Whether it’s a MOOC or a live-streamed tutoring session, the bottleneck for most edutech startups is usually the same: teachers.
“Chat [is] this two-way medium. It’s really well-suited for education,” says David ‘DC’ Collier, the CEO of Rikai Labs, an English education startup based in Shanghai. “You’re chatting with a chatbot and…actually learning at the same time.”
Rikai Labs is using chatbots to boost the scalability of its English education platform. Instead of only interacting with either a computer or a human, like in a one-on-one video session, Rikai Labs implements something called “Artificial Artificial Intelligence”, which blends the two.
For example, during any given lesson, Rikai Lab students will converse with both chatbots and human teachers, running through structured content as well as more open-ended role play. Lessons are conducted through WeChat’s chat interface.
“We don’t really regard the AI as necessarily replacing the teacher,” says Mr. Collier. “It’s more like a teacher’s assistant.”
The concept of “AAI” comes from Amazon’s Mechanical Turk, which outsources tasks that are difficult or unsuitable for computers to human beings. These tasks are known as “Human Intelligence Tasks” (HITs) and include things like labeling pictures and transcribing spoken audio files. For Rikai Labs, “Human Intelligence Tasks” include open conversations between students and teachers, who don’t need to be trained in order to talk to humans. Rikai Labs’ chatbots, which the company calls “teacherbots”, are responsible for simpler interactions, such as providing practice material and responding to student answers.
“[Chatbots are] taking out all of the drudgery,” says Mr. Collier. “Hopefully, [teachers] would be able to deal with twenty students at the same time.”
As Rikai Labs’ chatbots get smarter, the student-to-teacher ratio should increase, says Mr. Collier. The platform’s chatbots, like Rikai Labs’ students, learn from teacher corrections, which happen not only during human-to-human interactions, but also bot-to-human interactions as well. When Rikai Labs’ students are practicing with chatbots, for example, teachers can observe and jump in at any time to correct mistakes. As the amount of training material increases, Rikai Labs’ chatbots will become better and better at catching grammar mistakes and teaching students, says Mr. Collier.
China’s Chatbot Industry Is Still Early Stage
In China, chatbots are nowhere near as hyped up as they are in Western countries, where tech giants like Facebook, Google, Microsoft, and Amazon are actively investing in chatbot technology. For example, Google and Facebook are developing powerful natural language processing (NLP) tools, which are the backbone of chatbots – they’re what enable them to understand human speech and respond accordingly.
In May, Google open sourced SyntaxNet, the company’s NLP engine that can parse text and understand grammar. Facebook has its own natural language tool called DeepText, which helps chatbots mimic human speech by learning from Facebook content. China’s tech giants, on the otherhand, have been pretty quiet when it comes to chatbots, though related fields, such as artificial intelligence and natural language processing, are obviously areas of interest, especially for Baidu.
Besides Rikai Labs, there are a few other WeChat chatbot accounts. There’s Microsoft’s Xiaobing (小冰), a flirtatious female bot that users can chat or play games with (she challenged me to a Chinese idiom competition. I lost). Another chatbot WeChat account is Turing Robot (图灵机器人), whose conversational abilities appear to be less mature than those of Xiaobing.
Still, when it comes to more commercial or service-oriented WeChat accounts, chatbots are rarely implemented, says Mr. Collier.
“WeChat is just an amazingly underused platform,” he told TechNode. ” [In] Slack, everyday people are releasing cool new applications that really use the platform. [In] WeChat, people are just shoving marketing material on webpages.”
WeChat’s Developer API supports many functions that are useful for chatbots, such retrieving text-to-speech output, says Mr. Collier. However, most WeChat applications stick to HTML5 pages and rarely take advantage of WeChat’s built-in chatbot potential. In addition, though other chat platforms, such as Line and Facebook Messenger, have the ability to utilize chatbots, WeChat is the only one that can really monetize it as a business, due to its micropayments system WeChat Wallet.
Of course, chatbots still have a long way to go, especially when it comes to open-ended and less structured chat dialogue. “We’re trying to take a fairly curated approach…so it’s not just random chat,” says Mr. Collier. “Over time, we’ll have to see how scalable we’re able to [teach] people.”
“We’ll probably need a bot to watch the teacherbot or something like that, so we don’t get a Microsoft Tay kind of situation,” he adds.
The Shanghai-based startup will face stiff competition from all directions, as China’s English education market is highly lucrative and fiercely competitive. The industry includes large, traditional education companies, such as Education First and Wall Street English, as well as startups, such as Liulishuo (流利说) and 51talk.
Currently, Rikai Labs’ service is free – students can start lessons by following the company’s WeChat account – but doesn’t have that much content. According to Mr. Collier, the company plans to launch another version in two months, and will charge students 20 RMB (about $3 USD) per lesson in future versions.