could be improperly shared or used to further train commercial models without consent.

The faculty should ensure that the AI tools used are compliant with privacy laws like FERPA and be transparent about what student information is being used.

Wherever possible, keeping sensitive data local, on university-controlled systems, and anonymizing it can help safeguard student privacy.

Assignments that can be done quickly by AI (e.g., factual recall or generic essays) do little to engage students in critical thinking. In contrast, designing assessments to target the upper level of Bloom’s Taxonomy, such as analyzing, evaluating, and creating, encourages students to apply knowledge in novel contexts, solve real-world problems, or produce creative work.

Rather than relying solely on written essays or exams, give students options to demonstrate knowledge in diverse formats to capture learning. Assignments such as oral presentations, multimedia projects, or group work can not only address different learning preferences but also make cheating via AI more difficult since students must actively participate and personalize their work.

Since we can neither avoid the presence of AI in teaching and learning nor realistically ban students from using it, the most effective approach is to leverage AI to enhance learning outcomes. This is where process-based assessments become especially valuable. Take a look at some creative ways to integrate AI meaningfully into your teaching:

• Output Analysis: Engage students in hands-on experimentation with AI tools. For example, have them generate AI responses to a given assignment prompt. Then, guide them in analyzing the quality of the output. You can even have students modify prompts to produce average or poor answers intentionally and then ask them to critique or improve those responses manually. The key is emphasizing that recognizing and producing high-quality work is the real learning goal.
• Reasoning Analysis: Some AI tools now include reasoning features that show the step-by-step thinking behind responses. Ask students to evaluate this reasoning and reflect on what they can learn.
• Flipped Model:  Josh Thorpe provides a creative example of the AI flipped model. In this approach, students interact with AI to deepen their thinking rather than just to generate answers. Instead of asking AI to write an essay or summarize a reading, students can prompt AI to ask them questions about the topic. This turns the interaction into a Socratic dialogue, where students must actively think, respond, and reflect. In doing so, they practice articulating their ideas and learn to evaluate the depth and accuracy of their own responses.

Cheating is a common concern among faculty regarding the use of AI. However, cheating may be a sign that students either do not fully understand or do not value the benefits of completing the work themselves (Bowen & Watson, 2024).

When students lack this appreciation, the temptation to cheat increases, especially now that AI makes it easier to outsource thinking. One way to reduce this tendency is to build relationships with students and show that you genuinely care about their learning.

Faculty should seek to understand what motivates their students and use this understanding to guide their teaching. Start by designing assessments that demonstrate value and relevance to students’ personal and academic goals—human-centered assessments.

Bandura, A. (1997). Self-efficacy: The exercise of control. Freeman.

Bowen, J. A., & Watson, C. E. (2024). Teaching with AI: A practical guide to a new era of human learning. JHU Press.

Burrow, A. L., Hill, P. L., & Sumner, R. (2016). Leveling mountains: Purpose attenuates links between perceptions of effort and steepness. Personality and Social Psychology Bulletin, 42(1), 94-103.
https://doi.org/10.1177%2F0146167215624196

Hu, K. (2023, February 2). CHATGPT sets record for fastest-growing user base - analyst note | Reuters. Reuters. https://www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01/

Hutson, J. (2024). Rethinking Plagiarism in the Era of Generative AI. Journal of Intelligent Communication, 3(2), 20–31. https://doi.org/10.54963/jic.v4i1.220

Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7).

Luckin, R., Holmes, W., Griffiths, M., & Forcier, L. B. (2018). Machine learning and human intelligence: The future of education for the 21st century. UCL IOE Press.

NeJame, L., Bharadwaj, R., Shaw, C., & Fox, K. (2023, April 25). Generative AI in higher education: From fear to experimentation, embracing AI’s potential. Tyton Partners. https://tytonpartners.com/generative-ai-in-higher-education-from-fear-to-experimentation-embracing-ais-potential/

Russell, S., & Norvig, P. (2021). Artificial intelligence: A modern approach (4th ed.). Pearson.

Sadasivan, V. S., Kumar, A., Balasubramanian, S., Wang, W., & Feizi, S. (2023). Can AI-generated text be reliably detected?. arXiv preprint arXiv:2303.11156. https://doi.org/10.48550/arXiv.2303.11156

Smith, C. (2006). The history of artificial intelligence. University of Washington. https://courses.cs.washington.edu/courses/csep590/06au/projects/history-ai.pdf (links to a PDF)

Yoder-Himes, D. R., Asif, A., Kinney, K., Brandt, T. J., Cecil, R. E., Himes, P. R., ... & Ross, E. (2022, September). Racial, skin tone, and sex disparities in automated proctoring software. In Frontiers in Education (Vol. 7, p. 881449). Frontiers Media SA. https://doi.org/10.3389/feduc.2022.881449