mxx1's avatar

AI and Education

The central discussion about AI in education, and LLMs in particular, revolves around three questions:

Do we use AI to design better learning, or merely to produce cheaper content?

Using AI to generate educational content is such an obvious shortcut that almost everyone in education is doing it, whether they admit it or not.

Eventually it will become apparent that students, who know nothing about a subject, could use AI to generate material resembling that produced by their teachers and institutions.

If anyone can manufacture the content, what is special about the content or its delivery? Are students paying for quality learning, or merely for the credentials of the teacher and institution?

Already, in most subjects, a student using GPT, Claude or Gemini could get a better and cheaper education than most university’s deliver, so long as they are prepared to dig and dig.

What might be missing though with this approach is personal motivation, the social cohort and the qualifications.

My thinking is that there will be a special form of structured LLM pedagogy that controls sequencing, prerequisite testing, retrieval practice, application, feedback, review and testing, and progression.

The research question then becomes whether an LLM-mediated instructional architecture can equal or outperform conventional teaching for defined learning outcomes.

That question of the use of AI to build content should be resolved by evidence. Educational research needs to establish whether AI-generated content is good enough, or not.

And, of course, there will be better and worse ways to get AI to generate and deliver content – so we need to know what these are.

Should we train students for jobs that AI will replace?

Firstly, you don’t just learn things so you can use them to generate income. Unless of course you live in Australia and your parents sent you to a private school.

Here are some other reasons to learn things;

  • For interests sake.
  • So you can become a global expert and then use creativity and innovation to extend the field.
  • So you increase your base capability and can then take on even harder subjects.
  • To keep all those educators employed.
  • To make better decisions. Knowledge improves judgment in areas where you may never become an expert, including finance, health, politics, technology and law.
  • To avoid being manipulated. The more you understand many subjects, the harder it is for advertisers, politicians, institutions, consultants or AI systems to successfully fool you by substituting rhetoric for substance.
  • To communicate with experts. You often need enough knowledge to ask the right questions, understand the answer and know when someone is avoiding the issue.
  • To connect fields. A lot of innovation comes from carrying concepts from one domain into another. Breadth increases the number of possible useful combinations.
  • To understand the world you live in. Science, history, economics, philosophy and technology provide explanatory models for events that would otherwise look arbitrary.
  • To preserve intellectual independence. If all difficult thinking is outsourced, your conclusions become dependent on whoever built, trained, filtered or controls the tool you use.
  • To teach other people. You need internal understanding to explain something flexibly, diagnose another person’s misunderstanding and adapt the explanation.
  • To experience mastery. There is a distinct satisfaction in becoming genuinely competent at something difficult, independent of whether it produces income.
  • To preserve knowledge across generations. Some learning exists because societies need people who actually understand fields rather than merely have access to records about them.
  • To discover what you are capable of. You often cannot know whether you have aptitude for mathematics, music, engineering or languages until you learn enough to get past the introductory layer.

So my answer is, yes, we should train students even though AI may be all over the subject. And that is because there are plenty of reasons to learn that don’t depend on future income.

And it is in these areas that the 20W human brain can still shit on a computer system that consumes 1×10^6 times as much power: for example, originating genuinely new ideas, form independent purposes and turn deep expertise into creative and innovative contributions.

My assertion is that LLMs can augment this work, but not replace it.

We therefore still need to educate the top research students deeply enough to identify and develop that small handful that are capable of genuine creativity and innovation.

These students will still require deep knowledge, judgement and expertise. They will also need to become expert LLM wranglers.

But for all those vocational students just after an income, we need education systems to confront what AI can and will do.

Training students to perform tasks that machines will soon perform faster, cheaper and better is indefensible when the alternative is to teach them how to manage and extend AI systems within their professions.

The education system is not responsible for technological unemployment. However, it is responsible if it continues to teach obsolete capabilities after their obsolescence has become obvious.

Should students be allowed to use AI?

This is the dumbest question of all. Should students be allowed to learn how to use and manage one of the most important technologies ever developed? D’oh.

What I love about AI is that it has exposed the laziness of many educators. Old pre-LLM content can be reproduced instantly by an LLM because much of their teaching consists of little more than transmitting information and testing rote memory. They have not used pedagogy to work out how students become knowledgeable, capable, creative experts.

Instead, they have relied on tired content, standardised delivery methods and institutional authority.

The more threatened they are by AI, the more likely it is that AI can already do most of what they were doing.