Language learning is experiencing a revolution. Large language models (LLMs) and AI-powered apps are sprouting like digital mushrooms, promising fluency with just a few swipes and clicks. And the buzz and excitement have now resulted in the global AI education market being valued at a colossal $3.79 billion in 2022. But one key question remains: What role do teachers play in this technological advancement?

It’s easy to see how AI can be a valuable tool in learning languages. From automating repetitive tasks, like grading a student's work, freeing up teachers' time for more one-on-one student attention, to facilitating learning at home and making language homework more engaging.

However, as the founder of a Spanish language institute, I believe there are still a number of dilemmas that AI poses in classrooms. From weak memories to not being thought-provoking and requiring extra monitoring in schools—making them unaffordable—these issues reveal that AI isn’t ready to replace human teachers just yet.

Let's examine where AI technology struggles and how human teachers overcome these issues.

Unaffordable for schools

It’s no secret that most state schools are on tight budgets. With the final portion of funding from the federal CARES Act of 2020 being allocated this year, schools are facing an uncertain future.

With this in mind, it seems very unlikely that schools will have the spare budget to invest in AI technology, especially when the difference between spending and funding is $15.3 billion or $310 per pupil.

Moreover, when specifically talking about LLMs, like Google's Gemini and OpenAI's Chat GPT, although these can be free of cost, these types of LLMs aren't suitable for classroom use. For example, the prevalence of hallucination in ChatGPT is estimated at a rate of 15% to 20%, making them unpredictable and inappropriate for children. Therefore, schools would need to invest in expensive educational-specific AI technology.

Furthermore, biases are prevalent in the data used to train LLMs, potentially leading to discriminatory or harmful outputs. Schools have a responsibility to ensure the materials used in education are unbiased and inclusive, so implementing and monitoring LLMs for potential biases would add another layer of complexity and ethical concerns.

So not only are teachers responsible for children's education, health, and well-being, but with LLMs in the classroom, they will need to ensure their teaching tools are safe. And it’s unlikely that human teachers will be compensated for this extra responsibility.

LMs Limited context window

When first released, LLMs were very impressive with their conversational abilities, but one issue came into focus quickly—their limited context window.

OpenAI's GPT-3.5 has a 4096-token context window, and with 100 tokens equalling approximately 75 words, that means it can process around 3,000 words. Since typical LLMs don't have human memory capacity, this means they will become confused and forget the conversation topic after around the 3,000-word mark.

This limited context window can lead to misinterpretations, inaccurate responses, and a lack of understanding of implicit meanings and cultural nuances. Additionally, this conversation capacity pales in comparison to the vast and nuanced context encountered in real-world conversation practice with human teachers.

LLMs are useful for providing factual information about widely known topics and generating simple responses based on given prompts. However, they struggle with complex discussions, critical thinking, and understanding open-ended questions. On the other hand, human teachers can guide students through challenging concepts, encourage divergent thinking, and personalize their approach based on individual needs.

For a fee, larger context windows are available, but this doesn't help make them more accessible to educational settings with strict budgets.

Technology isn’t all fun and games

Put simply, using AI for language learning is essentially all about input and output, which is rather dull.

For example, let’s look back at the invention of the printing press and the internet. Although college students could find the exact books they needed in libraries or online, meaning they could study their entire semester's books alone, students still went to college.

And even if students could receive long emails every day from professors on precisely what their lesson was on that day, I can guarantee if asked, students would still prefer to commute and go to school and have human teachers. We are human and crave variability and interactions; we're not input and output models.

While people love keeping up their Duolingo streak, if you asked how interesting the app made language learning, the answer is probably 'not very.' An app can’t explain the etymology of a word using fabricated tales from time gone by; it can only provide the facts. A language platform can’t comfort children after hard language exams and give heartful pep talks of encouragement. A language game can’t give a classroom of children an authentic cooking class helping children understand a country's cultural nuances and learn new vocabulary.

While technology holds the promise of entertainment and undivided attention, in reality, this just isn’t the case.

Can technology and humans complement each other in language education?

That’s a difficult question to answer. Could anyone have predicted that 20 years ago, we would have handheld computers in the form of phones and applications on them that were designed for engagement and to aid language learning? Probably not.

Currently, while applications like Duolingo do a great job at using hook techniques to keep us engaged and remind users to practice, whether that helps with real language acquisition is unlikely. Even the Duolingo CEO and Co-founder Luis von Ahn has admitted when talking about the app: "I won't say that with Duolingo, you can start from zero and make your English as good as mine."

The human nature of a teacher is their main asset. Technology fads will come and go, and they can be valuable tools for homework or grammar practice or marking piles of exam papers. While these technology examples can complement human teachers, making their lives a little bit easier, I don't see AI as having a significant impact on classroom language learning any time soon.

AI technology presents exciting possibilities for language learning, but it's not a silver bullet. Human teachers bring an irreplaceable depth of understanding, cultural insight, and the ability to adapt to individual student needs. They can bridge the gaps where AI's context falls short, foster critical thinking, and create a vibrant, supportive learning environment essential for true language fluency. AI technology is not to be scoffed at or overlooked, but when it comes to language learning, teachers rule supreme.

About the author

Dan Berges, Co-Founder and Managing Director of Berges Institute, the fastest-growing Spanish language school for adults in the US. Dan is an entrepreneur, teacher, writer, programmer, podcaster, and musician from NYC. With a Master's degree in Teaching and a background in experimental semantics, Dan is also an author of The Graf Method for Spanish Language.