With artificial intelligence [AI] growing in importance throughout our lives, a current classroom teacher explores a free-market future for artificial intelligence education and schools.
This blog was produced as a reflection after our event on AI and education with the Goodison Group in Scotland.
Large language models (LLMs) will in time lead to a wealth of diverse and high-quality educational offerings that parents and children will be able to choose from. It will enable schools themselves to be provided by private markets without any state provision or state subsidies. There are clear arguments for how this would be brought about and why this would be desirable.
Large language models as a driver for change
LLMs are inexpensive to develop and are becoming cheaper over time. For example, programming an LLM like ChatGPT 3.5 from scratch today would only cost $1.4 million (1). However, access to “high quality training data” is scarce (1) and the data that makes an LLM commercially useful is held in a silo within individual organisations. It is reasonable to expect that LLMs will provide inexpensive or ‘free’ access to organisations that then proceed to train an LLMs on specific data they collect, thus adapting it for specific roles or uses in the organisation.
In this way, private organisations will be able to provide tuition and eventually high schools using LLMs. A start-up would train an LLM created by Meta, Open AI or Google on data such as online videos and textbooks. It will also train it on data collected by observing physical schools, recording the subtleties of interactions between teachers and students, scanning in real students work and teachers’ feedback.
The LLM will then be refined through rigorous internal testing, but also through interactions with students over time. Simulated teachers will be able to not only provide consistently outstanding and individualised teaching, but they will also be able to develop ever-clearer and ever-more engaging explanations with use.
The social aspect of high school
An important aspect of the learning experience of a high school is the social experience. A school would be incomplete if it did not offer the ability to make friends, to chat in breaks, at lunch (or during lessons).
The least costly method of simulating peers in a virtual school would rely on data from social media companies. Platforms like Facebook, Instagram and Snapchat contain the largest repositories for colloquial language from young people from every cross section of society. An LLM trained on this data will be able to provide a realistic simulation of any personality that has ever used social media.
Parents would seek out a ‘good’ virtual school factoring in not only the teaching but also the quality of the peers the school is able to offer. Virtual schools would compete on the basis of the diversity, the values, or the intelligence of the peers it simulates.
Accessibility of virtual schools
The preceding paragraphs establish the premise of LLMs being used to provide realistic simulations of teachers and peers — chatbots. To succeed, it will be necessary for video and audio-generating LLMs (e.g Synthesys) to produce higher quality video and enable audio interactions. This would enable teachers and children to engage in a virtual classroom environment.
Experiencing the detail of these environments fully would require more expensive devices, but this does not restrict virtual schools to the affluent. Just as the internet can be accessed from a variety of devices with vastly different prices, , students might engage with a virtual school through: a smartphone, a computer, and/or an augmented or immersive reality headset. As such, virtual schools will be accessible to students of every socioeconomic background.
Quality – and style – of teaching
The observable and consistently high quality of teaching and pastoral care will attract parents to virtual schools. But parents will also be drawn by the choice that virtual schools will be able to offer.
They will be able to choose a traditional education or a progressive style of teaching for their child. They will be able to choose a school that emphasises what parents value, such as high expectations, diversity, the arts, discipline — or all of the above.
They could simulate a boarding school such as Eton, an American high school, a diverse north London school, or Hogwarts — with proprietary film characters such as Snape and Dumbledore teaching Higher Chemistry or English.
Parents will be able to receive rich real-time updates on their child’s progress based on questions answered in class, class participation, progress in 1:1s, written tasks, exams. They will be able to replay moments of misbehaviour, or be shown pieces of work that their child has done particularly well in (or badly) — in time to discuss at dinner each evening.
Changing the economics of education
The scenario described above would lead to a large fall in the cost of providing education. Labour is the largest cost in the provision of education. Physical schools are bound by constraints on productivity growth in teaching. A teacher is able to teach (and mark the work of) the same number of students today as a hundred years ago.
Low productivity growth implies an increasing cost of labour over time i.e. Baumol cost disease. But virtual schools fundamentally change the economics of schools. While virtual schools will entail high up-front costs in collecting data and training LLMs, they have no need for a large workforce of teachers or a bricks and mortar school.
Given their scalability, firms will be able to spread the fixed costs over a larger number of students such that a virtual school may have hundreds of thousands or millions of students enrolled. The increase in variable costs for enrolling each additional student enrolled in a virtual school will be small: that of hosting the student on a server.
We could therefore expect the cost of educating a child to decrease from the £7,500 (2) it costs the British state today each year to the price of an annual app subscription like Spotify or YouTube.
State regulation rather than state provision
If high quality virtual education becomes affordable for all then the state would no longer need to provide education based in physical schools. LLMs therefore enable education to be provided by free markets. The role of the state can be limited to mandating enrolment in a virtual school.
While the key role of regulating the quality and monitoring compliance will continue, it becomes easier as virtual schools are much more transparent than bricks and mortar schools.
For parents, this form of privatisation represents a liberation from the constraint of choosing from bad schools in their catchment area or paying the housing premium of a good school. It also liberates their children from having to be in a certain place at a certain time (a ‘time-place’ mandate) “5 days a week, 7 hours a day, 9 months a year, for 12 years” (3).
An App Store-like marketplace of virtual schools restores the sovereignty of parents in deciding the education they want for their children. This might involve integrating work experience, national service, visits to other countries, outdoors experiences like camping into the curriculum — whether real or virtual.
Or it might mean a yet more traditional education which condenses the years of schooling while going into greater depth than current government specifications.
Such an expansion of genuine choice in education is to be welcomed by parents and children.
Sources
(1) https://sunyan.substack.com/p/the-economics-of-large-language-models
(3) https://www.richardhanania.com/p/the-old-school-reformers-case-for
Scotland’s Futures Forum exists to encourage debate on Scotland’s long-term future, and we aim to share a range of perspectives. The views expressed in this blog are those of the author and do not necessarily reflect the Futures Forum’s views.