Who decides over AI?

AI is not built by «society» in general. It is built by specific actors, with specific interests.

Who	What they do	What they want
Large technology companies	build the biggest models	make money, be first
Researchers	develop the methods	understand, publish
The open source community	free models and tools	accessibility
Politicians and authorities	set the rules	protect citizens
Users	use it, give data	get a benefit
Those affected	are affected by the decisions	often not asked

The last row is the most important. Somebody who gets an automatic rejection of an application has rarely had any say about the system.

Three questions that recur in every AI debate:

1. Who makes money?
2. Who pays the price?
3. Who gets to decide?

And those three groups are rarely the same people.

Who sets the rules in practice?

The level	What
The EU	the AI Act, the GDPR, the DSA — the most far-reaching rules in the world
Sweden	applies the EU rules; IMY and other authorities supervise
The companies themselves	terms of use, content policies — they affect everyday life the most
The school	what may be used in the lesson
You	what you choose to use and share

The companies' own rules govern the most in practice. A policy change at a large platform affects millions of people immediately, while a change in the law takes years.

Three tensions with no simple answers:

The tension	Arguments for both sides
Fast development against caution	the benefit comes sooner — but so do the mistakes
Open against closed models	open gives insight and competition — but also misuse
National rules against global technology	rules protect — but the development happens somewhere else

The concentration of power is the underlying question. Training the biggest models requires resources only a few actors have. That means a few companies make decisions affecting very many people — about what the models may say, who gets access, and what they cost.

The counterweights exist: open models, public funding of research, and regulation. How strong that counterweight becomes is a political question, not a technical one.

Your role is larger than it feels. You can choose which services you use, read what you agree to, speak up when something feels wrong, and — most importantly — understand enough to be able to take part in the discussion. That last one is in fact what this platform is about.

Hold a structured discussion in the class. Four statements, 15 minutes each.

How to do it: split into two groups per statement. One prepares arguments for, the other against — regardless of what you yourselves think. Then swap sides and argue the opposite. Finish with what you actually think.

#	The statement
1	«Large AI models ought to be open for anybody to download»
2	«Schools should ban AI tools in exams»
3	«It should always be evident when something is made by AI»
4	«Companies that profit from AI should pay tax that funds the transition»

Arguments to keep in the back of your mind:

Statement 1, for: insight, competition, research, no single actor decides. Against: it cannot be taken back, misuse becomes easier, safety work becomes harder.

Statement 2, for: the exams measure what the pupil can do, not what the tool can do. Against: the tools exist in working life, bans are hard to enforce, better to change what is examined.

Statement 3, for: fundamental for trust, already a legal requirement in the EU for synthetic content. Against: where is the line — spellchecking? translation? — and how is it checked?

Statement 4, for: the profits are concentrated, the costs are spread. Against: hard to define, risks moving the activity elsewhere.

Finish individually. Write a few sentences:

• Which statement did you change your mind most about?
• Which argument from «the other side» was strongest?
• What would you need to know more about to be more certain?

The last question is the most important. Being able to say «I do not know enough about this» is a strength, not a weakness — and it is often what separates a considered opinion from a loud one.