Name: Will AI Replace English Teachers In China?
Uploaded: 2017-12-06
Duration: 9 min 6 s
Description: Liulishuo founder and CEO Li Yang and Fortune’s Clay Chandler discuss

Right now, you know, after we started in September 2012 so early about five years, over the years we've collected the worlds largest speech corporates, speech database of Chinese people speak english, and based on that we built, which we believe the most accurate Speech pronunciation evaluation engine, and the speech evaluation engine in the world, especially for language learners. See, we are able to tell someone with a bad pronunciation from someone with a good pronunciation but bad grammar. We have You know, we can give people very accurate score, and pinpoint the mistakes on their pronunciation, grammar, vocabulary, and coherence. Tell us how many learners you have on your database now? We have over 50 million registered users. That's our free user base. We launched our first, and you know, pretty much the only paid product Which was the worlds first AI teacher, Amobofong, for English, back in July of 2016. By this July, so 12 months in, we have over 600, 000 paying customers. And how much does it cost? Yeah. It's pretty disruptive. We charge people 99 RMB a month. So that's basically $ 15 a month. So for those of you who don't live in China typically learning English cost about between, on a low end $ 1500 Yeah >> Or on high end I would say $ 4000 per year. Okay give uptake. So that's basically what we're saying, it's basically one- thirtieth of the cost annually. And we are actually also offering something more effective and efficient. So we basically tell people, it's one- thirtieth the cost and one- third of the time. We use ETS, which also is in Princeton. And we use ETS's. Test called a toy bridge. It's score maps to the canonical sort of language proficiency system called CEFR, the Common European Frame of Reference. And basically we use that as a benchmark. We realize okay, within two months our 400 Test corporate users have 16 % of them have increased at least one level. From A1 to A2. Or from A1 to B1. And that they have spent about 36. 5 hours across that two month period. But then CEFR would say hey it at least takes you about 100 hours to up one level. So that's the traditional method. So do you know about the demographics of these people? I mean, what their ages are, where they live. Yeah, so for our free user pool, we have one third about K- 12, K- 12 students, one third college, and one third sort of adults, mostly adults. Really? Yeah. That's remarkable. And Wow, I mean, so you've built these algorithms, this voice recognition system that helps Chinese speakers to learn English. That sounds like a fairly specific thing. Do these algorithms, could you easily reverse them around so that you could- [ CROSSTALK ] As is, we've been They have been very satisfied with our level of accuracy. Not only on pronunciation but also on translating to grammatical error detection and across different spectrums and now they are moving forward with co- branded app with us. So- It sounds like one of the things that he has really got going for him is this corpus of data though. It sounds like having all this data, having all these examples of good and bad- Correct. So we started from collecting bits and pieces of pronounciation data, or speech data. Right. Yeah. But then after we launched this AI teacher course, it's a structured course, it's a multi- level course. Now we started, basically quote on quote, recording people's full study traits. Imagine you were studying at school, when we were in school we took exams, those are snapshots of your learning status, okay? These are sporadic, these are not continuous. But now we just have people learning just every step along the way. You know, for these guys. And remembering now they are spending on average five hours a week, yeah, so it's a lot of data. But, in China tho you have a lot of people that their most comfortable language is [ UNKNOWN ] or that it's [ UNKNOWN ]. People that are Kantonese, Chinese. Speakers, really different right. And so the way in which they approach English, the things they get wrong, the things they get right. They have difficulty pronouncing, must be a broad spectrum, does it matter, or doesn't matter at all what the original dialect is? Well it does matter if wanna have any proof right? So that's why we actually, We understand a lot about that. So we know people from Guandong, you know, will tend to have certain mistakes. You know, people from, say Analos will have other types of mistakes. Other types of mistakes. So you know, we, from, you know, pronunciation courses through other courses, we Give them personalized feedback kind of curriculum. So what I didn't mention, which I should, is that in that AI English teacher course, at the back it's adapted learning energy at work. So you know yeah it basically would analyze you know on the one hand the product would it's highly interactive so a lot of data coming in and the engine would analyze it and start pumping out you know expert generated content but in a individualized way like [ INAUDIBLE ]. So interesting thing. So you're going to put a lot of American English teachers in China out of work, no doubt, with this app. [ LAUGH ] Well, I mean, one funny thing is that, it's a huge market, okay. Some people would, I think, for the time being still prefer to talk to a real human teacher, which is perfectly fine. But you know 80 % more than 80 % of our paying system base, was paying for the very first time. Third of the time, third of the cost and you can do it on your mobile phone is a pretty hard proposition to resisit. It's a serious question thought Tom, this is An isse for all types of Ai is what does it do to job prospects? I mean, we hear these numbers out there. Everybody cites the Oxford study that says that 47 % of jobs are gonna be wiped out. I know McKenzie's done some intersting stuff that even something simple. Like if voice recognition technology Improved about ten percentage points, and just got to the mean of human capability, that it will wipe out all these call centers in Bangalore and Manila. What do you think about that? And is that Too dire or do you think that's- I think jobs changed. And I think they change in every revolution we go through. I think the industrial revolution caused all kinds of job change. Actually, we're not plowing fields anymore by hand, that changed a lot of jobs. So we moved from agrarian to industrial, to the tech industry. And now to AI. So I think it will absolutely, fundamentally change the way we were. I think it will change a lot of people who are consumer facing Whether it's, they're in retail or whatever area, whether it's a call center or whatever they're doing. But I think it doesn't, the work we're doing in health care doesn't replace the doctor. The work we're doing, having all this data. I mentioned the data because he has the data. We bought a lot of data in health care. It's not publicly available data to have all these images to scan to build up at this learning repetition of what's a cancer cell and what's not? So you really need all that data. We're improving the odds, improving the efficiency, improving the efficacy of people who are reading these images, not replacing them. So I think for professionals in this field, I don't think they're at any risk at all. I think if you look at truck drivers, I'm not so sure that autonomous Vehicles aren't gonna replace truck drivers. I think that part of it, certainly, they're gonna have to change from the world in some of these roles, to other positions. But I don't think we're gonna replace teachers, even if we have outstanding language learning apps. I don't think we're gonna replace doctors. With Watson Health we can do an awful lot of things to help those doctors make accurate diagnoses, he's serving an underserved market Think about all the people in western China who don't have access to top, all of China who don't have access [ INAUDIBLE ]. Throughout China. [ BLANK_AUDIO ]

Liulishuo founder and CEO Li Yang and Fortune’s Clay Chandler discuss