(Emma Li)

OpenAI released ChatGPT and gained 1 million users in less than a week on Nov. 30, 2022. Now, ChatGPT has 200 million active users every week. In the academic scene, 46% of high school students in the U.S. admitted to using AI tools in 2023. I believe the rise of AI is eroding the foundations of education and encouraging shortcuts over substance.

Providing essays or homework answers at the click of a button, AI fosters an over-dependence on automated tools that impede the development of critical thinking and problem-solving skills in students. Like any type of cheating, using AI shortcuts the learning process and undermines the course. Is the convenience of AI worth the loss of authentic learning? 

For instance, rote exercises such as mastering Spanish conjugations or crafting basic paragraphs may seem tedious, but they build essential skills. These are the building blocks for tackling more complex and nuanced assignments. In my view, by outsourcing this process to AI, students risk stunting their creativity and self-expression, which are fundamental to personal and intellectual growth. 

Large language model (LLM) usage also raises ethical concerns by blurring the line between assistance and academic dishonesty. ChatGPT can quickly turn into a form of cheating if students ask the model to generate ideas and then pass them off as their own work.

Although ChatGPT can be useful for simple tasks, it is limited by inaccuracies. A study from Purdue University in 2024 showed that ChatGPT-3.5 was incorrect 52% of the time for computer programming questions. LLMs often experience ‘hallucinations,’ where the generated text sounds reasonable but is completely senseless. 

According to an experiment done by WebFX, ChatGPT-4 struggles with basic math and logic problems. After being fed several sample problems, ChatGPT-4 failed to expand basic equations, produce accurate historical information or generate original ideas: all potential prompts a student could ask. 

Even though researchers have designed OpenAI o1 and other LLMs to excel in these fields, the model responsible for these errors, ChatGPT-4o, is still the most accessible and used AI tool. Especially in academics, students should not rely on potentially inaccurate information for the sake of convenience. Doing so deeply impacts a student’s understanding of topics by confusing them with false information.

Considering the increasing capabilities of AI, falling down the rabbit hole of AI abuse is easy. High school students do not want to spend needless effort on assignments they perceive to be pointless and prefer to use AI instead. While AI tools can be beneficial, I believe it’s essential to balance their use with personal effort to ensure meaningful learning. They are no substitute for the hard work and critical thinking that true learning and education require.

(Emma Li)

Students and teachers are often overwhelmed by busywork, but AI can lighten the load. I believe that from educational tools to tutoring, AI can offer accessible and powerful assistance for confused students and exhausted educators alike. 

A quarter of public school teachers’ work hours in the U.S. goes uncompensated, and yet there’s plenty of work that doesn’t require a teacher’s time and expertise, like simple grading tasks and assembling teaching aides. I think that in schools where teachers are overworked and overwhelmed by busywork, AI assistance could ease their burdens through programs like ChatGPT and MagicSchool which might offer help in assembling teaching aides and grading.

Even better, AI has the unique potential to perform certain tasks that are not just difficult, but impossible, for humans to complete. For example, a teacher can’t realistically compare all of their students’ essays to their sources to look for uncited or improper paraphrases, but a Large Language Model (LLM) like ChatGPT can easily evaluate whether two passages of text have equivalent meaning – even if a student inserted synonyms or made structural changes. 

Students also stand to benefit from AI, which can streamline studying and prevent time from being wasted on making flashcards or other study aides. I recommend using LLMs to extract information from textbook passages or study guides to generate practice tests, create quizlet sets and answer questions. By giving an LLM your study materials, the chance of it making mistakes, or ‘hallucinating,’ is lower since it references your input, not just its general knowledge.

AI could also play a transformative role in the accessibility of education. When students don’t have support from teachers, tutors or parents, AI can assist with homework or studying. While I wouldn’t recommend using a general LLM like ChatGPT to replace a teacher, LLMs can offer some guidance, like helping students get unstuck on problem sets, giving feedback on writing or quizzing facts from a study guide. I believe that having access to an imperfect tool – even if users need to be wary of its mistakes – is better than nothing.

Although LLMs ‘hallucinate,’ or generate factually incorrect information, specialized models can perform their tasks with higher accuracy. OpenAI’s recent STEM-optimized series, OpenAI o1, performs well on several benchmarks, including the math competition AIME, programming problems on Codeforces and the GPQA –– a test designed to evaluate models on graduate physics, biology and chemistry problems. Notably, o1 exceeds expert performance on the GPQA, indicating it possesses Ph.D.-level knowledge. It’s also capable of explaining topics with various levels of complexity, as you could request an explanation of the same topic, but tailored to a middle school or high school or university student audience. 

Students may develop an overdependence on AI tools, but I believe they have the potential to create more value than harm. When qualified educators oversee the application and development of AI assistance, such aides can easily be modified to promote independent thought and effective learning, like how education bot Khanmigo offers gentle guidance but is programmed to never give away answers.

Ultimately, while AI does make mistakes, LLMs like ChatGPT and Meta’s LLaMA have the capacity to teach the basics of many subjects and to be used as an aide and sounding board for experts, who can benefit from generated ideas but also have the ability to notice inaccuracies in LLM responses. I believe that AI, when used in good faith, can be immensely useful for education, especially as specialized educational platforms like MagicSchool and ChatForSchools emerge.

Good or bad, AI is here to stay, so I believe students and teachers should adapt to AI tools to support learning.