Where AI Can't Help with Language Learning

It's all about context

The recent GPT-5 announcement from ChatGPT showed off its coding capabilities to write an app to teach someone French. It took all of four minutes to write the app. It put up some flashcards, mainly, based on vocabulary hard-wired into the app. and could do quizzes. It looked quite polished.

And there was another mention of learning languages, when they introduced GPT-5’s new Study Mode, which they demonstrated helping someone learn Korean, like how to order a coffee in a coffee shop. That one had some very cool features, like the ability to speak the Korean very slowly.

So—assuming people are still going to want to learn foreign languages, or expand the knowledge they’ve already acquired—do these kinds of capabilities sound the death knell for language learning products and apps?

I think not. Both of these examples of GPT-5’s application to language learning are missing the key aspect of context.

What’s missing: context

By context I am referring to

• what you want to learn—your learning goals

• how you want to learn it—your learning style

• what you’ve already learned—your learning history

• what other people are learning—shared learning experiences

Context matters for retention, transfer to real-world use, and motivation.

These are things that—at least at the present state of AI development—the AI can’t handle very well or maybe not at all. Of course LLMs can track goals, history, and preferences if engineered correctly, but AI today handles context inconsistently and shallowly; apps must provide the scaffolding for deeper, persistent context.

Managing this context and shaping the AI interaction based on that context will be the core feature of AI language learning apps going forward. The things we’re used to seeing in language learning apps, like predefined lists of vocabulary, or hard-wired flashcards, or even spaced repetition systems (SRS), will shrink into a supporting role at best. The next generation of language learning apps will mainly tee up the context and hand it off to the AI, or more accurately to many AIs.

What, no more SRS? That’s right. SRS was a great solution for knowledge that can be expressed in the form of flashcards. But learning a language is fundamentally different than mastering a deck of flashcards. Even if you do master it, there’s no guarantee you’ll be able to recall it in real life. The best you can hope for is to do well on some test. Many people have worked very hard and made a huge amount of progress on SRS algorithms and helped many learners in the process, but in the future the AI will be able to do a much better job of determining what knowledge to present in what form and test for in what order and at what frequency to maximize learning.

Some legacy apps have a whole bunch of predefined scenarios and dialogs built in, like going to eat at a restaurant. But those are static and repetitive, even if you use AI to help create those pre-built scenarios, and some legacy app vendors are doing. But that’s a suboptimal approach—the scenarios should be constructed in real time, based on the the learner’s context.

One good example of dynamic content creation is Google’s otherwise underwhelming Little Language Lessons.

You describe a situation—maybe it’s “asking for directions” or “finding a lost passport”—and receive useful vocabulary, phrases, and grammar tips tailored to that context.

This same principle of moving toward generating instructional content in real time with AI, in personalized fashion, will also gradually obsolete many or most other aspects of the current generation of language apps.

Kanjiism, the AI-powered kanji learning app

To explore these ideas, we are building an AI-powered kanji learning app called Kanjiism. It’s in testing now and not yet released. We discussed it in more detail in a previous post.

Kanjiism is an example of this new generation of language learning apps, albeit narrowly limited to learning kanji. It essentially provides a structured pathway to a variety of AI-based experiences that work together to promote learning which is effective and fun. It maintains the overall context important to the learning process and orchestrates AI flows around it. Of course it’s a work in progress. We have much to learn about how to wrap and shape AI for the language learning problem.

I’m looking forward eagerly to this new generation of apps and seeing how they can accelerate language learning.