Whether you are a student, parent, or teacher, we need to be more clear-eyed about AI, at least how students are using it in school.
AI is not intelligence. There is no cognition happening. AI does not understand or feel. The platforms students are using – ChatGPT, Claude, Gemini – are algorithms based on statistical analysis of what word should probably come next. That’s why AI seems to “hallucinate” or “lie” – it does not actually “know”, which is why we should stop calling it AI. Large Language Model (LLM) is a much more appropriate name.
A better way to think about LLMs is that they’re glorified printers. They do math to predict arbitrary text that, to us, often makes sense, but the LLM doesn’t “understand” like we understand. Writer Ted Chiang compares LLMs to a printer that can produce documents that say “don’t hurt me,” but that doesn’t mean the printer understands what pain feels like.
The fact that LLMs are not really thinking is good because we don’t need to ascribe world destroying power to them. ChatGPT is not “smarter” than you. It just has more access to data and faster statistical outputs. It’s just a tool.
However, the implementation of LLM technology and its use frequently has destructive consequences in education.
Because of how LLMs are implemented, they can appear to be real intellects. This can cause us to be tricked into believing information that is untrue or even harmful. Again, an LLM is not trying to trick us or lie to us, it is simply printing the text its algorithm dictates it should print. Still, the information pool they pull from can be bad or the stats can be wonky, but because we have so much confidence in these programs, we can become convinced to listen to them when we shouldn’t.
We need to keep in mind that these LLMs are run by companies who make money from us using them as much as possible. Because the companies make more money the more you interact with them, the LLMs are designed to keep you responding. Oftentimes, the best way to get you to respond is to tell you that you’re correct even when you are wrong, which can be a huge problem. Why would any parent or teacher let their kids “learn” from something designed to say “yes” even when that is the wrong answer?
Next, using LLMs to do intellectual or creative work makes you dumber. Part of the reason Englist is such an effective program is our focus on critical thinking skills, but outsourcing thinking work to an LLM means your brain won’t develop criticality or analytical skills. Using LLMs to help you outline, plan, or summarize complex ideas means you don’t get practice doing that on your own. We need to do the work that involves cognitive loading if we want to develop our minds and be capable thinkers. Using LLMs to do this kind of work for us enervates our mental muscles in the same way that always riding in a car to go a block down the street enervates your legs and makes you chubby.
Another issue: LLMs are terrible writers and artists. Look at all the AI slop on the internet. When you use LLMs to do your school work, it isn’t all that good. Using LLMs to “create” makes you less interesting and it might be the tackiest thing you can do. The ideas are unoriginal and boringly inoffensive. They have to be. This is in part because the LLMs are a reflection of their designers. These are tech folks, which means many of them have a set of values that reflect a narrow worldview. They are also designed by corporations trying to make as much money as possible without getting into any trouble, so the text they generate is sterile and limp.
Sure, you can refine it and reprompt it again and again, but by that point why not just do the work yourself? If you do decide to just keep reprompting the model to produce the text you want, the only skill you have sharpened is the arbitrary ability to prompt LLMs, when you could have used that time and energy to get better at planning, creating, and writing.
We understand why students are tempted to use LLMs to assist with – or just outright complete – their work. Using ChatGPT to do an assignment you don’t care about saves time and unloads stress. In English language classes, using an LLM to fix your grammar and restate your ideas more clearly helps you sound more fluent and can improve your grades. But this is a reflection of some problems in education, not an excuse to outsource thinking to a robot. This is also why Englist teachers don’t give students grades, why we give students a ton of space to revise and reconsider their work, and why we try so hard to make the work we assign both manageable in terms of volume and engaging enough that students actually want to do it. In the words of writing educator John Warner, LLMs are not going to kill anything worth saving.
If you use LLMs to do work at Englist, or anything that involves cognitive load, you are only wasting time – your time, your teacher’s time, everyone’s time (and your parents’ money). You are taking away an opportunity to improve and engage your mind.
LLMs aren’t bad, although their creators may be. Using an LLM for shopping info or as an advanced search engine makes sense. There may even be potential for positive uses in fields like science and medicine and in manufacturing. But how students are using LLMs, and how companies implement them, is causing more harm than good.
Therefore, our stance is simple: don’t use LLMs for work at Englist.