Neuvottelija.AI

EP394 · Society · first published 2026-07-09

Europe as Easter Island | Risto Linturi, Kari Angeria | Negotiator 394

This is a summary on Neuvottelija AI. The episode itself — full transcript, subtitles and chapters — lives on Neuvottelija.com, which is its canonical home.

Technology futurist Risto Linturi and Kari Angeria join Sami Miettinen to work through their book Elämän tarkoitus ja tekoäly. The question is not when AI surpasses people but what is left for them afterwards: where meaning comes from, where work comes from, and above all where power comes from. Linturi describes the 2,246-page technology report he curated for the Finnish Parliament’s Committee for the Future — written to be read by an AI rather than by a person — and makes two uncomfortable observations: withdrawal is possible but costs independence, and rebellion requires that the rebel’s labour still matters. Against that, concrete proposals — a five-robot quota, universal ownership — and practical lessons on why models hallucinate and how to stop them.

Sami Miettinen · Sections: AI and Society + AI and the Economy + Tools and Implementations

Europe as Easter Island | Risto Linturi, Kari Angeria | Negotiator 394

Summary: Technology futurist Risto Linturi and Kari Angeria join Sami Miettinen to work through their book Elämän tarkoitus ja tekoäly. The question is not when AI surpasses people but what is left for them afterwards: where meaning comes from, where work comes from, and above all where power comes from. Linturi describes the 2,246-page technology report he curated for the Finnish Parliament’s Committee for the Future — written to be read by an AI rather than by a person — and makes two uncomfortable observations: withdrawal is possible but costs independence, and rebellion requires that the rebel’s labour still matters. Against that, concrete proposals — a five-robot quota, universal ownership — and practical lessons on why models hallucinate and how to stop them.

The guests

Risto Linturi is a technology futurist who has written radical-technology assessments for the Finnish Parliament’s Committee for the Future since the 2010s. Kari Angeria runs Neuvottelija’s AI working group with Sami Miettinen — nine people and nine bots, each bot with its own name and its own accumulated competence. Together they wrote the book Elämän tarkoitus ja tekoäly (“The meaning of life and artificial intelligence”).

The question in this episode is not when AI surpasses people. That is taken as given. The question is what is left for people afterwards — and the hour divides into three answers, about meaning, about power and about work. The middle one is why this is worth reading.

1. The method deserves its own section

Before the content, look at how the book and the report were made, because the method explains the conclusions.

The book: one meaning at a time, tested against history

Angeria’s description of the method is unusually clear. Instead of asking “what is the meaning of life”, they enumerated what people in different eras have experienced as their purpose — under slavery, or as a Sumerian king of Uruk who already had everything in abundance — and tested each against a world of AI and robots: does this meaning survive when the machine does it better?

That is an inventory method rather than a philosophical argument, and it is checkable for exactly that reason: a reader can disagree about one item without the structure collapsing.

Angeria also says his own starting point changed mid-project. He began technology-positive — “prosperity increases, this works out for everyone” — and wrote himself into a different position. That matters to the reader: the book’s more pessimistic tone is the result of the work rather than its premise.

The report: 2,246 pages, written to be read by a machine

This is the most original single thing in the episode, and it deserves stating precisely.

The Committee for the Future material Linturi curated as second author runs, he says, to 2,246 pages, containing roughly 20,000 separately itemised and assessed threats or opportunities and thousands of source works. Written out for a lay reader it would be around 20,000 pages. The previous report, from 2018, was about 500.

And then the point: it is not written to be read by a person. It is stored in a relational structure where items and the links between them are weighted — which connection matters and which matters less — so that an AI can traverse everything relevant to a given question and produce summaries and cross-checks.

The justification is density rather than technology: when the material does not have to be compressed to human reading capacity, nothing essential has to be left out. Linturi says no significant bottleneck they found was omitted.

The structure has two levels and it is the most useful single tool in the episode:

  1. Twenty value-creation networks — human needs grouped (movement, nutrition, health, the desire for power, meaningfulness…), each with the values being sought. In movement, for instance: a person from one place to another, preferably safely, without emissions, on time and comfortably.
  2. The hundred most important technological bottlenecks — things that are too expensive, too inefficient or otherwise limiting, and where development is expected.

When a bottleneck opens, the structure tells you which value networks it touches — which is to say which parts of the economy, of daily life and of the statute book. Anyone can use it, and it is the clearest transferable thing here.

The hit rate — and how to read it

Linturi says the 2013 report’s forecasts have been re-scored after the fact: of a hundred, two or three are still in the hype phase and 97 have clearly advanced. The more important claim is that the ordering held: the quartile judged most important advanced faster than the next quartile, which advanced faster than the one after that.

That second claim is methodologically stronger than a hit rate, because it tests the scoring rather than merely the direction. Both, however, are the authors’ own retrospective assessment of their own work, and the episode does not say who did the scoring or how “clearly advanced” was defined. The claim is interesting and checkable — but it is not checked here.

2. Power — the hard core of the episode

This section is worth reading even if AI discussion bores you, because it is a description of a mechanism rather than a forecast.

The observation everything follows from

Linturi cites a statistic whose source he cannot recall exactly: at least half of Americans are people whose income development no longer matters to national output. Companies can go on producing more and more profit even if that half disappeared entirely.

The figure is marked here as unverified. But the reasoning built around it does not depend on the exact percentage, and it is the strongest argument in the episode:

Historically, an ordinary person’s power has rested on their participation being necessary. A worker could strike because production stopped without him. An Indian could practise passive resistance against British rule because the administration needed the cooperation of the governed. A consumer’s opinion mattered because the firm lived on his purchases.

Remove that necessity and the lever disappears — not prohibited but redundant. You cannot strike in a robot-staffed factory. You cannot practise passive resistance against robot police. And if a voter’s information arrives through a feed somebody else curates, the vote no longer connects to decisions in the same way either.

The argument is structural and can be made without reference to AI at all: it is about what a citizen’s bargaining power has always rested on. It can be answered — for example, that the legitimacy of government has never been purely economic — but no answer appears in the episode.

The price of withdrawal

The other half of the argument concerns the option that is always available: refuse. Three cases are worked through, and the treatment is honest because it concedes that the option works before naming its price:

Then the price, and it is specific: if Finland made the same choice, it would not be buying medicines, hospital supplies, fighter aircraft, drones or artillery shells. And because the productivity gap to competitors would widen while debt grows, this would not merely be impoverishment:

“It won’t be American gunboats. It will be Russia walking across the border.”

The same logic is applied to Europe’s regulatory road, and here the panel is at its harshest about its own continent: AI regulation, server requirements, taxing tokens and subsidising human labour are described as a losing path, because this is a competitive game between states. The consequence they draw is unusual and therefore worth marking: states would begin joining the United States one at a time — rather than the reverse.

That is a prediction, not an observation, and its counter-case — that regulation can also create competitive advantage or prevent real harms — goes unaddressed. I mark it here because the episode does not.

Two proposals that are not the familiar ones

The episode does not stop at complaint. Linturi puts forward two structural proposals, both rarer than a basic income.

1. A five-robot quota. Every citizen would hold a certain number of humanoid robots — say five — over whose physical operation they hold a veto. The robots would be rented to companies for a day, a week or years, so economies of scale survive; but if an employer had them do something the owner would not accept, the owner could withdraw them.

Linturi stresses two things, and both answer the obvious objections:

The point is that physical power is not the same as software power: anyone can run unlimited software agents on their own hardware and no police force can prevent it, but a robot is an object whose operation has a location and an owner.

Sami Miettinen challenges it directly: restricting through scarcity is a hard road in a world where everything else scales without limit. The disagreement is left open.

2. Universal ownership instead of a basic income. The justification is not welfare but power. In a basic income somebody gives, and the episode answers that coldly: who gives, what is it earned from, and what apparatus of force distributes it? Under ownership, money and decision-making circulate without anyone having to dispense goodwill.

Miettinen challenges this too, and his counter is the sharpest thing in the episode: if the army, the factories and the police are robots, and somebody can change their program code, it does not matter what anyone else thinks or what the law says. Ownership is ultimately a question of enforcement.

No answer comes. That is an honest outcome.

3. Meaning, responsibility and transhumanism

Why an AI cannot be responsible

The best conceptual passage, and it is short:

An AI wants nothing. It has no drives. And it cannot carry responsibility, because you cannot take revenge on it — it cannot be hurt or punished. You can switch it off, and that means nothing to it.

From which comes Linturi’s argument, which turns transhumanism into a safety question rather than a question of destiny: a human with AI integrated can carry responsibility, because they can be punished. Society can therefore give a human–AI pair more power than it can give a machine alone — and because there are many such pairs, power distributes at the same time.

For comparison Linturi cites the old line: to err is human, but for a real catastrophe you need a computer. Centralised optimisation makes its mistake once and everywhere; a distributed system makes many small ones.

Attached to this is the paperclip example — an AI asked to raise as much money as possible and produce as many paperclips as possible does precisely that. The example is familiar from AI-safety discussion; what the episode does with it is the useful part: the danger is not the machine’s will but its lack of one.

Harari, and a third way

Miettinen sets out two termini: transhumanism (the human is extended) and dataism (the human accepts being worse at everything and leans on the system). He says plainly that he thought Harari’s book good as a question and weak as an answer.

Linturi’s third way is the one above: an integrated human, because responsibility requires a body.

Voluntary androidisation

Miettinen’s own term, and he describes it as case-by-case rather than ideological: in some decisions the machine’s reasoning is superior and he defers to it, in others what decides is human judgement and the fact that somebody is answerable. He gives as an example having run the process on himself — a half-joking brief, “make me a tech bro in 30 days” — and says the model gives its user considerable power.

And then one very concrete point worth taking away:

An AI would probably decide social assistance cases more impartially than a person, because you can give it the criteria and it follows them — and being somebody’s neighbour does not affect it. Finnish law nevertheless requires a human decision-maker.

Attached is a qualification more important than the claim itself: the system decides, not the model. A free general-purpose chatbot with no memory structure is not the same thing as a curated system with memory built for it, whose mistaken inferences get cleaned out over time. Miettinen describes his own “intelligent notebook”, which he prunes afterwards — “I looked and thought, what a stupid idea I had in January.”

Meaning in extremity

The subtlest passage: a person will find meaning in anything if they have to — including slavery and the camps. That is not consolation but warning, and the episode uses it as such: that people adapt is not evidence that the conditions are good.

Set against it is the opposite extreme: the addicts of virtual worlds, whose life is enjoyment with nothing else in it — and the note that this is not the future but already partly the present.

4. On the technology — three things worth taking

Mid-episode there is a practical stretch useful to anyone working with models.

“Does it only predict the next word?”

Linturi’s answer is the best formulation of this I have heard:

How could an AI write German, where the verb comes at the end of the sentence, unless it knew the verb before writing the qualifiers at the beginning?

The argument is structural rather than metaphysical: producing language forwards requires a representation of where it is going. He adds an observation about reasoning models — they visibly hesitate, backtrack, discard their own proposal — and a dry remark: “when people say they don’t think, I’m not always sure who does.”

Miettinen continues in the same direction about hallucination: he says he knows plenty of people who repeat an outdated worldview regardless of the input.

Why a model hallucinates — and how to stop it

The most useful practical passage, and its core is conceptual:

A hallucination is often the right answer to the wrong brief. Ask for evidence that the moon is cheese and the model continues the genre it has been started in — and in literature the continuation of such a sentence is a fairy tale. The model is behaving exactly correctly; the contradiction is in the brief.

Three corrections follow, all reproducible:

  1. Don’t order it to find — ask whether there is. “Find me the evidence” compels evidence to be produced. “Is there evidence for this, and what is the fact” does not.
  2. Don’t say the idea is yours. If the system prompt tells the model to be supportive and constructive, owning the idea invites agreement. Put as an open question, the same thing gets an honest answer. Linturi notes that this is how science is supposed to work: evidence first, conclusion second.
  3. Use a reasoning model and a proper harness that prunes.

A language question arises as a side-path and the episode does not settle it: what is harness in Finnish? The candidates offered are suitset (reins), komentokeskus (command centre) and ohjaava pakkopaita (a steering straitjacket).

Androidisation in practice

Angeria’s example is the most concrete thing in the hour: his son built himself an AI coach, which he continuously specifies and improves himself. The hard part is not the model but the modelling of the data and its coherence — getting the tool to find the few real insights in a mass of it. After two and a half weeks the son’s verdict was that this is better than any human coach could be, because it will analyse every single training session tirelessly.

The observation Angeria draws from his own management career is the same thing inverted: in managing people there is a human dimension — fatigue, strain — that these do not have.

5. Economics: what to do with all this

The closing stretch is the most usable part for an investor, and it rests on one sentence:

Everything machines do by themselves becomes worthless. Once the capital is amortised and the volumes are vast, the price of machine work approaches zero — an infinite deflationary cycle.

From which comes a question Linturi says was under-served even in his own report: the report looked for opening bottlenecks, but less attention went to where new ones emerge. That is where the scarce resources are, and that is where value rises.

The example is instructive because it marks the difference: many invested in Nvidia, because the opening bottleneck was visible. How many invested in Samsung on realising memory would run short? The second is an emerging bottleneck. Miettinen admits he only bought expensive laptops.

This is the part of the episode you can use today, and it does not depend on what you think about AI.

Money for agents

A new structural question, handled briefly but precisely: when payment platforms begin giving agents the right to transact, and if an agent can own, who is there to punish? The human behind an agent network can become untraceable. Miettinen is not claiming people disappear; his point is that the frame changes — ownership, responsibility and liability no longer land on the same object.

SpaceX, and a question rarely asked

The closing scenario starts from a stock market listing whose valuation figure is stated in the episode rather than verified here. What matters is the argument, not the number:

Taken alone, SpaceX is a collection of disparate technologies. Together with Tesla it is a value-creation network end to end — compute, chip fabrication, space, communications, robots, passenger and freight transport — self-controlled, across several states.

Miettinen says he asked whether the United States could, in twenty years, subordinate such a combination. The analysis came back probably not, and the reasoning is more interesting than the answer: the administration would fall before antitrust legislation passed — and if the key individual were placed under house arrest, the robots would stop. All industry, commerce and transport would go on strike at once.

That is speculation and is marked as such. But it returns the episode to its own main theme in a way neither speaker states aloud: the strike did not disappear. It changed owner.

Claims presented as claims

What survives the episode

  1. A citizen’s power has always rested on their participation being necessary. The strike, passive resistance and the consumer’s voice are expressions of one mechanism — and the mechanism does not vanish by prohibition but by redundancy.
  2. Withdrawal is possible and it costs independence. Bhutan works; Bhutan does not buy fighter aircraft.
  3. Responsibility requires a body that can be punished. It is the best single argument for keeping a human in the loop — better than any appeal to humanity.
  4. A hallucination is often the right answer to the wrong brief — and the three fixes are available today.
  5. Ask where new bottlenecks emerge, not only which ones open. Nvidia was an opening bottleneck; memory was an emerging one.

How the episode runs

Markdown: index.md · Suomeksi