At a time when so-called generative AI is astonishing us with its capabilities, how can we properly assess the learning outcomes of students? Is the emergence of the ChatGPT chatbot likely to revolutionize assessment practices? Do these challenges present themselves in a radically new way, or is there ultimately nothing new under the sun when it comes to these value judgments that are grades and assessment evaluations?
The significant point is that generative AIs are capable of creating text, images, or even music based on instructions— known as “prompts”—given to them. Machines are taking over what seems to be uniquely human. Won’t they eventually become capable of performing any human cognitive task—and doing so better than we can?
The risk is that “generative AI” tools will be used on a massive scale to cheat. If the learning outcomes targeted by educational initiatives involve precisely these kinds of complex cognitive tasks, the temptation may be strong for some to have intelligent machines do what they are supposed to have learned to do during their training—such as writing a thesis.
Any assessment conducted outside the strict conditions of an exam—particularly “at home”—becomes suspect. Admittedly, the solutions are fairly obvious: require assessments to be conducted exclusively in a “controlled” environment; prohibit—or, better yet, limit or supervise—the potential use of generative AI during exams.
However, this possibility of cheating—which is simply a modern version of the classic practice of “having someone else do it” through identity theft—should not distract us from the central issue, which remains the same: how can we enable the learner being assessed to “prove themselves”—that is, to provide authentic, compelling evidence of what they have actually learned?
As was the case before ChatGPT, two conditions must be met. The first condition is to have a clear idea of the educational goal being pursued. In other words, one must be able to define, in operational terms, the ability or competency that the educational activity—and, consequently, the test or exam—is intended to assess, as these are designed to determine whether the goal has been achieved.
It should be noted that it is not enough simply to identify a body of knowledge; rather, we must explain what the “possession” of this knowledge enables the student to do in a tangible way. We must explain what mastering this knowledge concretely enables the student to accomplish: how can we clearly distinguish between those who possess this knowledge and those who do not?
The second condition—which goes hand in hand with the first—is to find exam “tasks,” that is, “assignments” that require students to face a situation in which they can, precisely, prove themselves. For example: simplifying fractions, writing an argumentative essay, or summarizing a philosophical text. This brings us back to the obstacle of the possibility of being replaced by generative AI: couldn’t it, in fact, “prove” myself in my place if I let people believe that its work is my own?
Beyond the moral and policing issues, let us note that this requires us, in each case, to identify the core skill targeted in every cognitive task that can be the subject of school or university learning. In other words, we must think in terms of concrete, practical abilities—that is, know-how.
These skills can be observed in students through the results they achieve (for example, being able to find their way around a city using a map), rather than in terms of content that can be listed in a curriculum (such as knowing the list of prefectures). This may lead to distinguishing between levels of learning objectives and assessment situations, depending on the type of skill involved.
Instead of lamenting the (relatively) new possibilities for cheating, it might be more useful to ask whether generative AI might offer opportunities for significant improvements in learning. This would involve making intelligent use of the tool—for example, in terms of content, illustrations, lesson planning ideas, or the creation of exercises.
[There are already more than 120,000 subscribers to The Conversation’s newsletters. What about you? Subscribe today to gain a better understanding of the world’s major issues.]
For students, AI can help facilitate personalized learning by providing tailored resources or incorporating interactive learning assistants.
Ultimately, by looking beyond the simple issue of assessment, we could identify areas of focus for AI-assisted education using generative AI. A first step would be to empower individuals to master digital tools—which are, after all, nothing more than tools.
Another approach would be to use this opportunity to try to understand how logical intelligence works; and, more broadly, how thought is formed—by looking beyond algorithmic problems and the workings of technical mechanisms to examine the ethical issues at stake. This can be achieved by co-developing the assessments with students. For the impact of generative AI—on assessment, as well as more generally—will depend largely on how it is used, for good or for ill.
