A service industry meets software

Education has always been a service industry. That is how Shield Xiao Dun describes the sector he has worked in for almost two decades, first in China and now in California. His first company, 17 EdTech, which he started in 2011 and listed on Nasdaq in 2020, grew to 20,000 employees. In his account it worked as a teaching assistant. The company called it a big-data-enabled personalised learning platform, and its focus was practice once a concept had been taught.

The scale reflected the size of the problem it addressed. Xiao Dun describes a country where parents in even the most remote village regard education as the route to a better future for their children, while good teachers and schools cluster in first-tier cities such as Beijing, Shanghai and Hong Kong. A family in Tibet or Yunnan, as he puts it, could not buy what did not exist locally, however much money it had. The arrival of smartphones and successive generations of cellular networks made it possible to reach those learners, including with streamed video. Solving that gap, he says, was how the company captured value.

His current venture starts from a different premise. He is building AI tutors in Palo Alto, and Michael Macfarlane puts it to him that AI is now doing what 20,000 people did less than a decade ago. Xiao Dun agrees, and goes further. He suggests that anyone can now have the equivalent of a 20,000-strong team to build or create with.

Learning that grows with the learner

The case for AI tutors, as Xiao Dun makes it, rests on conversation. The current wave of generative AI began with language models, so a tutor can converse with a student in a personalised way. Even when a student asks a bizarre question, he says, the AI can answer patiently and knowledgeably, at any hour, and follow the student's interest and curiosity as far as the student wants to go. He considers this the best way to learn something.

He uses the image of a tree to explain what personalisation means over time. Every child needs shared roots, the foundational skills of literacy and numeracy. Above that the branches divide, and every leaf is different, down to its veins. A tutor that has grown up with a learner can recommend, from an almost unlimited body of material, whatever best suits that learner's interests and their zone of proximal development. He adds that it costs a fraction of hiring the best human tutor, and describes the effect as something like having Einstein teach you physics or Chopin teach you the piano.

Macfarlane draws out what this could mean for the structure of schooling. If learners can move at the pace of their own aptitude and motivation, the fixed sequence of one year's syllabus after another loses much of its purpose, and someone in a remote village with a device and a connection could in principle reach a very high standard. Xiao Dun accepts that framing, while noting that young children still need the common foundations first. The vision, in his words, is an AI tutor that grows with you.

Motivation becomes the scarce resource

If high-quality content becomes abundant, the constraint moves elsewhere. Xiao Dun points to the heavy use of general-purpose chatbots by adults for learning, writing help and question-and-answer as evidence that lifelong learning is already happening. He agrees with Macfarlane that an education taken at the start of adult life can no longer be expected to last a whole career. The harder question is whether people want to keep learning.

I think the most important goal for education in this era is for someone to be motivated.

— Shield Xiao Dun, Wealth Office

He sees two different strengths in the world's education cultures. East Asian countries, in his view, are better at systematic, long-term execution, helped by more stable political structures that allow longer plans. Palo Alto, where he now works, excels at a different kind of motivation, the self-driven exploration that breaks a system and builds something new. He compares the room a government should leave for that to the blank space in traditional Chinese calligraphy and painting, and suggests China can learn from places like Palo Alto on this point.

Credentials after the final exam

The weakest link in self-directed learning is recognition. Macfarlane raises the case of someone who uses AI to teach themselves to the standard of a Cambridge degree and ends with no credential to show for it. Xiao Dun expects the answer to come from outside today's institutions. Those institutions, he says, face the innovator's dilemma, in which they would have to cut their arms in order to grow wings, and publicly funded systems must cut elsewhere to pay for change.

His reasoning draws on hiring. Interviews are among the more modern ways to assess people, he says, yet still not very effective, and he admits to having made mistakes many times. Judging someone by their output or creative work over time, in the manner of a work trial, is far more reliable. Schools have not worked that way because designing realistic challenges and assessing every student's response was too labour-intensive before AI. Now that AI can judge essays quite well and understand video reasonably well, he believes that has changed.

So I think there will be a new mastery based system coming out of this new generation or new era.

— Shield Xiao Dun, Wealth Office

He calls the coming period an age of creation, in which students are set challenges closer to real life than an abstract maths question and assessed on how creatively they solve them.

For families thinking about the next generation, the thread running through his argument is that the bottleneck in education is moving from access to motivation, and from examination to demonstrated output. Xiao Dun does not claim the technology guarantees a good outcome. He describes two dystopian possibilities, one in which people are kept consuming without growing, and one in which a small group controls the compute and every strategic decision. He sees his own work as one part of the response, offering people a way to find pleasure in growing and learning new things.