AI's effect on language teachers

[Jan Finster]

There is a lot of talk about how ChatGPT and AI is killing jobs in IT and XYZ professions.

 

I think one profession that will be heavily affected by AI is language teaching. ChatGPT 4 is already like writing back and forth with a really good teacher. It will only be a matter of time until it can speak. Here is an amazing example of AI making David Attenborough speak German (he does not):

https://www.reddit.com/r/singularity/comments/132vi0y/this_is_surreal_elevenlabs_ai_can_now_clone_the/

 

So, soon we will have AI talk to us in a perfect and humanlike Beijing accent. The next step is to be able to actually orally chat with AI and for AI to analyse our pronunciation and give us feedback on it. At that stage, language teachers (for adults) will have become largely unncessary.

[becky82]

On 5/6/2023 at 2:28 PM, Jan Finster said:

ChatGPT 4 is already like writing back and forth with a really good teacher.

 

It has some concrete advantages over a flesh-and-blood teacher too:

• It doesn't sleep, nor get impatient or argue.  Your next class is whenever you want, and you don't have to get dressed up and take the subway to get there.  It's never rude---there's no risk that the teacher will ridicule you or play favorites.  It doesn't force the student to use methods they don't want to use---it can suggest many alternatives.
• It understands English, Chinese, and linguistics to a high level.  It uses words very precisely.
• It's crazy functional:  "Can you write a paragraph using [these Chinese words]?"  "Can you make me an Anki deck with 100 insect names in Chinese?"  "Can you write a version of Clozemaster in C++ for me to practice Chinese?"  "Pretend you're a medical researcher, and I'll interview you in Chinese."  "Write a play in Chinese for me to act out."
• It is able to teach Chinese in every domain, e.g. "Can you teach me chemistry in Chinese?"
• Some versions are free.

So are Chinese teachers going to become extinct?

 

It's hard to say, but I don't think it'll be that simple.

 

Quote

You wasted $150,000 on an education you coulda got for $1.50 in late fees at the public library. --- Good Will Hunting (1997)

 

For a long time, people have been willing to pay money to learn things from a human, despite having free alternatives.  But maybe ChatGPT will change that.

[Moshen]

I agree with the skeptics.  Just because AI might seem to make language teachers unnecessary, that doesn't mean they'll become extinct.  As Becky said:

Quote

For a long time, people have been willing to pay money to learn things from a human, despite having free alternatives. 

 

You can check out yourself at most supermarkets or department stores now, but nearly all of them also offer staffed checkout stations where human beings do it.  Now why is that?!

 

Another example: Would you rather learn to drive from a computer simulator or a robot - or from a human?  Despite the fact that he sometimes lost his temper at me, I'm glad I learned to drive from my father.  He had lots of helpful hints along with conveying the mechanical skills.  And because I was one of 4 kids, it was one of the few situations where I was alone with my father.  Which brings up the significance that a teacher-student relationship can have.

[markhavemann]

I don't know, I wouldn't trust a human teacher who would confidently give me a list of 5 items when I ask for a list of exactly 7. 

 

ChatGPT is an amazing tool which can definitely complement your learning, but it delivers every reply with 100% confidence. As someone who is only learning whatever topic you are trying to get it to teach you, how can you possibly know when it's just feeding you garbage?

 

I couldn't see it replacing human teachers and other resources any time soon.

 

 

[hoshi]

https://github.com/liou666/polyglot

[Jan Finster]

On 5/7/2023 at 4:36 AM, markhavemann said:

ChatGPT is an amazing tool which can definitely complement your learning, but it delivers every reply with 100% confidence. As someone who is only learning whatever topic you are trying to get it to teach you, how can you possibly know when it's just feeding you garbage?

 

True there is confident hallucination going on. However, there are also many teachers, who have their blindspots and might unknowingly teach you something wrong. I believe the progress from ChatGPT from January to April has already been a giant leap. Since now the entire world has become obsessed with AI models, there will surely be an ultrafast development of even better models. I am not saying human teachers will become extinct, but probably the market will shrink very significantly. People, who would have booked 5 lessons on Italki per week, may now only book one lesson (e.g. to practice with a human being what they have practiced with ChatGPT before).

 

 

On 5/7/2023 at 2:37 AM, Moshen said:

Would you rather learn to drive from a computer simulator or a robot - or from a human?

 

I was really focussing on language teachers, not teachers in general. Of course, human teachers will still have their place. You will not find an AI martial arts sifu any time soon.

 

 

On 5/7/2023 at 6:08 AM, hoshi said:

https://github.com/liou666/polyglot

 

As I am not a coder, I have never been able to run any GitHub programme yet. It might be super easy (I guess I could ask ChatGPT to help me). Does it exist as a chrome extension?

[roddy]

I don’t see much value in looking at ChatGPT today and saying any Zoom-doable career is safe, if not from elimination then major disruption. The rate of change across generations is stunning and the models you’re interacting with haven’t been specifically trained to be good at any specific thing. Once we get GPT-X and it’s been tweaked to be a good language teacher… it’ll be a good language teacher. 
 

The outcomes are hard to predict. Maybe we end up with teachers able to handle three times as many students with AI assistants. But maybe the associated reduction in costs means four times as many people start learning languages and overall there are more “teaching” jobs. Maybe live translation tech become so good demand for language learning plummets. Etc. 

[becky82]

On 5/7/2023 at 5:40 PM, roddy said:

But maybe the associated reduction in costs means four times as many people start learning languages and overall there are more “teaching” jobs.

 

This is kind of what happened with chess engines.  People predicted the demise of chess, but it had the opposite effect.  Chess engines made it easier to get started, so when you're playing in your first tournament, you're not a total patzer.  ChatGPT is amazingly helpful for L2 language learning.  Maybe this lowers the bar to entry enough that many people start learning Chinese in droves.

 

 

On 5/7/2023 at 3:18 PM, Jan Finster said:

However, there are also many teachers, who have their blindspots and might unknowingly teach you something wrong.

 

It is also worth noting that online study materials also contain bugs---you identify these bugs more and more as your level increases, and you learn to auto-correct.  Especially worrisome is the large amount of "translation-ese"---it's technically correct but doesn't sound like something a Chinese person would say.  Sure, AI is imperfect, but many students are using study materials that contains errors.

 

Humans are also slow.  The other day, my friend was using ChatGPT to study the English word "intricate".  ChatGPT's response was without error, and gave five example sentences from diverse contexts.  It would have taken me quite a long time to do the same thing, and I doubt I could come up with better example sentences.

[roddy]

On 5/7/2023 at 3:36 AM, markhavemann said:

I don't know, I wouldn't trust a human teacher who would confidently give me a list of 5 items when I ask for a list of exactly 7.

I meant to comment on this earlier. What everyone's looking at these days are "large language models", not mathematics models (we've got those already, they work quite well in calculators and suchlike). If you think of ChatGPT or similar as a recent college grad who has somehow managed to avoid any degree of numeracy, but has picked up some concepts from wide reading, you might get close to it. I had a beautiful one a while back when it got a mega>giga conversion turned around and gave me an answer out by a factor of a million. I might have a screen shot somewhere (attached now).*

 

But I suspect it's being allowed to make those mistakes at the moment as the researchers are fascinated by what it can figure out without being taught, and what kind of errors it makes. It won't take long to give ChatGPT a calculator and tell it to check any numbers in its output. 

 

* As an aside - that bit where it apologised and corrected itself felt like some proper uncanny valley stuff on my side. 

[Moshen]

Quote

People, who would have booked 5 lessons on Italki per week, may now only book one lesson 

 

I wonder if I missed something.  Can the chatbots now talk to you (in a realistic voice) and correct your usage and pronunciation?

[roddy]

I don't know if they can do it now, in any market-ready form. Realistic voices maybe, I don't know what state-of-the-art is with that. Correcting usage and pronunciation - I can't see technical barriers here. Speech recognition already involves recognising the sounds people are saying, identifying deviation from some defined set of norms shouldn't be tough. Same for language models and usage.

 

Speech recognition can tell the difference between 'rich' and 'reach'. So if it hears a non-native speaker say "My cousin is really reach" it shouldn't be too hard to identify what was meant and draw attention to it. Similarly a language model knows we say "I went to the beach" and not "I go-ed to the beach", so it can flag that up as an error (even more easily once it's trained on the types of mistakes learners of various L1s are likely to make).

 

But even without that - what it can do with text chat (which could be embedded in apps you already have on your phone) could be remarkable. 

[Jan Finster]

On 5/7/2023 at 3:43 PM, Moshen said:

I wonder if I missed something.  Can the chatbots now talk to you (in a realistic voice) and correct your usage and pronunciation?

 

Currently it is text chat. But, to me this is "outputting" and while it does not improve your pronunciation it should improve your active vocabulary and it should translate to improved speaking [of course once you have the tones, pronunciation and prosody right, so maybe not for beginners].

 

On 5/7/2023 at 11:40 AM, roddy said:

four times as many people start learning languages and overall there are more “teaching” jobs.

 

Actually, I believe even fewer people will learn languages, because this will be a luxury effort with a poor effort-reward ratio. In a few years time, you can probably talk into a small headset and this will be translated by ChatGPT X and  the person you are talking to can hear the live translation in their headset with "your" voice. It will be like the "babel fish" ?

[Dlezcano]

So maybe you could think in the same way if people won't need to visit chinese-forums.com anymore since you can ask chatGPT anything. A robot lacks too many things a human has: it hasn't any feelings, emotions, friends, childhood memories, etc. It might emulate it very well but you will always know those are not real. Anyway I regard it as a very good tool to assist students for learning languages and it is very helpful for teachers too!

[Lu]

On 5/6/2023 at 8:28 AM, Jan Finster said:

So, soon we will have AI talk to us in a perfect and humanlike Beijing accent.

Nah. If that will ever happen, it certainly won't be soon. ChatGPT isn't even all that good at Dutch, a European language not too far from English, official language of two countries, with a large written corpus and lots of translations from English. Who is going to teach it Beijinghua, with what corpus?

 

AIs can certainly be useful teaching and learning tools, but they won't eliminate teachers. Textbooks haven't replaced teachers, Youtube hasn't, Italki and other programs haven't. Chat GPT doesn't 'know' anything. If you ask it a question, what it puts out is 'what an answer to that question would look like'.

[matteo]

I was having this conversation with my girlfriend and she said, the recent Covid and lockdown proved that we can effectively work from home using emails and team calls, and it looked like that was the new way of working.

Then we realized that we didn't like it one bit and we were all missing human interaction, and (largely) went back to the way it was before because we are social animals and we crave interactions with other humans. 

 

I think this point is really valid, in our technologically driven society we routinely miss these "soft" but extremely important considerations when weighing a problem. 

 

 

[becky82]

My thoughts on this matter keep changing rapidly; there are so many aspects to consider.  There are financial motivations (some people can't afford it; some people have disposable income); there's more to going to Chinese class than learning Chinese; maybe ChatGPT will lead to more people deciding to learn Chinese (sometimes considered among the hardest languages for native-English speakers); some students may simply prefer to (or not to) learn with a non-human; LLMs can be trained on vastly more material than any human could read; perhaps teachers can utilize ChatGPT in class to save time, and make it more efficient; what's the distinction between ChatGPT's neural network and my own neural network (i.e., my brain)?; are people expressing negative opinions about this technology because it's unfamiliar or destabilizing ("new, therefore bad")?;  are people's negative comments based on their experience with 3-month-old ChatGPT (which indeed was awful)?; is there even room to improve ChatGPT (it's possible it's already "used up" all available training data---maybe we've already reached peak performance)?;  what's the difference between "ChatGPT hype" and e.g. "blockchain hype"?

 

I feel it's safe to say that ChatGPT (and friends) offer an as-essential-as-Google way to study a language---the toothpaste is out of the tube now.  Nevertheless, I've always espoused "diversifying one's study methods", and I don't see why that would change.  I guess we'll see soon enough if people can learn Chinese ultra efficiently using ChatGPT et al.

 

Anyway, I asked ChatGPT to comment on this thread:

 

Quote

Hi, can you comment on this forum thread? https://www.chinese-forums.com/forums/topic/62832-ais-effect-on-language-teachers

 

ChatGPT:

 

Quote

Certainly, I can provide some thoughts on the forum thread you linked.

 

The thread discusses the potential impact of AI technology on language teaching and learning. Some participants express concern that AI language models like myself could replace human language teachers, while others argue that AI technology should be embraced as a tool to enhance language teaching and learning.

 

In my view, AI language models like myself are not intended to replace human language teachers, but rather to complement their expertise and provide additional support for language learners. While I can generate responses to user input that are contextually appropriate and grammatically correct, I don't possess the same level of cultural and contextual knowledge that a human language teacher can provide.

 

Furthermore, language learning is a complex process that involves more than just grammatical accuracy. Human language teachers can provide personalized feedback, guidance, and motivation, which are essential for effective language learning. AI language models like myself can provide support and resources for language learners, but we can't replace the human touch that is necessary for successful language learning.

 

Overall, I believe that AI technology can be a valuable tool for language teaching and learning, but it should be used in conjunction with human expertise and guidance, rather than as a replacement for it.

[Moshen]

Quote

I guess we'll see soon enough if people can learn Chinese ultra efficiently using ChatGPT et al.

I guess we should also soon expect the hypesters to start advertising dubious claims of incredible language-learning efficiency with their AI-enhanced (and not cheap) new products.

[Jan Finster]

On 5/8/2023 at 1:34 AM, becky82 said:

are people expressing negative opinions about this technology because it's unfamiliar or destabilizing ("new, therefore bad")?;  are people's negative comments based on their experience with 3-month-old ChatGPT (which indeed was awful)?; is there even room to improve ChatGPT (it's possible it's already "used up" all available training data---maybe we've already reached peak performance)?;  what's the difference between "ChatGPT hype" and e.g. "blockchain hype"?

 

Yes, I believe there is a huge scepticism. However, I am sure in 5-10 years time it will become so "normal" that no one will ever talk about it anymore.

[Lu]

On 5/8/2023 at 1:34 AM, becky82 said:

I guess we'll see soon enough if people can learn Chinese ultra efficiently using ChatGPT et al.

They won't.

 

- ChatGPT isn't changing people's brain. To learn Chinese, students have to cram words, learn grammar, practice building sentences, etc etc. ChatGPT can assist them with all that, but it can't do it for them, so people won't suddenly learn Chinese at twice the speed. They still have to put in about the same amount of effort and time,

- ChatGPT can't come up with new teaching methods. Coming up with new ideas is not something this machine is for.

 

Also, this is Chinese we're talking about. Google Translate is now how many years old? and it still can't handle Chinese well. I just fed it a paragraph of 平凡的世界 and for some reason it just skips an entire line and mistranslates one phrase into the opposite meaning.

 

Spoiler

在校园内的南墙根下，现在已经按班级排起了十几路纵队。各班的值日生正在忙碌地给众人分饭菜。每个人的饭菜都是昨天登记好并付了饭票的，因此程序并不复杂，现在值日生只是按饭表付给每人预订的一份。菜分甲、乙、丙三等。甲菜以土豆、白菜、粉条为主，里面有些叫人嘴馋的大肉片，每份三毛钱；乙菜其它内容和甲菜一样，只是没有肉，每份一毛五分钱。丙菜可就差远了，清水煮白萝卜——似乎只是为了掩饰这过分的清淡，才在里面象征性地漂了几点辣子油花。不过，这菜价钱倒也便宜，每份五分钱。

 

Under the south wall of the campus, there are now more than a dozen columns lined up by class. The students on duty in each class are busy distributing meals to everyone. Everyone's meal was registered yesterday and the meal ticket was paid, so the procedure is not complicated. Now the students on duty only pay each person for a reservation according to the meal table. The dishes are divided into A, B, C and so on. Dish A is mainly potatoes, cabbage, and vermicelli, and there are some big meat slices in it, which cost 30 cents each; The third dish is not far behind. Boiled white radish in clear water-it seems that just to cover up the excessive blandness, a few spicy oil flowers are floating in it symbolically. However, the price of this dish is also cheap, five cents per serving.

 

[Guest realmayo]

For AI to be game-changing for learning Chinese, it either has to invent a new and better way of teaching a language, or it has to do things that humans have already invented.

 

For the former, that means developing a much better way of understanding how humans learn languages. I don't think the current style of LLM-AI can do that at all.

 

For the latter, I guess it would just be like a cheap and tireless teacher who always does what you want it to. Compare my tones to that speaker's tones. Write for me 10 paragraphs of text on Suzhou gardens with 20 pieces of specialist vocbulary introduced in advance and the remaining text at 98% comprehension. Tell me 10 把 sentences involving common household items. Then prompt me to say 10 substitution-and-extension 把 sentences and correct them in real time if necessary.

 

So, either a radical reinvention of language learning, or just a few more resources?

 

Also: it is hateful (to me) playing online chess against a chess computer. It's huge fun playing against real human beings. I don't know if that will apply to AI.