AI Investor Forum: Gus Tai on AI in Education (Part 2)

Sramana Mitra: I want to double click on English as a second language. In March last year, we were in India, and my father took us to see a school in rural West Bengal, quite far from Calcutta, which is the main metropolis in that region. These rural Indian towns and villages are not places where fluent English as a second language is that common. We were blown away by these ninth grade kids who were giving us tours of their labs and their learning process, and they were so enthusiastic, and the whole thing was in fluent English.

It is highly unusual in a remote place like that these kids are getting access to a level of knowledge that is the Internet’s level of knowledge, which is largely in English. Most remote places don’t have that kind of access. So, if English-as-a-second language can be scaled in some way – whether it’s with the Speak methodology or the Duolingo methodology or a combination of both and variations of that, this is a very high potential area. I think there are nuances from what is the original language from which this English as second language applications are derived, right? So in the case where we were – the original language or vernacular would be Bengali, and English is the second language. In India, there are several other languages and many dialects. Africa has a number of languages to work with, so does East Asia. Latin America is Spanish mostly. I think there’re private opportunities in all those permutations and combinations to build English-as-a-second language applications.

That’s one observation I have. The second observation is on pricing. If you want to scale something, what works in America is not going to work in rural India or rural Africa, or rural Latin America, or rural East Asia. All that needs to be priced at a much lower level. So that already puts a differentiation in and a go-to-market strategy that has to be somewhat different. It looks to me as though this whole domain of personalized tutoring teaching has many different segmentation possibilities around which new private companies can be built, and many of them can be scaled to significant levels. Do you agree?

Gus Tai: I absolutely agree, Sramana, and I think entrepreneurs would benefit by looking at past, similar industrial developments as inspiration for what can happen. I could view this similarly to the rollout of mobile, the sim rollout of the internet, the rollout of PCs where there were elements of the infrastructure of this capability of intelligent automation. You can expect it to get cheaper and cheaper, just like you could expect a microprocessor to become more powerful and less expensive every couple of years.

Today, there’s a higher cost for having this patient, generative teacher that is really the core and crux of how people learn. When people learn, they learn because they have intentional practice that’s customized to their needs, and they put in the repetition. It’s reinforcement learning; that’s how we learn most of our skills.

Now, we have the availability of someone who’s fatigue less. That has a certain cost, but the cost is declining. The business model is – what is the ability needed to monetize it today against the cost structure? Today, there’s a higher cost, and today English-as-a-second language would support that. But if you have an open LLM model, which Facebook is helping to drive that cost to be very, very inexpensive. The internet, the phone provides infinite distribution, which allows this aggregation so that this capability is basically free.

So down the road, you could charge one eighth of the cost of English as a second language and you have a viable business model because of the efficiency of aggregation of demand against a single platform of capability.

So I think from putting on a startup hat, what are the windows of opportunity and the pricing pressure, the deflationary aspect of the supply of the capability you’re offering, and how do you make sure that you don’t get replaced by someone who comes in with a cheaper cost structure?

This segment is part 2 in the series : AI Investor Forum: Gus Tai on AI in Education
1 2 3 4 5 6