文字起こし 🔗
WEBVTT
1
00:00:00.120 --> 00:00:08.620
In nineteen-thirty, Keynes promised his grandchildren a fifteen-hour week. We are the grandchildren. How is your week going?
2
00:00:10.400 --> 00:00:12.920
Fifty hours. But that's not the machine's fault.
3
00:00:13.520 --> 00:00:14.240
It never is.
4
00:00:14.860 --> 00:00:17.180
And that, I think, is where we start today.
5
00:00:26.560 --> 00:00:52.520
Hello and welcome to ideation.zone. I'm Elisabeth, I'm a science journalist in Vienna, and this season we're asking one question from several sides: what does artificial intelligence do to our time? Our starting point is the work of the sociologist Hartmut Rosa – and I have two guests who know that work, and its weak spots, very well.
6
00:00:53.180 --> 00:01:00.700
Hello, I'm Krzysztof. I teach sociology in Warsaw, and I study time – mostly why there is never enough of it.
7
00:01:01.160 --> 00:01:09.680
And I'm Francesca. I lead the AI team at a tech company in Amsterdam. So I'm the one who installs the thing Krzysztof is worried about.
8
00:01:10.500 --> 00:01:28.140
Good, then the roles are clear. Today it's speed. AI is sold as a time machine – it writes the memo, sums up the report, drafts the code. Five topics, and I'll warn you now: we will not agree on all of them.
9
00:01:28.720 --> 00:01:31.200
I hope not. That would be suspicious.
10
00:01:31.660 --> 00:01:38.500
Before we start – Krzysztof, for anyone who has never heard of Rosa. Two sentences.
11
00:01:39.020 --> 00:01:57.060
Two sentences. Hartmut Rosa says that modern societies can only stay as they are by constantly growing, speeding up and innovating – they have to run to stand still. And the strange result is that the faster our machines get, the less time we feel we have.
12
00:01:57.600 --> 00:01:59.500
That was three sentences.
13
00:02:00.980 --> 00:02:03.540
I'm a professor. That's very disciplined for me.
14
00:02:08.080 --> 00:02:14.580
Let's start with the simplest example in Rosa's work. Krzysztof, the letter and the email.
15
00:02:15.200 --> 00:02:41.460
It's almost too simple. Forty years ago, a letter took you maybe ten minutes. An email takes five. So if you used to write ten letters a day, you should have gained – what, fifty minutes? And what happened? You now write twenty emails. Twenty matters, twenty people waiting for an answer. The time per task went down, and the number of tasks went up faster.
16
00:02:42.080 --> 00:02:44.640
And Keynes? You opened with him.
17
00:02:45.300 --> 00:03:27.280
Keynes was right about the economy and wrong about people. He said productivity would grow so much that three-hour shifts would be enough. The productivity came. The three-hour shifts didn't. We took the gains as more – more goods, more services, more emails – not as more time. And, uh, there's a lovely older example. The washing machine. Historians asked why women with washing machines did not do less housework. The answer: the standards went up. Clothes were washed after one day instead of one week. The machine saved time per shirt, and then everyone wore more clean shirts.
18
00:03:27.320 --> 00:03:28.380
Okay, but –
19
00:03:29.020 --> 00:03:32.860
AI is a washing machine for text. That's all I'm saying.
20
00:03:34.280 --> 00:03:52.320
That's not all you're saying, it's never all you're saying. Look – the savings are real. I can show you a big public example, not ours. The British government gave an AI assistant to about twenty thousand civil servants for three months. On average they said they saved twenty-six minutes a day.
21
00:03:53.040 --> 00:03:53.540
Said.
22
00:03:54.100 --> 00:04:07.820
Said, yes. Self-reported. Fair. And – here's the honest part – more than a fifth of them said they saved less than five minutes. So it's not magic. But for the people who drafted a lot, it was real time.
23
00:04:08.540 --> 00:04:13.340
So where did those twenty-six minutes go? That's what I'd want to know.
24
00:04:14.100 --> 00:04:28.440
Nobody asked them. That's the – that's the problem. We measure the minutes saved, we never measure what happens with them. In our own rollout I could tell you how many prompts people sent. I could not tell you whether anyone went home earlier.
25
00:04:29.040 --> 00:04:33.120
Because nobody wants to know. If you knew, you would have to decide.
26
00:04:33.680 --> 00:04:59.440
And it's not only work. There's a Swedish economist, Staffan Linder, who wrote a book in nineteen-seventy called "The Harried Leisure Class". His point was simple and a bit cruel: the more productive our working hours become, the more expensive every free hour feels. So we try to make leisure productive too. We buy the faster bike, the shorter holiday with more in it, three series at once.
27
00:04:59.460 --> 00:05:00.300
Guilty again.
28
00:05:01.900 --> 00:05:06.540
And now we have AI to plan the holiday, so we can fit in a fourth series.
29
00:05:07.160 --> 00:05:13.580
Francesca, from your side – do people in your company feel they have more time since the rollout?
30
00:05:14.180 --> 00:05:31.620
Honestly? They feel they get more done. That's not the same. When we asked our teams after six months, the most common answer was something like: "I finish the boring part faster, so I start the next thing earlier." Nobody said: "I go for a walk now."
31
00:05:32.260 --> 00:05:45.080
Let me try a thought experiment on both of you. Your company announces: thanks to AI, everyone gets one hour a day back. Officially. What happens in the first month?
32
00:05:45.700 --> 00:05:55.720
In the first month? People are happy. Some leave at four. Some use the hour for the things they never had time for – the documentation, the training.
33
00:05:56.300 --> 00:05:57.280
And in the third month?
34
00:05:58.280 --> 00:06:05.440
In the third month somebody notices that the team at four o'clock is "less available" than the team that stays.
35
00:06:06.216 --> 00:06:27.696
Exactly. The hour is not taken away by a decision. It leaks
Load full transcript
away. Because being busy has become a sign of importance. There is a British sociologist, Jonathan Gershuny, who showed this with time diaries – a hundred years ago the upper class showed off its leisure. Today the people at the top show off how busy they are. Busyness is the badge of honour.
36
00:06:28.676 --> 00:06:34.276
I'll give you that one. In tech, "I'm so busy" is basically a status update.
37
00:06:34.976 --> 00:06:36.916
So – one point for Krzysztof?
38
00:06:37.556 --> 00:06:46.276
Half a point. Because the hour only leaks away if nobody protects it. That's a management question, not a law of physics.
39
00:06:46.856 --> 00:06:49.956
Management is where laws of physics go to be ignored.
40
00:06:54.536 --> 00:07:22.736
That brings us to the second essay, and to companies. In twenty twenty-five, a large share of executives said quite openly that fear of missing out was driving their AI spending. And in the same year a report from MIT found that most company AI pilots had produced no measurable return. Francesca, you're inside one of these companies. Is that fear?
41
00:07:23.356 --> 00:07:38.416
Partly. I'll be honest. When a board asks for an AI strategy, the question underneath is often not "what problem do we solve" but "what will happen if the others have one and we don't". I've been in that meeting.
42
00:07:39.016 --> 00:07:56.416
That is Rosa's down escalator. You stand on an escalator that moves downwards. If you stand still, you go down. So you run – not to arrive anywhere, only to stay where you are. And the more everyone runs, the faster the escalator goes.
43
00:07:57.016 --> 00:08:00.876
And when you hear "ninety-five percent no return"?
44
00:08:01.416 --> 00:08:09.116
Then I say: the investment did its job. Its job was not profit. Its job was to stay on the escalator.
45
00:08:09.796 --> 00:08:36.296
No, that's too cynical. There is another way to read the same story. In organisation research there's the Red Queen – from Alice in Wonderland, you have to run as fast as you can just to stay in the same place. But the researcher who used that image, William Barnett at Stanford, found something interesting: firms that survive that kind of competition actually get better. They learn. Running makes you a stronger runner.
46
00:08:36.316 --> 00:08:37.416
For the ones who survive.
47
00:08:38.036 --> 00:08:50.856
For the ones who learn. And in that same MIT report, the minority that did get results – they did very specific things. They let the people who do the work choose the tools. Not the board.
48
00:08:51.416 --> 00:08:57.956
Give me an example from your own work. A pilot that was escalator, and one that wasn't.
49
00:08:58.736 --> 00:09:28.696
Okay. Escalator: we once built a chatbot for our internal wiki because, uh, two competitors had announced one. It answered questions nobody asked, in a tone nobody liked. We switched it off after four months. Not escalator: our support team asked for a tool that sorts incoming tickets by language and urgency. They knew exactly what they wanted, they tested it themselves, and it saves them real, boring hours every week.
50
00:09:29.256 --> 00:09:30.736
And what did they do with the hours?
51
00:09:32.196 --> 00:09:34.936
They answered more tickets. You're not going to like that.
52
00:09:35.556 --> 00:09:37.096
I love that. I'll quote you.
53
00:09:37.716 --> 00:09:41.736
Krzysztof, you lived through a country that was running very fast.
54
00:09:42.756 --> 00:10:05.636
Poland in the nineties, yes. Everything at once. New currency, new companies, new words every month. And I will say something that may surprise Francesca – some of that running was necessary. We had a lot to catch up. But look at what stayed: Poles today work around eighteen hundred hours a year. Germans, about thirteen hundred and fifty. We caught up on many things. Not on time.
55
00:10:06.256 --> 00:10:08.516
That's – yeah. That's a strong number.
56
00:10:09.236 --> 00:10:20.396
One of the open questions in this essay is almost provocative. How much of today's AI investment would survive if you took the fear away? Francesca, do you dare a guess?
57
00:10:21.036 --> 00:10:39.976
A guess, clearly a guess. Maybe a third. And I can tell you how I'd test it. For every AI project, I'd ask one question: would we do this if no competitor were doing it? If the answer is yes, keep it. If the honest answer is "only because they do", you've found the escalator.
58
00:10:40.576 --> 00:10:42.776
And who answers honestly in a budget meeting?
59
00:10:44.316 --> 00:10:49.136
Nobody. Which is why you'd ask it in writing, anonymously, before the meeting.
60
00:10:49.796 --> 00:10:53.176
I like the question. I don't believe the third. I think it's less.
61
00:10:53.936 --> 00:10:58.916
So we have a test and two different guesses. That's more than most boards have.
62
00:11:03.556 --> 00:11:32.476
Third topic. The end of the working day. For most of industrial history it was a physical fact: the factory gate closed. Today, Microsoft – which, let's say it, sells the tools – published data from its own software: forty percent of users were going through their email at six in the morning, and many came back to it late in the evening. Francesca, is that you?
63
00:11:33.456 --> 00:11:45.516
Guilty. I commute by train, forty minutes, and I – I prompt on the train. A first draft before I'm in the office. And I like it. That's the uncomfortable part.
64
00:11:46.076 --> 00:11:59.416
That is exactly the uncomfortable part. Nobody forces you. That's how acceleration works today – it doesn't need a boss with a whip. You do it yourself, because starting a task now costs one sentence.
65
00:12:00.016 --> 00:12:21.552
But flexibility is real, Krzysztof. There was a big experiment in China – a travel company let half its staff work from home two days a week. People quit a third less, and their work didn't get worse. My parents had a shop in Milan, they couldn't leave it for one hour.
For many people, being able to work at odd times is freedom.
66
00:12:22.152 --> 00:12:28.692
Freedom of timing, yes. I'm not against that. I'm against a day without an end. That's a different thing.
67
00:12:29.452 --> 00:12:48.072
I have my own little experiment. I go hiking in the Alps on weekends, and in many valleys there's no reception. The first hour I keep reaching for the phone. By the afternoon I've forgotten it exists. And on Monday – nothing has collapsed.
68
00:12:48.772 --> 00:12:56.012
Because you're allowed to be unreachable there. The mountain gives you an excuse. In the city, nobody accepts the excuse.
69
00:12:56.612 --> 00:13:11.752
Or you build the excuse into the system. Volkswagen has done that for years – for many staff, emails stop arriving on their work phones half an hour after the shift ends and start again in the morning. The mountain, as company policy.
70
00:13:12.372 --> 00:13:16.652
Isn't the end of the day simply each person's own responsibility?
71
00:13:17.432 --> 00:13:35.892
No. That's the – sorry – that's the trick. If I stop answering at six and my colleague doesn't, I fall behind. Personal discipline is a weak answer to a structural problem. That's why France made a right to disconnect into law, and Australia did the same in twenty twenty-four.
72
00:13:37.372 --> 00:13:46.352
And that was my phone. My son. He studies computer science, he works at night, and he assumes everybody does. Sorry. Go on.
73
00:13:47.072 --> 00:14:14.012
Perfect timing. But here's what worries me more than the evening email: the cost of fast work lands on someone else. There's a study from last year – people were asked about AI-generated work from colleagues that looks finished but isn't. They called it "workslop". About forty percent said they'd received some in the last month, and they said it took them around two hours each time to sort it out.
74
00:14:14.652 --> 00:14:18.252
So the time saved by the sender is paid by the receiver.
75
00:14:18.851 --> 00:14:20.291
Yes. With interest.
76
00:14:20.852 --> 00:14:33.292
The essay ends with a question that fits you two perfectly. Soon AI agents will work through the night and present results in the morning. Will that make the human workday longer or shorter?
77
00:14:33.952 --> 00:14:45.492
My guess – longer. Because the agent finishes at three in the morning, and the result is there. Waiting. And a result that waits is a question that someone must answer.
78
00:14:46.032 --> 00:15:04.832
Unless you design it differently. I'd go the other way. Give the agents a right to disconnect too. They can work all night – but they deliver at nine. Not at three, not at six. One delivery, in working hours. Then the night work actually gives you a shorter day.
79
00:15:06.312 --> 00:15:10.172
A right to disconnect for machines. I would not have thought of that.
80
00:15:10.732 --> 00:15:12.232
That's because you don't build them.
81
00:15:12.892 --> 00:15:21.152
Fair. But I'll tell you what happens. Someone will discover that the nine o'clock delivery is ready at three, and they'll want to see it at three.
82
00:15:21.852 --> 00:15:30.092
So – a design idea, and a doubt that it survives the first ambitious manager. I think I can live with that.
83
00:15:34.672 --> 00:15:54.012
Fourth topic, and it's about experience. In twenty twenty-three, "prompt engineer" was the job of the future, with spectacular salaries. Two years later it had more or less disappeared – it became something everybody was expected to do. Krzysztof, what does Rosa make of that?
84
00:15:54.732 --> 00:16:14.452
Rosa says the present is shrinking. The period in which what you learned yesterday is still valid tomorrow – that period gets shorter. In early modern times, the big changes happened between generations. Your grandchildren lived in another world. Now it happens within one working life, sometimes several times.
85
00:16:15.052 --> 00:16:23.432
But I have to push back on one thing. Everyone repeats that skills have a "half-life of five years". Do you know where that number comes from?
86
00:16:24.032 --> 00:16:25.452
From a study, I assume.
87
00:16:26.051 --> 00:16:33.952
No! From a book and an opinion piece. Nobody measured it. It sounds like physics and it's – it's a slogan.
88
00:16:35.732 --> 00:16:39.032
All right, you got me. I've used that number in lectures.
89
00:16:39.672 --> 00:16:56.832
Everybody has. And the World Economic Forum says employers expect about forty percent of core skills to change by twenty-thirty. But look at which skills grow: analytical thinking, judgement. The tools change every six months. The judgement doesn't.
90
00:16:57.532 --> 00:17:02.772
So there's a surface that changes quickly, and something underneath that stays?
91
00:17:03.452 --> 00:17:16.372
Yes, and a German philosopher, Hermann Lübbe, said something nice about that: the people who cope best with a shrinking present are the ones who know languages and mathematics – because that kind of knowledge doesn't age.
92
00:17:16.972 --> 00:17:32.472
There's an open question here I want to ask you both, because it's personal for one of you. What does a shrinking present do to younger workers, who have no stable past to fall back on? Francesca, you're thirty-three.
93
00:17:33.752 --> 00:17:59.252
Yeah. I'll try an answer, and it's half data, half my own life. The data first – there's a Stanford team that follows payroll records in the US. In the jobs most exposed to AI, employment of young people, early twenties, is now almost a fifth below where it would otherwise be. The authors call it an early warning, not a final result. But it's the entry jobs that go first.
94
00:17:59.892 --> 00:18:01.632
And the half from your life?
95
00:18:02.352 --> 00:18:20.836
My guess is that my generation doesn't suffer from having no past. We never expected one. What we lose is something else – the boring first two years. The years where you do the simple stuff and learn the craft by
doing it badly. If the machine does the simple stuff, where do you learn?
96
00:18:21.495 --> 00:18:45.136
That – I agree with completely. And I'll add one thing from my generation. In nineteen-eighty-nine my teachers' experience became useless overnight. Everything they knew about the economy, about careers, gone. I know what it means when the past stops being a guide. What saved us were apprenticeships in the old sense – people who took you along.
97
00:18:45.716 --> 00:18:59.356
Can I tell you how I learned? My first job was at a bank in Milan, data cleaning. Eight months of fixing addresses in spreadsheets. It was terrible. And it's the reason I can smell a wrong number from across the room today.
98
00:18:59.976 --> 00:19:00.516
And today?
99
00:19:01.176 --> 00:19:09.756
Today a model does the address cleaning in a minute. And our juniors are faster than I ever was – but when the model is wrong, half of them don't notice.
100
00:19:10.356 --> 00:19:12.996
So your answer is: protect the apprenticeship.
101
00:19:13.556 --> 00:19:26.796
My guess is: protect the boring years. Deliberately. Let young people do things the slow way, even when the machine could do it faster. Not out of nostalgia – because that's where judgement grows.
102
00:19:27.416 --> 00:19:34.856
I'd add: let them do it with the machine, but make them explain every answer. Then the slow part is the explaining.
103
00:19:35.456 --> 00:19:38.076
That's the first time today you've built on each other.
104
00:19:38.696 --> 00:19:39.716
Don't get used to it.
105
00:19:44.456 --> 00:19:55.276
Last topic, and perhaps the most political. Time is never saved in general. It's saved by someone, and it goes to someone. Krzysztof?
106
00:19:55.916 --> 00:20:14.016
There's a large Danish study – twenty-five thousand workers. The researchers estimate that only three to seven percent of the productivity gains from chatbots reached the workers as higher pay. The rest went – somewhere else. To the firm, to customers. Not to the person who saved the time.
107
00:20:14.576 --> 00:20:17.416
That's early data. Adoption was two years old.
108
00:20:17.996 --> 00:20:20.996
It's always early data when the result is uncomfortable.
109
00:20:21.836 --> 00:20:41.556
Touché. But there's a second finding in that same research that I find more important. Women were much less likely to use the tools – about twenty percentage points, even in the same jobs. And the reason isn't skill. It's time. To learn a new tool you need quiet hours, and those hours are not equally distributed.
110
00:20:42.196 --> 00:20:58.876
I can speak to that, actually. When my children were small, I had no uninterrupted hour. Ever. Every new technology at the radio station, I learned it last – not because I was slower, because I was the one who had to leave at five.
111
00:20:59.516 --> 00:21:09.996
And the fix is boring and cheap. The same researchers say a day of training within working hours closes a lot of the gap. Within working hours – not as an evening course.
112
00:21:10.556 --> 00:21:23.336
And it's not only training. AI tools reward deep, uninterrupted sessions – you sit with the model, you go back and forth for an hour. Who has that hour? Not the person picking up the kids at four.
113
00:21:23.956 --> 00:21:37.116
That's the thing Rosa calls desynchronisation. The tool runs on one rhythm, a life with children or parents to care for runs on another. And whoever can't keep the tool's rhythm falls behind – even if they're very good.
114
00:21:37.536 --> 00:21:52.116
And then there's the time we don't see at all. AI is fast because somewhere people work slowly. Data labellers in Kenya were paid between one and two dollars an hour to read the worst content on the internet, so that the chatbot would be polite to us.
115
00:21:52.796 --> 00:22:11.956
That's true, and it's shameful. But I don't think the answer is "AI only takes". There's an example I like: in Britain, sixty-one companies tried a four-day week. Same pay. At the end, fifty-six of them kept it. Saved time can be given back – if someone decides to.
116
00:22:12.596 --> 00:22:23.576
Exactly. "If someone decides to." That is my whole point. The four-day week was not a gift from technology. It was a decision. And it was not made by the software.
117
00:22:24.156 --> 00:22:25.816
Then we agree that it's political.
118
00:22:26.476 --> 00:22:30.056
We agree that it's political. We don't agree that it will happen.
119
00:22:30.676 --> 00:22:38.236
I think it will. Slowly. Because people will compare – and they'll ask why their saved hours vanished and their neighbour's didn't.
120
00:22:38.755 --> 00:22:41.616
I hope you're right. I've seen many hopes like that.
121
00:22:42.276 --> 00:23:18.416
Let me try to hold this together without pretending you agree. You both accept that the time savings are real – smaller than the marketing, but real. You both accept that, left alone, they leak into more work. Krzysztof thinks that's how the system is built. Francesca thinks it's a design and management problem that can be solved. You found one idea together – protect the slow years of the young. And one idea that only Francesca believes in: give the machines a right to disconnect.
122
00:23:20.096 --> 00:23:21.416
I'll get you there, Krzysztof.
123
00:23:21.996 --> 00:23:24.496
Ask me again in a year. If I have time.
124
00:23:25.056 --> 00:23:42.176
What stays open is the question nobody in this conversation could answer for you. So here it is. If AI gave you one hour a week back tomorrow – who would decide what happens to it? You, your employer, or nobody?
125
00:23:43.176 --> 00:23:59.496
The essays behind this conversation, with all the sources, are on ideation.zone. Next time we look at what happens when society runs faster than nature and democracy can follow. Krzysztof, Francesca – thank you.
126
00:24:00.116 --> 00:24:00.936
Thank you, Elisabeth.
127
00:24:01.476 --> 00:24:02.836
Thanks – this was fun.