Is the Completed Homework a "Task" or "Learning"?—School Evaluations in the AI Era

Well-structured essays. Logical compositions. Visually appealing materials.

When a child submits such a report, adults might perceive it as "well-studied." However, in an age where generative AI can create text and structure, the quality of the work does not necessarily reflect the depth of the student's understanding.

This issue cannot be resolved solely through technology that detects cheating. The criteria for what schools recognize as learning outcomes and what families see as evidence of effort are being questioned.

In the October 14, 2026 edition of the Irish newspaper "The Irish Times," Conor Murphy expressed strong concerns about the education system through his account of attending the educational conference "ResearchEd."

The article addresses more than just AI. It covers memory, review, attention, and classrooms where mistakes can be made. The common theme is the question of whether we are confusing "appearing to have learned" with "actually learning."

Whose Effort Does the Impressive Submission Reflect?

Bradley Busch, who introduced psychological insights at the conference, touched on the difficulty of reliably distinguishing whether AI was used in students' writing.

Murphy connects this to doubts about the expanded task evaluations in Ireland's "Leaving Certificate," not just homework.

This exam is also related to university admission. The "Additional Assessment Components (AACs)" introduced in the reform include reports based on research and projects.

According to an explanation published by the teachers' union ASTI in March 2026, AACs will account for at least 40% of the final grade in nine newly established or revised subjects targeted in the 2027 exam. At the same time, teachers pointed out the difficulty of verifying whether the submitted assignments are the students' own work.

There is value in assessing skills that are difficult to capture in written exams. However, as the impact on grades increases, it becomes important to have a system to verify how much the student thought, researched, and judged on their own.

What needs to be distinguished here is the use of AI as a research aid versus creating a product without personal understanding. Lumping the two together makes appropriate utilization and fair evaluation difficult.

The Limits of Expecting "Teachers Can Spot It"

As material for considering this concern, there is research from the University of Reading in the UK.

In research published in 2024, AI-generated answers were mixed into psychology undergraduate exams to see how they would be handled in the actual grading process. As a result, 94% of AI answers went undetected and, on average, received higher evaluations than actual student answers.

However, this is a result under specific university, subject, and exam conditions. It does not mean the same percentage would apply to all schools in Japan, nor does it indicate the performance of all AI detection technologies.

Nevertheless, it conveys the danger of relying solely on the expectation that "experienced teachers can tell by reading the text" to support the system.

What we want to consider as an evaluation system is combining the final product with short oral explanations, classwork, and records of intermediate stages. It is necessary to verify the student's understanding from multiple perspectives, rather than making any one of them a universal proof.

Why Highlighting with a Highlighter Makes You Feel Like You've Studied

What is striking in the original article is the connection between discussions about AI and traditional study methods.

Highlighting textbooks. Neatly organizing notes. Reading through many pages.

These actions are visible, making the effort easily recognizable to both the individual and those around them. However, coloring the pages and being able to recall the content are different matters.

In a review of study methods by Dunlosky et al., practice tests and spaced learning were highly rated, while highlighting and rereading were rated as having low utility across a wide range of conditions.

This does not mean you should not use highlighters. It is necessary to have opportunities to extract what was important on your own after highlighting.

For example, closing the textbook and listing three key points. Solving problems without looking at the answers. Briefly explaining what was learned yesterday. Checking what couldn't be done and trying again.

When reading easy-to-understand explanations created by AI, the same question should be asked. Did you just agree with it, or can you explain it even after closing the screen?

What You Can Do Today May Not Be Possible Next Week

At the conference, the importance of distinguishing between immediate performance after a lesson and long-term learning retention was also discussed.

You can answer right after hearing an explanation. You can solve similar problems right below the example. However, a few days later, you may be stuck when tackling it without any hints. Such experiences are not uncommon.

The original article introduced a method of distributing learning and repeatedly verifying it while combining related content. Short learning periods of about 30 minutes were also cited as examples, but it is not a story that "30 minutes is optimal for everyone."

If incorporating it at home, you can think of going back a little to what was learned today the next day and checking it again over the weekend. In mathematics, instead of continuing with only problems that have the same solution method, try mixing different types of problems that have already been learned.

What is important is not to use the feeling of smoothly progressing as the sole criterion for success. However, it is not enough to make it difficult and cause trouble. Tasks that can be tackled and explanations that can correct mistakes are needed.

On Social Media, Concerns About "Ghostwriting" and Questions About "One-Shot Evaluations" Intersect

The online debate surrounding this issue is not simply divided into pro- and anti-AI.

The following are opinions that could be confirmed in related discussions on Reddit regarding Ireland's educational reform and AI, not direct responses to this article in The Irish Times. These are anonymous posts and do not represent a survey of the entire public opinion.

In a thread dealing with AI use in research assignments in December 2024, there was a concern that as long as high points are assigned, students might have AI complete their assignments even if it is prohibited by rules.

The same post also explains that it does not completely deny evaluations other than written exams, such as oral exams and music performances. The issue is the design of how to verify the student's ability for each type of assignment.

In another related thread, there were concerns about the differences in support due to home environments and the burden of overlapping assignments. On the other hand, there were opinions questioning the concentration of grades on a single large exam and suggesting that dividing evaluations into smaller parts could reduce student pressure.

What emerges from this discussion is an issue that cannot be organized by just the binary choice of "exam or assignment."

There is value in task evaluations that recognize the inquiry over time. However, if the help received at home varies and the student's understanding is difficult to verify, it may create another form of unfairness. Both methods have advantages to be preserved and weaknesses to be supplemented.

In Japanese Schools, We Should First Decide "What to Evaluate"

This question also applies to book reports, research studies, exploratory activities, and university reports in Japan.

For example, even with the same homework of "summarizing local issues," the purpose is not singular. Is it to cultivate information-gathering skills? The ability to compare materials? Writing skills? Depending on that, the tasks that can be entrusted to AI will change.

If the task is to practice writing, generating the entire text would lose the core of the practice. On the other hand, if the task is to compare information, activities such as investigating discrepancies between AI explanations and official materials can be considered.

The second edition of the guidelines revised by the Ministry of Education, Culture, Sports, Science and Technology in December 2024 also focuses on human-centered utilization and the development of information utilization skills considering the existence of AI. The idea is to consider the developmental stage, the status of ability development, and measures against risks in the use by children and students.

To incorporate this direction into lessons, it is not enough to simply convey "AI use is allowed." It is necessary to indicate which uses are permitted for each task, such as assistance in ideas, correction of expressions, and text generation.

Towards Evaluations That Show "Traces of Thought" in Addition to the Finished Product

So, what kind of tasks make it easier to verify the student's learning? Here are proposals based on the original article and related materials.

One method is to share the process leading up to submission little by little. The initial question, the materials referenced, the thoughts changed midway, the final conclusion. It is not necessary to record all of them in large quantities. Even just the important turning points can capture judgments that are not visible from the finished product.

Another is a short explanation after submission.

"Why did you choose this material?"
"What changed from your initial thoughts?"
"Would the same conclusion be reached under different conditions?"

These questions are not just for cracking down on AI use but also serve as opportunities to deepen understanding.

However, evaluation should not be based solely on the speed of immediate responses or the ability to speak in front of others. It is necessary to provide time to think and methods that allow explanations in short texts or diagrams, according to learning objectives.

Moreover, if all verification work is an additional burden on teachers, it will not last long. Reviewing the number and volume of assignments and securing time to verify the process during class are also part of evaluation reform.

In a Classroom Where Mistakes Cannot Be Made, the Thought Process Also Becomes Invisible

The importance of being able to fail in a safe environment is highlighted towards the end of the original article.

This is deeply related to education in the AI era. If only completed correct answers are evaluated and interim hesitations and errors become shameful, students have no reason to show their incomplete thoughts.

Conversely, if what was corrected from mistakes is recognized, there is meaning in showing the thought process.

"I understood up to this point."
"I couldn't understand this part even after reading the AI explanation."
"There was a problem with the material I initially used."

In a classroom where such statements can be treated as starting points for learning, teachers can more easily grasp where support is needed.

The practices of the teachers introduced in the original article also aimed to capture students' attention and connect it to time for deep thinking. Despite differences in methods and generations, there was a shared trust that students have the potential to learn.

What to Ask After "How Many Pages Did You Do?"

There are changes that can be made at home.

In addition to asking "How many pages did you do?" try asking "What did you understand today?" Instead of just "Did you finish everything?" ask "What parts are still unclear?"

It is not necessary to conduct an oral exam every time. Even when a child struggles to explain, it should be used as an opportunity to check together, not as material for blame.

Generative AI can create visually appealing outcomes in a short time. However, the creation of such outcomes and the retention of knowledge and judgment in the learner are not the same.

What we want to verify in future education is not just the completeness of the submitted pages. It is what the child understood, what they questioned, and what they can do on their own next.

The value of schools in the AI era lies in carefully observing these differences.

Sources and References

1. The Irish Times:
   Descriptions related to AI and task evaluation, memory, attention, and environments where failure is possible.
   https://www.irishtimes.com/ireland/education/2026/10/14/we-are-walking-our-way-into-a-fundamentally-flawed-education-system-and-we-have-been-warned/

2. ASTI: Explanation of the teacher survey published on March 31, 2026.
   Reference to AAC scoring, verification of student-created assignments, and concerns about the burden.Association of Secondary Teachers Ireland
   https://www.asti.ie/news-campaigns/latest-news/the-student-or-ai-teachers-warn-of-verification-issues-in-new/

3. University of Reading: Research presentation on mixing AI-generated answers into actual grading processes.
   Source of the result that 94% of AI answers were not detected in specific university exams.University of Reading
   https://www.reading.ac.uk/news/2024/Research-News/AI-generated-exam-answers-go-undetected-in-real-world-blind-test

4. Dunlosky et al.'s review of study methods / Introduction of research from Kent State University.
   Materials related to the utility of practice tests, spaced learning, highlighting, and rereading.pubmed.ncbi.nlm.nih.gov
   https://pubmed.ncbi.nlm.nih.gov/26173288/
   https://www.kent.edu/kent/news/success/all-study-strategies-not-created-equal-according-kent-state-researchers

5. Ministry of Education, Culture, Sports, Science and Technology: On the use of generative AI in school settings