DBA 8080 Module 4 Final Project: The Future of Assessing Adult Learners When AI Can Write the Answer Example

When the Machine Can Write the Answer: The Future of Assessing Adult Learners in the Age of Generative AI

Student Name

Doctor of Business Administration, Thomas Edison State University

DBA-8080: Teaching Adults: Archetypes, Tools, and Tactics - Andragogy Lab

Mentor Name

Month Day, Year

When the Machine Can Write the Answer: The Future of Assessing Adult Learners in the Age of Generative AI

Introduction

In the spring of 2023, a reviewer in our bank's compliance training noticed that several written case analyses submitted by branch managers used nearly identical phrasing. The managers had used a generative AI tool. None had broken a rule, because we had none, and several argued, reasonably, that they would use such tools at work. The episode showed that our assessments, built around written answers, no longer told us what learners could do. This paper reviews the trend behind that moment, the arrival of generative AI in adult learning, and asks what assessment of adult learners will look like as it matures and what it means for instructors.

The Trend

Generative AI tools can produce fluent text, summaries, code and analysis in response to prompts. For assessment, the consequence is simple: any task whose product is a written answer that does not depend on the learner's own context can now be completed without the learning it was meant to show. Cotton et al. (2024) examined the integrity challenges that large language models create in higher education and argued that machine-written work is hard to detect, so institutions need to rethink assessment design alongside their policies. The same pressure applies in workplace learning, where adults have strong practical reasons to use the tools and little patience for rules that seem to ignore how work is done.

What Assessment Is For

Any forecast should begin with the purposes of assessment. Black and Wiliam (1998), in a review of research on classroom assessment, found that formative assessment, information used during learning to adjust teaching and guide learners, produced substantial gains in achievement. The value lay in feedback and in what learners and teachers did with it, not in grades.

Roediger and Karpicke (2006) found in experiments that students who were quizzed on a passage remembered it longer than students who reread it, the effect now called test-enhanced learning. Assessment, on this view, is not only a measurement but a learning event.

Gulikers et al. (2004) proposed a five-dimensional framework for authentic assessment, in which the task, the physical context, the social context, the result or form and the criteria resemble those of professional practice. Authentic assessment asks adults to do what the job requires, in conditions like the job's.

Sitzmann et al. (2008), in a meta-analysis of trainee reactions, found that reactions were only modestly related to learning, so satisfaction cannot stand in for evidence that learners can perform. Workplace learning has relied heavily on reactions and completion data, which generative AI does nothing to improve.

Knowles et al. (2015) add that adults learn best when they see why a task matters and when it connects to problems they face. Assessments that feel like hoops are the ones adults are most tempted to hand to a tool.

How the Trend Is Likely to Develop

Four developments seem likely over the next five to ten years.

First, assessment will move from written products toward observed performance. Role plays, simulations, presentations and supervised work, judged by trained observers against clear criteria, are harder to outsource and closer to the authentic tasks Gulikers et al. (2004) describe. In banking, that means a manager is assessed on how they handle a customer complaint in a simulation, not on an essay about complaint handling.

Second, evidence of process will matter as much as the product. Drafts, decision logs, recorded reasoning and short conversations in which learners explain their choices show whether learning took place. An instructor who asks a learner to walk through how they reached a recommendation, including how they used an AI tool, learns more than a finished paper shows.

Third, frequent low-stakes retrieval will grow. Short quizzes, flash scenarios and spaced questions delivered on phones serve the testing effect and give formative feedback, and the low stakes reduce the incentive to outsource them. AI tools themselves can generate and vary these questions, which makes the practice cheaper to provide.

Fourth, AI use will be built into tasks rather than banned. Many adult learners will use AI at work, so assessment will increasingly ask them to use it well: to check its output against policy, to find its errors and to adapt its drafts to a specific customer. The skill assessed shifts from producing text to judging and applying it.

Effects on Instructors

These changes will alter what instructors do. Observing and judging performance takes more time than grading essays, so instructors will need well-designed criteria, trained co-assessors and better use of peers. Process evidence requires conversations with learners, which suits small groups but strains large online courses. Instructors will need fluency with AI tools themselves, both to design tasks that use them and to recognize their limits. And they will need to explain to adult learners why assessments are changing, since adults who see the purpose of a task are more likely to engage honestly. The role shifts from grader of products toward coach and judge of performance.

There are risks. Observation can be biased, and criteria must be applied consistently across observers. Low-stakes quizzes generated by AI can contain errors. Building AI into tasks may favor learners already comfortable with technology. Each risk calls for deliberate design rather than a return to the written test.

Implications for Our Bank

For our programs, the review leads to three decisions. The compliance case analysis will be replaced by a recorded simulation scored by two trained observers. Each branch manager course will include weekly retrieval questions sent to phones for three months after the course. And our guidance will allow AI tools in preparation, while requiring learners to explain their reasoning in a short conversation with the instructor.

Conclusion

Generative AI has not made assessment of adult learners impossible; it has exposed assessments that measured products instead of learning. Research on formative assessment, retrieval practice and authentic tasks points to a future that is in many ways better for adults: assessment that looks like the work, that teaches as it measures and that treats AI as a tool to be used with judgment. Instructors who move in that direction now will keep their ability to know what their learners can do.

References

Black, P., & Wiliam, D. (1998). Assessment and classroom learning. Assessment in Education: Principles, Policy & Practice, 5(1), 7-74. https://doi.org/10.1080/0969595980050102

Cotton, D. R. E., Cotton, P. A., & Shipway, J. R. (2024). Chatting and cheating: Ensuring academic integrity in the era of ChatGPT. Innovations in Education and Teaching International, 61(2), 228-239. https://doi.org/10.1080/14703297.2023.2190148

Gulikers, J. T. M., Bastiaens, T. J., & Kirschner, P. A. (2004). A five-dimensional framework for authentic assessment. Educational Technology Research and Development, 52(3), 67-86. https://doi.org/10.1007/BF02504676

Knowles, M. S., Holton, E. F., III, & Swanson, R. A. (2015). The adult learner: The definitive classic in adult education and human resource development (8th ed.). Routledge.

Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249-255. https://doi.org/10.1111/j.1467-9280.2006.01693.x

Sitzmann, T., Brown, K. G., Casper, W. J., Ely, K., & Zimmerman, R. D. (2008). A review and meta-analysis of the nomological network of trainee reactions. Journal of Applied Psychology, 93(2), 280-295. https://doi.org/10.1037/0021-9010.93.2.280