Will A.I. Replace Language Teachers?

Will A.I. Replace Language Teachers?

Jeffrey D. Shaffer

 

I spend a lot of time reading about computers, technology, and AI. I’m fascinated with the idea of teaching a computer to do things FOR me, and so naturally the recent improvements in AI have intrigued me. I think it’s great to get help with translating emails or getting help with solving a tricky math problem (hey, I like to learn a wide range of subjects). But as a language teacher, I have mixed feelings. On the one hand, I wonder how I might be able to leverage AI to help my students and I improve our language skills (I am a learner of Japanese), but I’m also concerned whether there won’t be any more language teachers in 10 or 15 years. So, I decided to sit down, and based on what I know about the current abilities of AI and LLMs (Large Language Models), I wanted to think through it.

 

The Upside of AI

With my own study and experiments, I’ve found that AI is actually a very good editor. An LLM is great at helping language-learners find, correct and understand their language mistakes. Where before we would have to search online and then sift through a confusing array of hit-or-miss webpages and often questionable results, we can just ask the LLM to explain each mistake, why we probably made it, how to fix it, and how not to make that mistake again. That’s an editor PLUS a tutor! How great is that? The only downside to this, of course, is that it takes time to do it properly, and I suspect many people would just ask the LLM to “fix their mistakes” and be done with it, which is one of the dangerous temptations of LLMs — let the machine do it. Yeah, you might get a decent grade, but you didn’t really learn anything.

Another big advantage I find with LLMs (AI) is that they make great brainstorming partners, so long as I start with some ideas and a direction I want to go. Given a clear topic, some examples, and some explanation to clarify what IS and ISN’T being talked about, then LLMs can be excellent brainstorming partners. I will often ask an LLM to look over a list of topics I have written (such as an outline for an essay) and ask it if it thinks I’ve forgotten anything important, or if it thinks a different order would be better. I don’t just accept the LLM’s answer, of course, but it’s great food for thought, and often it catches some topic or angle that I had forgotten about.

So, AI seems to make a great editor, tutor (think-explainer), and brainstorm partner, but even when fulfilling these roles, it has its limits.

 

The “Middle” Side of AI

LLMs have a very DEEP “depth of knowledge”. Buried somewhere in all their weights and parameters is the entire knowledge of the internet. Literally. AIs know something about everything — from the major themes and juicy gossip found in all the Marvel Movies to Beowulf history and trivia, not to mention a storehouse of programming, math, and science knowledge, examples, and formulas, too! It’s like being friends with a talking library. A BIG library!

But, it sometimes gets those facts and details a bit jumbled. It, just like we humans, can jump to conclusions that just aren’t there. For example, just yesterday I did a web search for a rare username that I occasionally use online, and the LLM confidently told me that this user (me, in reality) was a major contributor to an online gaming forum (I had written perhaps three comments, and that was years ago).

Setting aside its human-like trait of getting a bit confused from time to time, some issues that might actually come up with a language learner are things like pronunciation and accents. I believe LLMs are improving in these areas, but it hasn’t been a major focus of the creators. Most LLMs come with a small selection of “voices” or “characters” you can speak with (if you go with the voice-based conversations). But they cannot, yet, change accents or pronunciation on the fly. Heck, most of them cannot even tell if you are whispering and cannot whisper back. Yet.

An even bigger drawback, at least for more advanced language learners, is the lack of an LLMs ability to read your emotions. They cannot, yet, read the tone in your voice, your intonation, word stress, or timing. They can only consider your word choice and grammar. It’s a start, but not enough to help someone become a high proficiency speaker, or listener! At more advanced levels, so much of language becomes sub-conscious in just the ways given here. I can tell if my wife is stressed, even over the phone, by her tone. I know in an instant. I can tell if one of my kids is scared (or sick) with the first “Dad?” in a conversation. LLMs (AI) cannot do this yet, and not even in the foreseeable future as the companies that make them are currently focused on teaching them to “think better” and “be more logical”, not “understand and mimic human speech patterns”. Maybe in the further future. Maybe.

 

The Downside of AI

But there are some things that LLMs are just not capable of doing, and these are the things that affect language-learning the most. Language is not just a code or a set of vocabulary rules to be memorized; it is a social and emotional human experience. LLMs not only lack the ability to pick up or mimic emotions via speech, but also the ability to pick up and mimic emotions via writing through word choice, grammar, syntax, etc. They are getting there, perhaps slowly, but they are currently not good enough. AI and LLMs do not pay any attention to visual clues, and so they completely miss out on body language and facial expressions, a large part of communication! Changing languages can often mean “changing cultural expectations” which include a shift in body language to boot! I’m less hand wavy when I speak Japanese, but my hands can be all over the place when I speak English!

LLMs also lack the training to push back on learners when there is a problem that needs to be corrected or overcome. They have been so trained as to gently accept (nearly) anything the user says or requests. Yes, they will tell you about your mistakes, when you ask, but they won’t point out the same mistakes the next time you make them, unless you remember to ask again.

LLMs are not able to set deadlines or encourage time management, or even help with timing a speech you want to practice. LLMs are laughably bad as timers! A recent video (recorded around the end of May 2026) showed a man asking ChatGPT to time his running sprints. He said “Go”, waited two seconds, and said “Done!” and ChatGPT confidently told him he had a great running time of several minutes!

LLMs also are terrible in group settings. They cannot tell the difference between different speakers (yet?) and therefore cannot take part, or lead a group discussion. So even if they have some distinct advantages in helping an individual learner improve their language ability, they are NOT going to be able to lead a classroom. The timing problem alone assures this, but lacking the ability to recognize more than one student, with more than one need, is a deal breaker.

And that leads to the biggest problem with considering LLMs as a replacement for language teachers — they lack the human touch. As an experienced EFL teacher, I try my best to pick up on every detail about each of my students — their clothes, body language, eye contact, volume, tone of voice, word stress, pronunciation, grammar, and word choice. AI can only pick up on the last two. Each student has individual needs. Sure, they all might be roughly at the same TOEIC level, but they do not all have the same needs. Some are talkative, but need help with syntax. Some are quiet and need a bit of encouragement. Everyone, and I really mean EVERY ONE of my students is unique and that’s something an AI is simply not able to decipher and teach to.

 

The Backside of AI

In the end, at least as the abilities of the currently available LLMs are concerned, AI makes a great editor, tutor, and brainstorm partner, but a terrible language teacher. Being able to work with an AI to fix and understand your language mistakes, at any time, is a major advantage in using LLMs for language learning, but their LACK of visual and auditory abilities means that students who want to improve beyond mere lexical and grammatical accuracy are going to need a real teacher, a trained teacher… a HUMAN teacher.