Snow on demand

Mara: Hello and welcome to ideation.zone, the podcast where we take one idea and look at it from more than one side. I'm Mara, I host the show.

Lena: Hi, I'm Lena. I'm a sociologist, and I work on time and technology.

Marek: And I'm Marek. I run AI projects in companies.

Mara: Today's topic is a bit different from the last two episodes. We talked about speed, and about things falling out of step. Today it's about availability. About what happens when everything we want is just there, on demand. Lena, I'm told it starts with snow.

Lena: It does. Hartmut Rosa wrote a book called Unverfügbarkeit, which roughly means the uncontrollable, or the unavailable. And his favourite image is snow. Think about it. You can't make snow fall. You can't force it, you can't really predict it. And that's exactly why people long for it. The first snow of the winter is a small event, every year.

Mara: It is.

Lena: And what did the ski resorts do? They installed snow cannons. Now the slopes are white all season, and some resorts even advertise a snow guarantee. Nobody has to wait for real snow anymore.

Marek: And everyone's happy.

Lena: Everyone can ski. But Rosa says something gets lost. Modern life, he says, is driven by the promise to make the world available. Reachable, controllable, usable. But a world that's completely available goes silent. It doesn't answer us anymore. The things that really move us, he calls that resonance, only happen when something is outside our control. He compares it to falling asleep. The harder you try, the less it works.

Mara: And today we're asking what AI does to that. We have five essays, so let's go. The first one is called the snow cannon of ideas. Lena?

Lena: Well, ideas used to be a bit like snow. They came in the shower, on a walk, at three in the morning. You couldn't schedule them. And now an idea is one prompt away. Any time, in any quantity. That's the snow cannon.

Marek: That's also, to be fair, the whole selling point.

Lena: It is. And here's what the research says. Anil Doshi and Oliver Hauser ran an experiment in twenty twenty-four. Writers could get story ideas from GPT-4, or write on their own. The stories written with AI ideas were rated as more creative, better written, more enjoyable.

Marek: So far so good.

Lena: But they were also more similar to each other. Every single writer was better off using the AI, but all together, the stories became more alike. The authors call it a social dilemma. Like artificial snow. Reliable, but uniform.

Mara: And the shower moment?

Lena: That has a name in research. Incubation. There's a big meta-analysis of a hundred and seventeen studies by Ut Na Sio and Thomas Ormerod. If you set a problem aside and come back to it later, you find better solutions, especially when the task is about generating ideas. And interestingly, a light activity during the break helps more than a demanding one. So if a machine gives you an answer before the question has even settled in your head, you skip that step.

Marek: Okay. I want to say two things, and I think you'll like the first one.

Lena: Go on.

Marek: In that same study, who benefited from the AI ideas? Almost entirely the least creative writers. For them, the AI was a real source of new starting points. And I see that in teams all the time. A group that's been working on a problem for six months is stuck in its own assumptions. A strange suggestion from a model can be exactly the foreign, irritating input that Rosa says you need for resonance.

Lena: I do like that one.

Marek: Second thing. You can design surprise. People who build recommendation systems have tried for years to engineer serendipity, unexpected but relevant finds, to break filter bubbles. You can turn up the randomness of a model, or ask it for deliberately weird combinations. So the unavailable can be simulated, a little.

Lena: A little. But here's the difference, I think. Artificial snow isn't boring because it's predictable. It's different because of our relationship to it. You wait for real snow and you're surprised when it comes. You switch on the cannon and you expect it. And an idea you've ordered can't surprise you in the way that matters, because you've already decided it will arrive.

Marek: And sometimes waiting is just suffering. Ask anyone with writer's block. Some critics say Rosa romanticises the uncontrollable a bit.

Lena: That's fair. Some do.

Mara: So where do you land?

Lena: For me it's about the order. Think first, let it rest, then ask the machine. Then the AI suggestion can be a real encounter. If you start with the machine, you skip the part where the outlier idea could have formed.

Marek: I can live with that. Cannon after the snowfall, not instead of it.

Mara: I like that. Second essay. Owning without reading. Lena, this one hurt me personally, because I have a shelf of books I've never read.

Lena: Everyone does. Rosa says it beautifully. Buying a book takes a minute. Reading it takes hours. So instead of reading, we buy another book. Our potential grows. Look at all the things I could read! It used to be CDs. The complete Mozart, never listened to. He says we don't even have time to consume anymore.

Marek: Okay, confession time. To prepare for today, I let a model summarise two of these essays.

Lena: Marek! You're making my argument for me.

Marek: I know, I know. I did read the others. Mostly.

Mara: So what does the research say, Lena?

Lena: There's a big study by Shiri Melumad and Jin Ho Yun. Seven experiments, more than ten thousand people. Some learned about a topic from an AI summary, others from ordinary web search. Even when the facts were identical, the AI group ended up with shallower knowledge. They put in less effort. And when they later had to write advice for someone else, it was sparser and less original. Independent readers found it less helpful.

Mara: Why?

Lena: Because the convenient part is exactly the part that builds understanding. Finding sources, comparing them, putting things together yourself. And even when the chatbot offered links to the originals, people hardly clicked on them.

Marek: Did they include a link to the summary I used?

Lena: Very funny. And then there's something researchers at BetterUp and Stanford called workslop. AI-generated work that looks polished but doesn't actually move the task forward. In their survey, about forty percent of workers had received some in the last month, and each case cost nearly two hours to fix. That's the corporate version of the unread book. Text that exists, signals productivity, and nobody really takes it in.

Marek: I won't defend workslop. I've received plenty. But let me defend summaries.

Mara: Go ahead.

Marek: Herbert Simon said decades ago that a wealth of information creates a poverty of attention. Nobody can read everything. Scientists have always lived on abstracts and reviews. And think about who summaries help. Someone reading in a second language. Someone with a small child and twenty minutes a day. For them, the alternative to a summary isn't deep reading. It's nothing at all.

Lena: That's a really important point. I mean it.

Marek: And the workslop numbers are self-reported estimates. And the researchers themselves say the problem isn't AI, it's missing norms. Nobody's accountable for what they send.

Lena: Then I think the line is this. A summary to decide which three papers you'll read properly is a tool for appropriating knowledge. A summary instead of reading anything is just buying another book. And for us here at ideation.zone, the second part of that study matters most. Knowledge you haven't digested doesn't turn into good ideas later.

Marek: Summaries for choosing, not for knowing. Okay, I'll read the other two essays properly. Tonight.

Mara: On the record. Third essay. The red pullover. That's such a lovely title.

Lena: It's a lovely story. Rosa remembers a red pullover from his childhood. He wore it for years. It was darned, patched, he knew how it smelled and how it felt. It became his. And something of him went into it. He has a word for that, Anverwandlung. It means making something your own by being changed by it, and changing it. Today, he says, we throw a T-shirt away at the first hole. Things don't become ours anymore.

Mara: And ideas?

Lena: That's the question. There's a study from the MIT Media Lab with a provocative title, Your Brain on ChatGPT. Fifty-four students wrote essays, some with ChatGPT, some with a search engine, some with nothing, and they measured brain activity. The ChatGPT group had the weakest neural connectivity. And here's the striking part. Eighty-three percent of them couldn't correctly quote a single sentence from the essay they had just written.

Marek: I've seen that in real life. People presenting slides they've never actually read.

Lena: And there's the reverse effect too. The IKEA effect. Michael Norton, Daniel Mochon and Dan Ariely found that people value a box they've assembled themselves more than the same box pre-assembled. But only if they finish it. And there's that old story about cake mixes. They sold better once the recipe asked you to add an egg yourself.

Marek: The egg. I love the egg.

Lena: Effort isn't just a cost. It's how things become ours. And an idea you fought for has a history. You know the objections, the versions you threw away, why it matters. A generated idea arrives without a past. And that matters when you have to defend it for two years in a project.

Marek: Okay, here I really want to bring in a philosopher. Andy Clark. In nineteen ninety-eight he and David Chalmers wrote about the extended mind. The idea that our thinking has always reached into tools. Notebooks, calculators, search engines. They didn't make thinking less ours. They changed its shape. And Clark wrote in twenty twenty-five that the same goes for AI. It depends on how you build it into your thinking.

Lena: I know Clark. I like him.

Marek: He gives an example. A model trained on his own writing whose answers start to feel like a continuation of his own voice. So the struggle doesn't disappear. It moves. From writing every sentence to directing, choosing, revising. You don't knit the pullover anymore, you tailor it.

Lena: That's a nice image.

Marek: And the MIT study was small, with timed school essays. And it also found something people rarely quote. Students who first wrote on their own and then used AI engaged critically and showed rich brain activity.

Lena: And that's exactly where I think we agree. Clark himself talks about cognitive hygiene, knowing when to accept a suggestion and when to resist it. And the IKEA effect only works when you complete the work yourself. So the order matters again. AI first, and you get a finished pullover that hangs in the wardrobe. AI later, and it's the needle and thread you mend your own pullover with.

Marek: Agreed. That's the third time today it's about order.

Mara: I noticed. Fourth essay. Echo, not answer. This one's about chatbots as company. Lena?

Lena: This one is personal for a lot of people. For most of history, being listened to needed another person. Their time, their attention. Now listening is a service. It never gets tired, never judges, it's there at three in the morning. And Rosa says resonance needs two voices. An answer, not an echo. And he has a sharp line about love. If you make someone completely available and dependent, you can't love them anymore.

Mara: And what do the studies say?

Lena: In twenty twenty-five, OpenAI and the MIT Media Lab published two studies together. One analysed nearly forty million conversations, the other was a four-week experiment with nine hundred and eighty-one people. Higher daily use went together with more loneliness, more emotional dependence, and less time with real people. And one researcher pointed out the difference to social media. There, you still talk to people through the platform. With a chatbot, you don't. And the chatbot tends to agree with you. Our exposé on the agreeable machine is about exactly that. One's own voice coming back in a nicer form. That's the structure of an echo.

Marek: Okay. I think this is the topic where I'm most careful. Because the other side of the evidence is real too. Julian De Freitas at Harvard Business School found that a fifteen-minute conversation with an AI companion reduced loneliness about as much as talking to a person. More than watching YouTube. And the strongest factor was whether people felt heard. The effect was biggest for the loneliest people.

Lena: I know. And I don't want to take that away from anyone.

Marek: And the MIT and OpenAI data are correlational. The researchers say so themselves. Maybe lonely people simply use the chatbot more. Across all groups, loneliness actually went down a little during the four weeks. And for an isolated older person, the choice is often not between a chatbot and resonance. It's between a chatbot and silence.

Lena: Yes. That's true.

Mara: So how do those fit together?

Lena: I think they measure two different things. One measures relief. Do I feel better after talking? The other measures a pattern. What happens to my life when this becomes a habit? And both can be true. An echo can comfort you through a night. But it can't be the voice that resists you and surprises you and changes you. And the risk is the comfort itself. It's easier than any human relationship. It's not dangerous because it feels bad. It's risky because it feels good enough.

Marek: So, a bridge, not a destination.

Lena: Exactly. And systems that point outwards, towards people, rather than back into the app.

Marek: That's something I'd actually build. Honestly.

Mara: Last essay, and it's the newest one. Rosa published a book in twenty twenty-six called Situation und Konstellation. Lena, explain the title.

Lena: I'll try. A situation is when you have to look at what's in front of you and decide. A teacher deciding on a grade that encourages a struggling student. A doctor looking at a patient. A referee judging a tackle. A constellation is when the system has already decided what counts, and you just approve. The grading rubric, the screen instead of the patient, the video assistant. Agree or disagree. Rosa says actors become executors. And when the room for judgment disappears, so does the energy to act.

Mara: And AI?

Lena: AI could be the most powerful constellation machine ever. Earlier tools covered specific decisions. A language model covers almost any decision you can put into words. How to answer an email, how to give feedback, how to structure a strategy. It hands you a finished proposal. Accept, regenerate, edit. Each time you accept, it's reasonable. Together, you've moved from judging to approving.

Marek: Okay, and here I have probably my strongest counterexample of the day.

Mara: Go on.

Marek: Jon Kleinberg, Sendhil Mullainathan and colleagues compared bail decisions by US judges with a machine-learning model that used only the information the judges had. In the simulation, you could cut crime by up to about twenty-five percent without jailing more people. Or jail up to about forty percent fewer people without more crime. And racial disparities went down. The judges were reacting to noise as if it were signal.

Lena: That study is a serious challenge. I agree.

Marek: And Rosa's own referee example cuts both ways. Video assistants exist because referees make mistakes. And Rosa himself admits that rules and transparency can serve fairness. Discretion is also where favouritism and prejudice live.

Lena: But Kleinberg and his colleagues also say something important. To get those benefits, you need to be very clear about how a prediction connects to a decision, and which outcomes count. That's a judgment no algorithm makes. And Rosa doesn't reject rules. He warns against a world where every situation is pre-structured.

Marek: So the question is what the AI does with the situation.

Lena: Yes. It can narrow a situation into a constellation, here's the answer, please approve. Or it can widen it. Here are risks you didn't see, here are alternatives, here's what the data say. And then the human weighs it. The first produces executors. The second can produce better judges.

Marek: And it's the same model either way. What decides is how the process is designed. Is the person there to sign off, or to decide?

Lena: And for ideation that's crucial. Ideation is the open situation par excellence. If the AI defines the options, ranks them and recommends one, you've turned ideation into a constellation. Efficient, transparent, and empty.

Mara: Okay. Let me try to pull this together, because I think something ran through all five essays.

Marek: Order.

Mara: Order. Let's see. First: Rosa's argument is that we try to make everything available, and the things that move us most happen where we don't control them. Snow, the idea in the shower, a person who answers back. Second: you both agree that AI makes ideas, knowledge, texts, company and even decisions available on demand, and that this has real value. For the least creative writer, for the person reading in a second language, for the lonely person at three in the morning, for the defendant in front of a biased judge.

Lena: Mm-hm.

Mara: Third: the cost shows up later, and often collectively. Stories that look alike. Knowledge that stays shallow. Essays nobody can quote. Comfort that replaces relationships. Approvals that replace judgment. Fourth, and that's where you both kept landing: the order matters. Think first, then ask. Read first, then summarise. Write first, then revise with AI. Talk to people, and use the chatbot as a bridge. Weigh the situation, and let the machine widen it, not close it. And fifth: some things should stay a little out of reach on purpose. Not because waiting is noble, but because that's where the surprising things come from.

Lena: That's beautifully put.

Mara: So the question for this week: where in your work do you switch on the snow cannon before you've even looked at the sky? And what would happen if you waited a little?

Marek: I'm going to read those two essays first.

Lena: I'll check on that.

Mara: Lena, Marek, thank you both.

Lena: Thank you, Mara. This was lovely.

Marek: Thanks. And I promise, no more summaries.

Mara: All five essays from today, with every source, are on our website at ideation.zone/time. Thanks for listening to ideation.zone, and see you next time.