Responsible AI use in schools

Artificial intelligence (AI) can be useful when intentionally deployed, particularly in schools. While there are several tools, such as plagiarism checker Turnitin, that claim to detect AI usage, it may be more useful to reflect on what genuine learning entails.

Harvard’s CS50 Introduction to Computer Science built its own tool with guardrails: a “CS50 Duck” the knowledge of which is limited to the course material and that explains how code works—but doesn’t give answers directly. Student surveys reported high usefulness under these constraints.

This is AI as a teaching assistant, not a shortcut for students.

Turkiye tested nearly 1,000 high-school students. One group had no AI, another used a standard chatbot (“GPT Base”) and a third used a constrained version (“GPT Tutor”) that offered hints but not full answers. During practice, both AI groups scored higher than control GPT Base by about 48 percent and GPT Tutor by 127 percent.

But on later proctored exams without AI, GPT Base students scored 17 percent lower than those who never used AI while GPT Tutor students performed about the same as control. This third group that practiced with a constrained, customized AI tutor—one that offered hints but was disallowed from giving answers—performed better than the control group on practice exams and about the same on the closed-book exam.

The gap came from behavior: unrestricted users copied full solutions instead of reasoning through problems. The study shows that unguided AI raises apparent performance but weakens learning while guardrails can retain most of the benefit without the loss.

In our problem-solving class, former Ateneo president Fr. Bienvenido Nebres and I expose students to nonroutine problems and give hints when needed to help them figure out strategies, rather than the answers themselves. Learning from mistakes is encouraged, and students sometimes come up with better solutions than the standard ones, which promotes interest, motivation and retention.

On policy, Ateneo de Manila has moved early. A university memo “GenAI in Higher Education gives guidelines for teaching, learning, research and creative work.

AI use centers on alignment with learning outcomes and instructor control. The policy frames permitted use at the course level, requires transparency from students and cautions against uses that replace core skills practice. It explicitly advises restricting or prohibiting GenAI when it would retard critical thinking, problem-solving, analysis or skills acquisition needed for the course.

That is a clear standard: design uses that support those skills, and block uses that replace them.

See Also

Sugarcane nursery, soils lab to rise in Victorias City

Applied, that yields a playbook for basic and higher education:

• Declare rules in the syllabus. State where AI is allowed for brainstorming or feedback and where it is banned. Require disclosure for any assistance.
• Design for effort. Use AI to prompt planning, hint next steps and explain errors. Do not let it output complete essays or full solutions for graded tasks. Tie major grades to unaided work.
• Verify learning. Pair AI-supported practice with oral checks, in-class writing and closed-tool assessments. Check retention, not only completion time.
• Protect integrity and security. Follow the school’s security and data-handling rules when piloting tools.

The policy direction is not anti-AI. It is pro-learning. The evidence is consistent: unguided use can help in the moment yet harm mastery later; guided use can preserve or improve learning.

Schools should take the win by putting teachers in command, building course-level rules and measuring what matters. In education, the unit of output is not a perfect paragraph today. It is a durable skill tomorrow.

Queena N. Lee-Chua is on the Board of Directors of Ateneo’s Family Business Center. Get her print book “All in the Family Business” at Lazada or Shopee, or e-book at Amazon, Google Play, Apple iBooks. Contact the author at blessbook.chua@gmail.com.