In the book Free Gifts, [[2025-12-09-free-gifts-review]] one of the concepts introduced was the Baumol cost disease.

Basically wages in sectors that do not increase productivity over time go down, while sector with constant technological improvements increase productivity and wages at the same time.

Health, educations, arts and culture are the most common examples of sectors that suffer from the cost disease. The reason is simple: It is really hard to automate human interactions.

Lately I have been re-examining my relationship with LLMs, surveillance capitalism and climate change. In my new rescoping rather than thinking of them as byproduct of some actor choice, I think of them as a property of the human system that can’t be changed individually most of the time.

[[2026-10-12-is-resilience-a-choice]]

LLMs are, in essence, a new way to externalise energy consumption for human purposes. The ability to elaborate symbols from other symbols with some energy means that humans can avoid doing some of the work of generating repetitive symbols.

I am not going to comment on the other problems of LLMs: power concentration, intellectual property appropriation and resource usage are not on the menu today (sorry). Instead I want to share a personal reflection on education.

Schools these days…

Sending kids to school is weird. My memories of going to school are blurry and my opinions on the system are based on glimpses of what happened then plus my parents opinions. And they had the same issue than I have now.

With that lack of objective framing it is really difficult to assess if schools are worse or better than when I was a kid. Certainly the ratio of kids seems similar, and the teachers seem underpaid as well. Probably worse just because there are less worker protections now than then (e.g.: zero hour contracts).

But one aspect that seems objectively different is technology indeed. I did not have any digital board in my school, the TV they had for special occasions was tiny and paperwork was manual most if not all of the time.

My parents could not get an email from the school because there were no emails, and every communication had certain amount of work attached to it (so less of it too).

Nowadays I get many emails, and a portal with privacy policy. The kids go to a school with hikvision cameras recording the common areas. Some assessments are digital, and some of the homework uses various degrees of automation, including AI tools.

I would say these are bad things. My suspicion is that human care and work is superior in most ways. I would rather got 1 email per month of a teacher that had enough time to think on it than 10 emails that say a lot of nothing. Or 5 minutes of dedicated teacher time will likely beat any tablet app no matter how good it is.

This goes all the way to university. Over my lifetime university got more expensive and there are more sponsorships and corporate mediation. These data points show a clear direction of travel of something that peaked well before I was born.

But in the same way that I was wondering if I really have a choice, I wonder if schools do really have a choice or can do anything meaningful about the system.

Governments are getting poorer over time and the fraction of total wealth dedicated to education keeps going down. I guess this is because of indebtedness, increase of resource costs, changes of populations, etc. So what can schools really do?

AI as a cure to the Baumol cost disease

Here is an idea: Baumol’s ideas are a reflection of how close certain sectors are to energy surplus growth. The more than a sector can benefit from converting energy into utility, the bigger it will become over time.

Here is a consequent of that idea: If education introduces more AI to their processes, will it be able to ride the energy surplus that we still have?

I think it is possible that the productivity per worker will increase at the cost of two things: Quality of education and number of workers.

During growth times these workers would be reallocated to do more care rebalancing the quality. But nowadays with a constantly shrinking education expenditure it will mean that schools will manage to survive a bit longer by reducing headcount.

And while I feel sad for the loss of quality for the future generations, I accept that perhaps it is how it needs to be while we live above our means.

If the thesis is correct we would see an increase of “corporatisation” of schools.

⚠️However this doesn’t mean that we shouldn’t fight for more accountability on AI and reduce inequality. It would be a bad outcome that the rich people get even richer just because there are less resources in general.

In my opinion all models that train on open data should be open weights, and monopolies should be destroyed rather than the main asset in our pensions.

*#llm #essay