A large proportion of students now use generative artificial intelligence (GenAI) tools to support their studies, a Hepi survey has found. And it is changing the way they learn.
Before GenAI, when a programming student faced a problem while coding in Python, they may have revisited lecture slides, watched a recorded class or YouTube video or searched for answers on online forums.
Now, the first step is increasingly to open a GenAI tool, paste their problem into it and receive a block of code back. All within the space of a few seconds.
In many cases, students then copy the code directly into a notebook and execute it. If the GenAI output appears to work, the student proceeds. If it does not, the interaction often continues through further prompting, with the student revising the query and the system generating alternative responses until its output is judged good enough.
This raises an important question: by what standard, and by whom, is that judgement made? Is the answer considered sufficient because it runs, because it produces the expected output or because the student genuinely understands how and why it works?
At first glance, GenAI is improving efficiency; students solve problems quickly, complete tasks and move forward with their work. It can also act as a personal tutor to help them break down complex concepts. Yet it also raises pressing questions about learning, assessment and the evolving role of expertise in higher education.
The risks of over-reliance on AI
GenAI does not merely provide information; it alters the sequence of learning activities. Instead of following a linear learning process, students may begin with a finished solution.
Such changes could influence learning outcomes; students who rely heavily on GenAI tools tend to perform slightly worse in traditional examinations, possibly because the tools reduce opportunities for independent reasoning and practice, recent research by academics at the University of Bremen has found.
GenAI tools are also unreliable. Although they can produce impressive outputs, they can also generate incorrect or inefficient solutions. Using them effectively often requires enough discipline-specific knowledge to judge whether the output is valid.
The challenge for educators is therefore to ensure students use AI to aid rather than replace the thinking process.
Users, supervisors and experts
Programming students take on different roles when interacting with GenAI. At one end of the spectrum is the “user”. Here, the AI tool produces code, and the student simply implements it. If it runs successfully, the task is complete.
A second role is that of the “supervisor”. Here, the student treats AI as a tool that generates suggestions rather than final answers. The student reviews the output, tests it, identifies problems and adapts the solution as necessary.
Finally, there is the “expert”, who possesses deep knowledge of the subject and can independently construct solutions while using AI only as an occasional aid.
In other words, AI might change how tasks are performed but it does not remove the need for expertise. In many cases, it may make expertise even more important.
- Spotlight guide: Boosting AI literacy across your institution
- A four-step process to embedding AI literacy in business courses
- Why GenAI helps some students but not others (and what to do about it)
What exactly are we assessing?
This observation leads to important questions: what competencies are students developing when they use GenAI? And, if students rely on it to produce outputs, what competencies are our assessments measuring?
Before GenAI, when a student submitted functioning code that produced the expected output, we assumed that student understood how to write the code.
However, now that GenAI is widely available, the student might indeed understand the code and have written it independently. Or the student may have used AI to generate most of the solution and simply implemented it. Another possibility is that the student refined prompts until the AI produced a working result.
Each of these scenarios involves different competencies.
The assignment might therefore test a combination of abilities:
- knowledge of programming concepts (programming competence)
- ability to formulate effective prompts (prompting ability)
- ability to evaluate and debug AI-generated outputs (supervisory skill)
- ability to use digital tools efficiently (tool use).
These are not the same competencies that traditional assessments were designed to measure but they are all valuable skills.
The workplace perspective
GenAI is reshaping programming roles but some might still require in-depth technical expertise. Software engineers and data scientists, for example, must understand systems at a fundamental level to design, maintain and improve them.
Others might involve overseeing GenAI outputs rather than performing tasks manually. In such environments, professionals will need to guide, evaluate and refine work.
This raises an important distinction between two professional profiles:
- Experts who perform tasks independently
- Professionals who can effectively supervise AI-generated outputs.
Higher education has historically focused on developing the first profile, and GenAI could be shifting the demand. To succeed in either, students need both AI literacy and subject knowledge.
A moment for reflection
GenAI is often framed as a challenge to academic integrity, and concerns about plagiarism or misuse are certainly important.
Across many disciplines, AI tools are subtly reshaping how students search for information, how they solve problems and how they complete assignments.
Educators must now ask themselves:
- If students increasingly rely on AI-generated outputs, what competencies should our assessments measure?
- Should courses explicitly teach students how to critically evaluate and supervise AI-generated output?
- How can we design learning activities that encourage understanding rather than simple tool use?
As GenAI becomes embedded across higher education, these questions are likely to become central to teaching practice.
Acting on them can help universities produce the knowledgeable, AI-literate and critical thinkers that they aim to cultivate.
Fatema Zaghloul is lecturer in business analytics at the University of Bristol Business School.
If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.
comment