AI is winning EdTech approval but is it truly there yet?
Indian edtech startups are going gaga over AI, especially generative AI and LLMs. The likes of upGrad, EdZola, Cuemath, SkillUp Online, and BYJU’s, are all busy experimenting with generative AI, introducing new products and services into the market.
In the global field too, companies are exploring the advantages of AI. In the US, the Southeastern Conference (SEC) adopted a new Auburn University-designed course to assist faculty in implementing AI into the classroom.
Recently, Bill Gates talked about the potential in AI with education, particularly ChatGPT and GPT-4.
“It’s kind of human, but in a naïve, untrained way. So if we think about the next ten years, I’m sort of back to my optimism that I had when the Gates Foundation first got into education, that both the absolute level of learning and the gap with lower-income, minority students, that with these new tools can both close the gap and raise up the overall level of achievement,” the Microsoft Co-founder said in a chat with Khan Academy Founder, Sal Khan.
If we think about the next ten years, I’m sort of back to my optimism that I had when the Gates Foundation first got into education, that both the absolute level of learning and the gap with lower-income, minority students, that with these new tools can both close the gap and raise up the overall level of achievementBill Gates, Co-founder of Microsoft
Sal Khan agrees, who thinks GPT4 could be the end of static curricula. Khan Academy has just launched Khanmigo, an AI tutor that helps manage groups of students together.
AI is helping students learn, teachers teach, and EdTech companies build better courses. Tools like social annotation are transforming the experience of education.
Dan Whaley, the Founder and CEO of Hypothesis, a social annotation platform with over 50 million annotations so far, told The Tech Panda that though annotations have always been around, with AI’s help, it has become especially relevant in virtual educational environments.
“Indeed, the practice of jotting down thoughts alongside texts isn’t new. Many of history’s greatest thinkers scribbled their insights, revelations, and critiques in the margins of their books. But what if those marginalia weren’t confined to dusty bookshelves? What if they were accessible to all, sparking global discussions and shared enlightenment?” he asks.
If you think about it, the transition from physical classrooms to online platforms was more than just about convenience; it brought its own set of challengesDan Whaley, the Founder and CEO of Hypothesis
“If you think about it, the transition from physical classrooms to online platforms was more than just about convenience; it brought its own set of challenges. One of the key challenges being: how do you preserve the essence of dynamic classroom discussions in an online setting? This is where Hypothesis comes into play,” he explains.
But is AI truly able to help in modern education, or are we truly able to leverage the strong points of AI?
AI Isn’t There Yet?
While the adoption of AI is good for enhancing productivity and increasing the efficiency of online learning and EdTech platforms, many think that we aren’t actually able to harness the power of AI yet when it comes to online learning.
Dan Berges, MD and Founder of Berges Institute, one of the online Spanish language learning platforms expanding in the US, Europe, and India, tells The Tech Panda that the role that AI can play in assisting language learners with their speaking practice and language acquisition isn’t clear yet.
“It is not clear yet, as large language models (LLMs) like ChatGPT, Bard or Llama are fairly new. So far, we know they can be really useful for casual conversation practice, as they allow a language learner to chat with a human-like entity that speaks the target language perfectly, and on top of that, can be very engaging.
Natural language processing models have important limitations as they are today, the most important one being the fact that it’s hard for them to differentiate between factual and fictional information. An AI chatbot cannot reliably explain the grammar to a beginner, since it will sooner or later hallucinateDan Berges, MD and Founder of Berges Institute
“For intermediate and advanced language students talking to a good AI chatbot can be very useful. That said, natural language processing models have important limitations as they are today, the most important one being the fact that it’s hard for them to differentiate between factual and fictional information. An AI chatbot cannot reliably explain the grammar to a beginner, since it will sooner or later hallucinate.”
He ponders about the accuracy of LLMs that are known to have failed now and then.
“We could think AI companies will soon find a way to get LLM models to be more accurate, but as of today, there is no clear roadmap on how to achieve this without redesigning the underlying neural networks completely.”
While some language learning apps have deployed AI for Text to Speech (TTS) and Speech to Text (STT) models, Berges isn’t convinced.
“Regarding speaking practice, Text to Speech (TTS) and Speech to Text (STT) models are not very good yet. While a written conversation with an LLM can be engaging and feel “natural”, connecting LLMs to TTS/SST models has not yielded great results so far, as 1. the resulting system doesn’t tend to sound very “human” when it speaks, and 2. there is still some awkward latency happening when the system has to process voice input,” he says.
Future EdTech & AI
Be that as it may, AI still has a bright future when it comes to online learning.
“AI is not there yet, but it will surely have a big impact in the not-so-far future,” says Berges.
“AI could potentially have a huge impact not only on language learning, but on the field of education and on society as a whole. 2023 LLMs are very impressive, and yet they are very limited. But we know there are many companies backed by large amounts of capital employing highly experienced AI engineers that will eventually develop the new generations of natural language processing (NLP) models,” he adds hopefully.
He takes a transportation analogy.
“It could have been reasonable when the first gasoline-powered cars came out to think that in the future we would have flying cars. We ended up having jet planes instead, which allowed us to fly from London to New York in just a few hours. If ChatGPT is an early gasoline-powered car, what will the Boeing 747 be like?” he asks.
A point to ponder indeed. Will AI become an integral part of education but in a way that we haven’t foreseen yet?