1 00:00:00,730 --> 00:00:02,940 My prediction is this. I'll go out on a limb. 2 00:00:03,090 --> 00:00:08,320 I'll say in five years I will not check my inbox anymore. 3 00:00:08,764 --> 00:00:13,924 Don't think of AI as an Einstein. Think of it as what would you do if you would 4 00:00:13,964 --> 00:00:15,614 have 100 assistants. 5 00:00:16,191 --> 00:00:17,921 Use the machine to test the hypothesis. 6 00:00:18,211 --> 00:00:20,741 Never allow the machine to give you the hypothesis. 7 00:00:26,700 --> 00:00:28,700 You are on mute! 8 00:00:28,800 --> 00:00:31,700 Where AI is taking us, and how to prepare. 9 00:00:32,700 --> 00:00:35,300 Episode 1: Think first. 10 00:00:36,480 --> 00:00:41,090 Hello everyone, and welcome to the first episode of You're On Mute. 11 00:00:41,800 --> 00:00:47,640 We're here together with Nicklas a former colleague from Google, and we have 12 00:00:47,680 --> 00:00:53,790 been discussing a lot lately where the world is heading and what we are seeing. 13 00:00:53,960 --> 00:00:57,560 And we get these questions all the time as well, is that so how should we be 14 00:00:57,600 --> 00:01:00,640 preparing? What- how should board work look like? 15 00:01:00,680 --> 00:01:05,290 How should education look like? How do you use these tools? 16 00:01:05,660 --> 00:01:10,070 And this is exactly the reason why we are here right now to discuss with you 17 00:01:10,220 --> 00:01:13,210 what we can do. Great to see you, Nicklas. 18 00:01:13,600 --> 00:01:14,060 And great to be here- 19 00:01:14,100 --> 00:01:14,760 How was your morning? 20 00:01:14,800 --> 00:01:18,480 Petri. My morning was really good. I'm on an island in the archipelago, and 21 00:01:18,520 --> 00:01:23,730 it's drizzling, but it's a beautiful autumn morning, so I really enjoyed it. 22 00:01:25,180 --> 00:01:27,310 What did AI do to you this morning? 23 00:01:27,880 --> 00:01:30,210 do for me, I hope. So- 24 00:01:30,880 --> 00:01:33,470 what it did, what it did do to me, I don't know. 25 00:01:33,620 --> 00:01:36,380 But no, so what it did for me this morning is several things, right? 26 00:01:36,420 --> 00:01:42,040 One is that I nowadays read the morning news through an, a news 27 00:01:42,080 --> 00:01:45,560 aggregator that I built myself. So the first thing I do in the morning, I go to 28 00:01:45,600 --> 00:01:49,380 this news aggregator. I check what the news is, what the top news is in the 29 00:01:49,420 --> 00:01:52,490 categories I care about, AI, politics, et cetera. 30 00:01:52,920 --> 00:01:57,170 so that's primarily how I encountered it this morning. 31 00:01:57,260 --> 00:02:02,780 And then I, and I also had it remind me what we had agreed we would discuss on 32 00:02:02,820 --> 00:02:07,420 this podcast. So there you go. which it which it happily did because it has 33 00:02:07,460 --> 00:02:10,560 access to my calendar and email and meeting notes and so on. 34 00:02:10,600 --> 00:02:13,920 so I s- I would say that's my morning AI. 35 00:02:13,960 --> 00:02:14,860 What about yours? 36 00:02:15,860 --> 00:02:22,160 Well, my morning AI routine has become to check out what it 37 00:02:22,200 --> 00:02:27,200 looked into my inbox and then decides on where, what to do a reply to, wh- what 38 00:02:27,240 --> 00:02:30,310 not to reply to, and also write drafts. 39 00:02:30,660 --> 00:02:34,070 I have not let it answer directly itself. 40 00:02:34,360 --> 00:02:39,270 So I use it as an assistant, as it should be, but it writes for me 41 00:02:40,060 --> 00:02:44,320 replies, and then I look through them and may might do some small changes, and 42 00:02:44,360 --> 00:02:49,690 then I'm ready to go. I also used as a news aggregator. 43 00:02:50,380 --> 00:02:54,380 work- been working in the, in the defense industry, so I have no clue of 44 00:02:54,420 --> 00:02:58,920 defense, so it's been f- amazing for us, for me to be able to have my own news 45 00:02:58,960 --> 00:03:02,070 aggregator to look at what's going on in that field. 46 00:03:02,560 --> 00:03:05,100 so tho- those are those things that really... 47 00:03:05,140 --> 00:03:11,180 Plus, I have to say that I used AI to optimize this beautiful setup that 48 00:03:11,220 --> 00:03:12,130 I have here with- 49 00:03:12,200 --> 00:03:12,480 It is- 50 00:03:12,680 --> 00:03:14,640 it is quite stunning. It's quite stunning. 51 00:03:14,680 --> 00:03:15,580 Very well done. 52 00:03:16,160 --> 00:03:20,600 I did, I did a few alterations, and it said, Oh, you know, you should use more 53 00:03:20,640 --> 00:03:24,140 light here and more light there, and don't let light come in through the window, 54 00:03:24,180 --> 00:03:28,600 and so on. again, it shows what all kind of stuff you can, you can do with this. 55 00:03:28,640 --> 00:03:31,620 I should probably learn from that. I feel worried about my background now. 56 00:03:31,660 --> 00:03:35,480 but I want to catch something you said that I think is really interesting. 57 00:03:35,520 --> 00:03:39,000 it's the email thing because I've started to do the same. 58 00:03:39,040 --> 00:03:44,710 And I, I've started to sort of ask myself a question, which is I wonder how much 59 00:03:44,840 --> 00:03:50,400 longer we will actually have email inboxes for humans, and when email 60 00:03:50,440 --> 00:03:55,040 inboxes should be for agents instead. Because the idea of looking into your 61 00:03:55,080 --> 00:03:58,880 email inbox with all of the noise and all of the different things that come in, 62 00:03:58,920 --> 00:04:03,240 some important, some not, and for me to take on that filtering cost and that 63 00:04:03,280 --> 00:04:05,670 filtering burden makes absolutely no sense. 64 00:04:05,960 --> 00:04:06,220 And I- 65 00:04:06,460 --> 00:04:11,640 and I really do believe, I mean, I hear you about only having a draft because 66 00:04:11,680 --> 00:04:15,840 you don't want it to send yet, but, I mean, for how much longer does it have to 67 00:04:15,880 --> 00:04:20,520 draft really well for you to say, Well, you know, I actually am quite fine with 68 00:04:20,560 --> 00:04:24,100 it responding for me as well? My prediction is this. 69 00:04:24,140 --> 00:04:25,190 I'll go out on a limb. 70 00:04:25,340 --> 00:04:30,570 I'll say in five years I will not check my inbox anymore. 71 00:04:30,680 --> 00:04:35,300 I won't open Gmail anymore. I will be doing all of my correspondence, whether 72 00:04:35,340 --> 00:04:39,830 it's, you know, email or any other correspondence that I do, through an agent. 73 00:04:39,880 --> 00:04:44,300 And I'm so looking forward to it. I think we might be closer to the end of email 74 00:04:44,340 --> 00:04:47,150 than we realize, and what a glorious thing that would be. 75 00:04:47,860 --> 00:04:53,219 that's a good prediction. And I have to say that also now I still go into my 76 00:04:53,259 --> 00:04:55,690 inbox then to glance through that was it correct. 77 00:04:56,540 --> 00:05:01,720 And until now the only mistakes that I have found was that it had obviously acc- 78 00:05:01,760 --> 00:05:06,040 access to my calendar too, and if I, if I put into calendar that there's an 79 00:05:06,080 --> 00:05:11,430 event that takes a full day, it thinks that day is free. 80 00:05:12,360 --> 00:05:16,090 so if someone comes in and said, Oh, could we have lunch during that day? 81 00:05:16,580 --> 00:05:19,760 it will tell, Oh, yeah, you have-- you don't have anything in, on that day, 82 00:05:19,800 --> 00:05:24,940 because I had put it into the event. So you need what I learned is that even if 83 00:05:24,980 --> 00:05:30,940 it's a day, a full day event, you need to block the full day in order for 84 00:05:30,980 --> 00:05:34,540 it to understand that yes you are actually busy. 85 00:05:34,580 --> 00:05:39,540 But I, I'm completely with you. I actually heard an interesting statistic a few 86 00:05:39,580 --> 00:05:44,560 days ago that said that already today two-thirds of internet traffic would be 87 00:05:44,600 --> 00:05:45,590 done by bots. 88 00:05:46,160 --> 00:05:46,500 Yeah. 89 00:05:46,540 --> 00:05:47,920 That makes total sense to me. And- 90 00:05:47,960 --> 00:05:52,310 So if you have a website today, who are you optimizing for? 91 00:05:52,580 --> 00:05:55,830 Are you optimizing for humans, or are you optimizing for bots? 92 00:05:56,260 --> 00:05:56,500 Oh, I would- 93 00:05:56,540 --> 00:05:56,980 And then if you go to Google's SEO- 94 00:05:57,020 --> 00:05:59,040 I would, I would certainly optimize for bots. 95 00:05:59,080 --> 00:05:59,340 Yes. 96 00:05:59,380 --> 00:06:03,580 Yeah. But if you go to Google's SEO rules they say, oh, you should own- or, you 97 00:06:03,620 --> 00:06:05,530 know, you should really optimize it for the humans. 98 00:06:05,760 --> 00:06:10,804 But if it's only a third of the traffic- I think that's a fascinating 99 00:06:10,844 --> 00:06:11,744 development. 100 00:06:11,904 --> 00:06:14,884 It is. And I, and I think there's more to that that's really interesting to dig 101 00:06:14,924 --> 00:06:19,184 into as well because as the web transitions into an agent web rather than a 102 00:06:19,224 --> 00:06:23,184 human web there is so much else that should happen as well with services. 103 00:06:23,224 --> 00:06:27,144 So if you're a service today and you don't have an API that agents can interact 104 00:06:27,184 --> 00:06:32,464 with or like a MCP server or something like that, I think that's a conscious 105 00:06:32,504 --> 00:06:36,164 choice because you're afraid of being out-competed by the agents or you're 106 00:06:36,204 --> 00:06:40,354 afraid of becoming just a resource under the agent layer. 107 00:06:40,783 --> 00:06:46,704 And I don't think that works because I think we, I know I want these services to 108 00:06:46,744 --> 00:06:47,644 just connect. 109 00:06:47,944 --> 00:06:51,344 The reason I, for example, have started to use Notion a lot more is that Notion 110 00:06:51,384 --> 00:06:54,514 has a really good set of agent integrations. 111 00:06:54,964 --> 00:07:00,974 So you know, I have- I do meeting transcripts, I do different kinds of updates, 112 00:07:01,224 --> 00:07:05,204 and Notion essentially works as an extended memory for all of the projects that 113 00:07:05,244 --> 00:07:08,334 I'm involved with. And my agents then just consult Notion. 114 00:07:08,704 --> 00:07:12,844 They update it. There's a work journal for every single project, so I can go in, 115 00:07:12,884 --> 00:07:18,064 see what has been done where. and that kind of structure really depends on the 116 00:07:18,104 --> 00:07:22,994 idea that we have services that are agent first. 117 00:07:23,544 --> 00:07:27,224 And I think if you're not agent first because you're a little bit afraid that 118 00:07:27,264 --> 00:07:30,444 the agents will come between you and the customer, you probably need to find 119 00:07:30,484 --> 00:07:33,094 another way to make sure the customers love what you're doing. 120 00:07:34,084 --> 00:07:34,984 I love it. 121 00:07:35,764 --> 00:07:38,194 Th- this really reminds me all the time is that 122 00:07:39,504 --> 00:07:43,604 well, I can tell our viewers now that also that you worked for DeepMind, and one 123 00:07:43,644 --> 00:07:48,944 of the DeepMind engineers once told me that, Petri, don't think of AI as an 124 00:07:48,984 --> 00:07:53,734 Einstein. Think of it as what would you do if you would have 100 assistants. 125 00:07:54,364 --> 00:07:58,504 And that has really stuck with me in the sense that, yes, if I would have those 126 00:07:58,544 --> 00:08:02,734 assistants, will I let them run immediately free? 127 00:08:02,824 --> 00:08:06,804 No. I would tell them exact- you need to tell them, You do this, and that, and 128 00:08:06,844 --> 00:08:09,394 when you've done this, you come and tell, show me what you've done. 129 00:08:09,804 --> 00:08:12,674 And then you correct and then say, Oh, can you do this, and that? 130 00:08:12,904 --> 00:08:14,254 And I think that, 131 00:08:15,644 --> 00:08:20,964 let's call it a metaphor I think is, has been s- stuck with me, and that's the 132 00:08:21,004 --> 00:08:24,644 way I've been looking into most of the stuff I'm doing is that. 133 00:08:24,684 --> 00:08:30,654 So what could my 100 assistants do for me while I'm 134 00:08:31,224 --> 00:08:32,124 you know, 135 00:08:32,704 --> 00:08:34,114 r- raking leaves in the garden? 136 00:08:34,684 --> 00:08:39,084 Yes. and reading and doing all those things that sort of make us smarter. 137 00:08:39,124 --> 00:08:42,464 But... And I think it's interesting because I agree with that although I think 138 00:08:42,504 --> 00:08:46,184 there's a, there's a twist to it, which is that if you had 100 assistants, you 139 00:08:46,224 --> 00:08:48,024 would appoint one of them head assistant. 140 00:08:48,064 --> 00:08:51,184 And then you would say to that person, Please take care of the rest so you can 141 00:08:51,224 --> 00:08:53,904 organize them. And that's essentially orchestration, right? 142 00:08:53,944 --> 00:08:55,034 Agent orchestration. 143 00:08:55,364 --> 00:08:55,984 Exactly. 144 00:08:56,024 --> 00:09:00,264 and I think we will see these orchestration patterns emerge that are super 145 00:09:00,304 --> 00:09:03,224 interesting, and people will adopt different orchestration patterns. 146 00:09:03,264 --> 00:09:06,984 You can have a main agent that you interact with and then delegates to 147 00:09:07,024 --> 00:09:10,724 sub-agents. You can have a main agent you interact with sub-agents, and then you 148 00:09:10,764 --> 00:09:15,004 can have a guy who checks it. Currently, you're the guy who checks the email 149 00:09:15,044 --> 00:09:16,574 inbox to see nothing was missed. 150 00:09:16,944 --> 00:09:18,274 You could have an agent do that. 151 00:09:18,564 --> 00:09:21,884 Or you're the one who reads the draft. But if you have one person drafting, one 152 00:09:21,924 --> 00:09:23,804 reading, and a third saying, Yeah, you're right. 153 00:09:23,844 --> 00:09:27,454 We should change this, then you start to sort of build these little teams. 154 00:09:27,504 --> 00:09:30,844 And I know Ethan Mollick, who writes about this beautifully and is a really 155 00:09:30,884 --> 00:09:35,464 smart professor of management, he said that the key skill in the future for AI 156 00:09:35,504 --> 00:09:38,654 is not going to be coding or computer science, it's going to be management. 157 00:09:39,124 --> 00:09:40,754 You have to be really good at management. 158 00:09:40,984 --> 00:09:42,784 How do you delegate? What do you not delegate? 159 00:09:42,844 --> 00:09:46,434 How do you set up teams? How do you make sure teams have clear goals? 160 00:09:46,704 --> 00:09:49,344 The kinds of things that you've been a manager for a long time, or the kinds of 161 00:09:49,384 --> 00:09:54,504 things that we, that we learned as managers are now actually the core skills for 162 00:09:54,544 --> 00:09:57,094 interacting with AI, which is kind of fascinating. 163 00:09:57,564 --> 00:10:02,934 It is. And if you take that argument or that thinking even further, 164 00:10:03,564 --> 00:10:08,394 traditionally in companies, even your salary bands have been more or less 165 00:10:08,504 --> 00:10:13,324 defined by how big is your team. And then the smartest one has realized that, 166 00:10:13,364 --> 00:10:17,994 oh, you know, I should, I should get the bigger team, and I get paid better and 167 00:10:18,504 --> 00:10:23,164 and so on. I might higher in, so to say, in the, in the hierarchy of these 168 00:10:23,204 --> 00:10:28,454 companies. So what if... And we will move into these hybrid teams. 169 00:10:28,883 --> 00:10:34,464 I will have those 100 assistants on my hybrid team, and I might have two other 170 00:10:34,504 --> 00:10:36,774 people there doing things. 171 00:10:37,944 --> 00:10:40,144 is that a big team or a small team going forward? 172 00:10:40,184 --> 00:10:40,444 Exactly. 173 00:10:40,484 --> 00:10:45,284 And the question is that maybe we should be really looking into that how can we 174 00:10:45,324 --> 00:10:48,234 build those hybrid teams in the future. 175 00:10:48,384 --> 00:10:49,084 And measure on outcomes. 176 00:10:49,124 --> 00:10:50,764 By the way, leadership will change also completely. 177 00:10:50,804 --> 00:10:53,884 Oh, it will. I- it certainly will. I mean, and measure on outcomes, I think, is 178 00:10:53,924 --> 00:10:57,804 the thing you have to do, and not look at, for example, you know, how many 179 00:10:57,844 --> 00:11:01,294 people do you have in your team, which I agree traditionally has been the case. 180 00:11:01,484 --> 00:11:06,344 Because now you know, a single person with a really good agent setup with 181 00:11:06,384 --> 00:11:11,314 excellent data can outproduce a 200, 300, 400 person team. 182 00:11:11,584 --> 00:11:14,664 And that's also why we're seeing the restructuring in many of the tech companies 183 00:11:14,704 --> 00:11:15,624 now where, you know- 184 00:11:15,744 --> 00:11:18,944 the vast restructuring that Microsoft is doing, et cetera. 185 00:11:18,984 --> 00:11:22,944 so coming back to this notion of what we do, I think one of the core things that 186 00:11:22,984 --> 00:11:27,594 we need to learn is that we need to learn how to think about this in a more 187 00:11:28,644 --> 00:11:29,614 traditional way. 188 00:11:32,664 --> 00:11:33,564 Meaning? 189 00:11:34,444 --> 00:11:35,894 Thinking a traditional way? What do you mean? 190 00:11:36,464 --> 00:11:40,164 So what I... Yeah, no, what I mean, we need to think about this in a way where 191 00:11:40,204 --> 00:11:42,214 we sort of go back to the basics- 192 00:11:42,484 --> 00:11:42,804 Mm-hmm 193 00:11:42,844 --> 00:11:47,704 and ask what do we do? What are the kinds of things that we produce? 194 00:11:47,744 --> 00:11:50,454 What are the kinds of services that we offer, for example? 195 00:11:50,784 --> 00:11:54,954 So that's largely what I was referring to. 196 00:11:55,364 --> 00:11:56,814 I have a question for you, Nicklas. 197 00:11:58,504 --> 00:12:04,504 Is AI a transformational technology, like for example digital 198 00:12:04,544 --> 00:12:06,774 was, or is it just a way to 199 00:12:08,664 --> 00:12:11,234 make us as individuals much more efficient? 200 00:12:13,512 --> 00:12:17,372 it's a good question because when we say AI, we typically take all of the 201 00:12:17,412 --> 00:12:21,152 different technologies that fall under that particular term, and we treat them 202 00:12:21,192 --> 00:12:25,712 the same. And I think there is some AI that is really good at sorting, for 203 00:12:25,752 --> 00:12:28,952 example. so you have spam, for example. 204 00:12:28,992 --> 00:12:31,852 Spam filters were a kind of AI, and they were really good at what they were 205 00:12:31,892 --> 00:12:35,912 doing. But if you think about AI as extended cognition, as us being able to 206 00:12:35,952 --> 00:12:40,512 think more differently, deeper, and m- you know, across more domains, then 207 00:12:40,552 --> 00:12:42,802 certainly it is transformational in a sense. 208 00:12:43,492 --> 00:12:49,452 for 'Cause to me, the technology that we are now building is adding cognition to 209 00:12:49,492 --> 00:12:52,792 the world, adding thinking to the world, and I think- that's something that's 210 00:12:52,832 --> 00:12:56,152 radically different from adding a new way of connecting or communicating or 211 00:12:56,192 --> 00:12:59,912 storing information. So you could, you could almost phrase this in saying that 212 00:12:59,952 --> 00:13:03,412 we're sort of, we have understood ourselves for the metaphor of the information 213 00:13:03,452 --> 00:13:08,292 society for a long time. Now, if you transition that into the new metaphor, the 214 00:13:08,332 --> 00:13:12,512 intelligence society, you suddenly realize that there's a huge difference 215 00:13:12,552 --> 00:13:15,912 between information and Intelligence is action. 216 00:13:15,952 --> 00:13:18,082 Intelligence is in different ways 217 00:13:19,152 --> 00:13:24,812 sort of reframing the world. And information is capturing the world and 218 00:13:24,852 --> 00:13:27,592 trying to describe or represent it in different ways. 219 00:13:27,632 --> 00:13:31,272 so I would, I would probably fall on the side of transformational when you look 220 00:13:31,312 --> 00:13:35,512 at the entire technological architecture we're building, because we're building 221 00:13:35,552 --> 00:13:39,362 something in the world that can think with us, and sometimes without us. 222 00:13:39,872 --> 00:13:40,412 What do you think? 223 00:13:40,452 --> 00:13:43,292 the reason why I'm asking this a little bit also is that obviously, you know, 224 00:13:43,332 --> 00:13:48,712 it's like we both remember the times when, so to say, in the, in the beginning 225 00:13:48,752 --> 00:13:53,502 of digital . and then s- suddenly we started seeing 226 00:13:54,312 --> 00:14:00,212 e-commerce, we started seeing digital banks that had no branches any longer, 227 00:14:00,252 --> 00:14:06,102 and so on and so forth. so it more or less built complete new ways of 228 00:14:07,372 --> 00:14:12,632 new ty- type of enterprises, new type of companies, new type new ways of serving 229 00:14:12,672 --> 00:14:17,942 your clients. and I don't know if we've seen an AI bank yet. 230 00:14:18,332 --> 00:14:24,432 We might have seen some insurance companies who can really automate the 231 00:14:24,472 --> 00:14:29,972 claims process and so on. but so the, so it really goes back to yes, I, I'm 232 00:14:30,012 --> 00:14:33,552 completely with you that it will, it's, and it's already transforming the way I 233 00:14:33,592 --> 00:14:37,922 work, and it transforming the way individuals are working. 234 00:14:38,712 --> 00:14:39,612 But is it 235 00:14:40,672 --> 00:14:46,542 is it transforming companies and the way companies work, and, 236 00:14:46,612 --> 00:14:49,662 or is it just so that we are so early in the development that we haven't- 237 00:14:50,092 --> 00:14:50,372 seen it yet? 238 00:14:50,412 --> 00:14:52,152 Yeah, no, I think, I mean, if you look at where it is- 239 00:14:52,192 --> 00:14:53,112 and the reason also I'm asking this a little bit- 240 00:14:53,232 --> 00:14:57,712 is that I see that those companies that in the early days of digital did not 241 00:14:57,752 --> 00:14:58,922 jump on the bandwagon, 242 00:15:00,112 --> 00:15:05,632 they got left behind. we look at some of those department stores who said, Oh, 243 00:15:05,672 --> 00:15:09,292 we will be waiting until, you know, I don't know what they were waiting for. 244 00:15:09,332 --> 00:15:13,272 They're still today waiting for it. and the Amazons and, of this world have 245 00:15:13,312 --> 00:15:18,732 passed them. and I see here the real, the same risk is that if this is truly 246 00:15:18,772 --> 00:15:23,602 transformational, if you don't jump on the bandwagon today, you'll be 247 00:15:24,832 --> 00:15:29,202 g- you know, you will be, not be able to catch the train any longer. 248 00:15:29,852 --> 00:15:33,552 and I actually think it's, I do think that it's the case. 249 00:15:33,592 --> 00:15:37,512 I think you're right. And I think the problem here is also that it's in a 250 00:15:37,552 --> 00:15:41,052 different way you will be left behind. Because in, if you look at the 251 00:15:41,092 --> 00:15:44,812 information society, if you look at the digital divide as we used to talk about, 252 00:15:44,852 --> 00:15:48,272 the digital divide you could bridge by connecting more and more people, or 253 00:15:48,312 --> 00:15:51,892 connecting companies, or connecting employees, and making sure they had access 254 00:15:51,932 --> 00:15:54,332 to the internet, and then starting to rethink the business model. 255 00:15:54,372 --> 00:15:58,252 So connection was the thing that you would do in order to bridge the digital 256 00:15:58,292 --> 00:16:01,512 divide. But if you now look at the kind of divide that we're building, it's a 257 00:16:01,552 --> 00:16:06,382 cognitive divide. It's sort of a divide that is based on how you learn. 258 00:16:06,872 --> 00:16:10,552 And learning is premised on organizing your data in the right way, asking the 259 00:16:10,592 --> 00:16:14,412 right kinds of questions, following up, compounding, making sure that every new 260 00:16:14,452 --> 00:16:17,202 learning is integrated into a new world model. 261 00:16:17,472 --> 00:16:20,382 And those things, they compound much faster. 262 00:16:20,532 --> 00:16:23,182 It's not just a question of giving somebody access to AI. 263 00:16:23,372 --> 00:16:27,012 You have to set up this learning infrastructure, and once you do that, it 264 00:16:27,052 --> 00:16:32,042 accelerates. So if you're left behind, you're left behind not a linear 265 00:16:32,511 --> 00:16:37,992 evolution of a field, but you're left behind somebody who is slowly going into 266 00:16:38,032 --> 00:16:43,412 an exponential learning curve. And that is, that is means that the cognitive 267 00:16:43,452 --> 00:16:47,292 divide is unlikely to just become smaller by its own. 268 00:16:47,332 --> 00:16:51,862 It's l- rather likely to grow. So between the companies that do or do not 269 00:16:52,052 --> 00:16:56,732 organize themselves for learning through AI or adapting through AI will probably 270 00:16:56,772 --> 00:17:02,632 be left even further behind, and their sort of gap will be growing if you 271 00:17:02,672 --> 00:17:06,402 compare to digital. So I think they're potentially in a worse situation. 272 00:17:07,652 --> 00:17:12,382 It's interesting you say about the exponential development. 273 00:17:13,172 --> 00:17:18,902 you also remember our ex-colleague Ray Kurzweil, who always argued m- 274 00:17:19,112 --> 00:17:23,582 his argument was clearly that yes, all these developments are exponential. 275 00:17:23,712 --> 00:17:26,892 And the interesting thing about exponential developments are obviously that in 276 00:17:26,932 --> 00:17:29,542 the beginning it looks like nothing happens. 277 00:17:29,952 --> 00:17:30,312 Yes. 278 00:17:30,352 --> 00:17:34,622 Because the beginning is, it looks slow, and then when the hockey stick hits, 279 00:17:35,032 --> 00:17:37,102 then that's when it's, so to say, late. 280 00:17:37,432 --> 00:17:42,932 Are we today in that phase now when... have we hit the hockey stick or yet, or 281 00:17:42,972 --> 00:17:45,742 are we still in the phase that it development looks slow? 282 00:17:46,100 --> 00:17:50,440 And there's a new term being thrown around in tech circles where they say we're 283 00:17:50,480 --> 00:17:54,580 in the foothills of the singularity. So the foothills is sort of this space 284 00:17:54,620 --> 00:17:58,790 between the ground and the mountain, the exponential. 285 00:17:59,100 --> 00:18:01,840 So this notion of being in the foothills of the singularity, I think is not 286 00:18:01,880 --> 00:18:04,580 completely wrong. I mean, I think singularity is a really... 287 00:18:04,620 --> 00:18:08,140 It's an interesting concept, and it can be contested because you can say there's 288 00:18:08,180 --> 00:18:11,880 been no technology in the history of mankind that has not turned out to be an S 289 00:18:11,920 --> 00:18:15,980 curve. So whatever we said was an exponential curve, electrification or 290 00:18:16,020 --> 00:18:21,520 computers or internet, et cetera, ultimately always turned out to flatten like 291 00:18:21,560 --> 00:18:24,530 this like an S curve, not an exponential curve. 292 00:18:24,860 --> 00:18:27,440 Now, the argument against that, though, from the people who believe in the 293 00:18:27,480 --> 00:18:30,700 singularity is really interesting because what they will say is that, yes, all 294 00:18:30,740 --> 00:18:35,240 individual technologies describe an S curve, but AI allows you to stack them on 295 00:18:35,280 --> 00:18:39,400 each other. So every time a technology flattens, a new technology will come 296 00:18:39,440 --> 00:18:42,280 along, and that will build the exponential over time. 297 00:18:42,320 --> 00:18:46,150 So compounding these S curves is what gives you the exponential. 298 00:18:46,520 --> 00:18:47,420 And so 299 00:18:48,660 --> 00:18:52,480 I mean, the jury's out on this one. You can or cannot believe in the singularity 300 00:18:52,520 --> 00:18:55,820 if you want to. But I think- one of the things that we should recognize is that 301 00:18:55,860 --> 00:18:59,630 the evolution, the development is now incredibly fast. 302 00:18:59,940 --> 00:19:00,200 And- 303 00:19:00,240 --> 00:19:00,600 Yeah 304 00:19:00,640 --> 00:19:04,900 we also seem to be at the cusp of this interesting moment where AI codes AI in a 305 00:19:04,940 --> 00:19:08,730 way that we don't quite understand, but we also believe will greatly 306 00:19:09,840 --> 00:19:12,810 will greatly improve the speed with which AI develops. 307 00:19:13,620 --> 00:19:17,770 Anthropic recently released a statement where they said that, you know, around 308 00:19:17,820 --> 00:19:21,220 90% of the improvements that we're doing right now in some of our models are 309 00:19:21,260 --> 00:19:24,900 actually done by models. checked by humans, they sort of added. 310 00:19:24,940 --> 00:19:26,250 It's like they're checked by humans. 311 00:19:27,400 --> 00:19:28,700 but that- 312 00:19:28,740 --> 00:19:30,410 It was an assistant. It wasn't me. 313 00:19:30,520 --> 00:19:35,860 Yes. but that sort of suggests, that suggests that we might be heading into the 314 00:19:35,900 --> 00:19:40,240 exponential to your point because I mean, I, and I do think this is interesting 315 00:19:40,280 --> 00:19:45,650 because y- when you can start to run different processes in computational speed, 316 00:19:46,080 --> 00:19:49,140 so you can run them at the speed of a computer, and you're no longer held back 317 00:19:49,180 --> 00:19:54,130 by biological speed, the speed of human beings, then things will go faster 318 00:19:54,300 --> 00:19:58,660 almost by default. there will be still some biological bottlenecks, but we're 319 00:19:58,700 --> 00:20:02,670 removing more and more of them across at least the field of coding. 320 00:20:06,280 --> 00:20:09,650 if I, if I... A little bit a- another, other question is like I think one of the 321 00:20:10,240 --> 00:20:14,880 advantages both you and I have right now is that we control, so to say, our own 322 00:20:14,920 --> 00:20:22,360 calendars, and we have the possibility to test, learn, take time to 323 00:20:22,400 --> 00:20:24,630 develop different things and so on. 324 00:20:25,940 --> 00:20:31,280 if I see the companies that I work together with on boards and also through 325 00:20:31,320 --> 00:20:36,580 different other means, they are completely overwhelmed and overheated by 326 00:20:36,620 --> 00:20:41,040 information, by tasks. Then they're saying, Oh, you know, we're cutting now 327 00:20:41,080 --> 00:20:45,540 people, but they don't know how to, how to you know, take advantage of these 328 00:20:45,580 --> 00:20:46,870 tools and so on. 329 00:20:48,060 --> 00:20:52,950 where do you think they should start? What would be-- What should they use their 330 00:20:53,480 --> 00:20:59,530 one free hour a day of in order to get over this 331 00:21:01,200 --> 00:21:07,170 how would I put it completely overwhelmed world as it seems like now in many 332 00:21:07,500 --> 00:21:08,450 cor- corporations? 333 00:21:09,260 --> 00:21:14,360 Well and it's such a good picture of reality as well because I do think that a 334 00:21:14,400 --> 00:21:17,180 lot of people feel overwhelmed. they feel they're working more than ever, and 335 00:21:17,220 --> 00:21:21,170 this notion that AI is taking their jobs, they're like, Yes, please, take it. 336 00:21:21,620 --> 00:21:23,950 Because it's like there's so much to do, right? 337 00:21:24,260 --> 00:21:25,900 but I but I think that if you, if you- 338 00:21:25,940 --> 00:21:27,790 by the way, that you s- you said that 339 00:21:28,880 --> 00:21:30,780 and partly as a little bit of a joke, but it interests me. 340 00:21:30,820 --> 00:21:31,040 yeah. 341 00:21:31,080 --> 00:21:36,880 I just saw in a survey where half of people in 342 00:21:36,920 --> 00:21:42,320 businesses did not dare to say what tools they are using because they're afraid 343 00:21:42,360 --> 00:21:46,600 of their jobs. So if someone sees that they're using a tool and that makes them 344 00:21:46,640 --> 00:21:49,310 more efficient, the question is, Oh, am I still needed? 345 00:21:50,140 --> 00:21:51,040 Half of the people- 346 00:21:51,140 --> 00:21:53,070 which I thought was scary. 347 00:21:53,480 --> 00:21:54,020 and it's crazy. 348 00:21:54,060 --> 00:21:57,300 So they'd rather, they'd rather keep it to themselves and say, Oh, yeah, I'm 349 00:21:57,340 --> 00:21:58,240 really busy. 350 00:21:58,320 --> 00:22:01,840 and that's, and that's crazy because what we really need are people who can use 351 00:22:01,880 --> 00:22:05,340 the tools and use them really well. But to your question then, what do you do if 352 00:22:05,380 --> 00:22:08,120 you have an hour a day and you want to develop and you want to grow and you want 353 00:22:08,160 --> 00:22:11,500 to understand how to deal with the information that is overwhelming you? 354 00:22:11,540 --> 00:22:14,990 I think the answer is surprisingly simple and slightly cheeky. 355 00:22:15,300 --> 00:22:20,630 It is you should ask the AI to teach you how to deal with information overload. 356 00:22:20,960 --> 00:22:24,940 Give it, you know, ask it to give you three tips a day and walk you through how 357 00:22:24,980 --> 00:22:28,960 to become more effective, and it will. At this point, we're in this unique 358 00:22:29,000 --> 00:22:32,210 situation where we have a technology that can teach you how to use it. 359 00:22:32,700 --> 00:22:36,700 And that's never been the case before. And so if you're really dead set on 360 00:22:36,740 --> 00:22:40,430 trying to figure out how things work, you can just start to ask it. 361 00:22:40,740 --> 00:22:43,800 That's when I realized that, which was when I was sort of playing around with 362 00:22:43,840 --> 00:22:49,100 chatbots, et cetera, that I can ask this thing to teach me it m- sort of made 363 00:22:49,140 --> 00:22:51,820 for a minor revolution in the way that I use technology. 364 00:22:51,860 --> 00:22:55,820 So I started coding. It taught me to use open source repos. 365 00:22:55,860 --> 00:22:59,760 I started to build my own skills, was the thing I started with, which is a 366 00:22:59,800 --> 00:23:04,220 simple way of telling the g- the AI to do the thing it's doing in the same way 367 00:23:04,260 --> 00:23:05,060 all the time. 368 00:23:05,120 --> 00:23:08,940 And then I went into using Claude code, and every time I ran into something that 369 00:23:08,980 --> 00:23:12,620 I didn't understand, I asked it to explain it to me as if I was a 12-year-old, 370 00:23:12,660 --> 00:23:14,550 and it happily did. And so- 371 00:23:14,760 --> 00:23:19,420 I would, I would I would, I would sort of constantly have the ability to learn 372 00:23:19,460 --> 00:23:25,220 more and more and more. my challenge now is that I need the AI to 373 00:23:25,260 --> 00:23:29,060 tell me where my ignorance is keeping me back because I think I learned 374 00:23:29,100 --> 00:23:33,190 something. I think I know how to do this thing, and so I do it the same way. 375 00:23:33,480 --> 00:23:37,440 And what I've learned to do is that I've learned to ask the AI, improve on my 376 00:23:37,480 --> 00:23:42,239 ideas and improve on my methods. Tell me what I could be doing to do this more 377 00:23:42,279 --> 00:23:43,179 effectively. 378 00:23:43,380 --> 00:23:46,800 Which means that I don't get stuck in thinking that I'm really good at this, 379 00:23:46,840 --> 00:23:50,240 which is like a very human error. I did that in the beginning. 380 00:23:50,280 --> 00:23:52,720 I was like, Oh, I'm really good at this and I'm doing this in the right way- ... 381 00:23:52,760 --> 00:23:55,980 et cetera, et cetera. And then when I ask the AI to sort of review what I had 382 00:23:56,020 --> 00:23:59,260 been doing, the project that I'd been working on, and give me a few insights 383 00:23:59,300 --> 00:24:03,080 into where I could be doing things more effectively it was slightly embarrassing 384 00:24:03,120 --> 00:24:05,440 because I was I was, yeah, using wrong technology. 385 00:24:05,480 --> 00:24:08,600 I was using wrong hosting. Too much of this, too little of that. 386 00:24:08,640 --> 00:24:11,830 Models that were much too expensive for what I was trying to do. 387 00:24:12,220 --> 00:24:15,940 And when we optimized it and my s- could continue to learn- 388 00:24:16,040 --> 00:24:22,040 the result was awesome. so this sort of i- two things: ask the 389 00:24:22,080 --> 00:24:27,830 technology and constantly be on guard against thinking that you know enough. 390 00:24:28,380 --> 00:24:30,340 I think those are the kinds of things I would advise. 391 00:24:30,380 --> 00:24:31,289 What would you advise? 392 00:24:32,520 --> 00:24:37,070 Well, yeah, I like that thing. As I wrote my first skills 393 00:24:38,120 --> 00:24:42,090 some time ago, and I also was thinking, So where do I start? 394 00:24:42,440 --> 00:24:45,790 And I started by saying, Tell me how to m- how to do this. 395 00:24:46,080 --> 00:24:46,380 Yes. 396 00:24:46,420 --> 00:24:51,060 so we're coming back to that. the other thing there I think also is fascinating 397 00:24:51,100 --> 00:24:54,890 really about asking those questions in order to learn. 398 00:24:55,880 --> 00:25:00,860 I have the chance to help the business school here in Helsinki around these 399 00:25:00,900 --> 00:25:05,100 topics. And one of the things that I always tend to ask them is that, so if I 400 00:25:05,140 --> 00:25:10,190 don't understand something, how many questions do I dare to ask the professor? 401 00:25:11,280 --> 00:25:12,410 Probably two. 402 00:25:13,620 --> 00:25:16,750 And if I still don't understand, I, I'll just say, Oh, okay. 403 00:25:16,880 --> 00:25:20,490 I and then I'll go home and trying to figure that out or I ask my friends. 404 00:25:21,320 --> 00:25:25,160 what's wonderful with these tools is that as you said, you know, you can, you 405 00:25:25,200 --> 00:25:29,280 can, you can continuously ask, So sorry, I still didn't understand. 406 00:25:29,320 --> 00:25:32,850 Can you please make this a more simple ways? 407 00:25:32,900 --> 00:25:33,800 or 408 00:25:34,560 --> 00:25:38,580 as you, as you said, I wouldn't dare saying that, just explain to me like I'm 12 409 00:25:38,620 --> 00:25:40,990 year old because I, I'm sure that those 12-year-olds- 410 00:25:41,340 --> 00:25:42,640 No, they're smarter than we are now- 411 00:25:42,760 --> 00:25:43,620 for sure. yeah 412 00:25:43,660 --> 00:25:44,590 how to use these- 413 00:25:44,680 --> 00:25:45,640 I completely agree with that 414 00:25:45,680 --> 00:25:45,900 these tools 415 00:25:45,940 --> 00:25:46,600 They're very smart. 416 00:25:46,640 --> 00:25:50,160 but I completely agree with you. I think that's one of those things that where 417 00:25:50,200 --> 00:25:55,550 you can really f- find amazing ad- advantages of... 418 00:25:56,240 --> 00:25:58,910 and also as you said what am I, what am I missing? 419 00:25:59,620 --> 00:26:05,580 I wrote a keynote a few days ago, and then one of the things I asked also 420 00:26:05,620 --> 00:26:09,670 was that tell me what areas of this keynote are dated. 421 00:26:10,500 --> 00:26:14,940 And it said, and it said, Oh yeah, you know, you, these examples are quite old. 422 00:26:14,980 --> 00:26:16,680 You know, you should, you should probably use this. 423 00:26:16,720 --> 00:26:21,690 I, I'll g- I'll find you a better example that is closer to today and so on. 424 00:26:22,080 --> 00:26:23,780 And I thought that was really good also. 425 00:26:23,820 --> 00:26:24,720 and 426 00:26:25,520 --> 00:26:30,790 it also added, said so you know, just don't talk about, you know in a way that 427 00:26:31,080 --> 00:26:35,860 using it as a chatbot or like a, like a instead of a of a, of a Google Search 428 00:26:35,900 --> 00:26:40,480 box, but really about how do you take that into agents and so on. 429 00:26:40,520 --> 00:26:40,960 And I- 430 00:26:41,060 --> 00:26:46,350 I thought it was it gave me good insights again in what I was missing. 431 00:26:46,660 --> 00:26:48,780 And I say, Oh yeah, you know, you, you're right. 432 00:26:48,820 --> 00:26:53,000 I am missing those things. so I think you know, keeping you alert all the, all 433 00:26:53,040 --> 00:26:57,550 the time also and say, you know, what am, what are areas did I not think of- 434 00:26:58,320 --> 00:27:00,400 I think are absolutely key also. 435 00:27:00,440 --> 00:27:04,340 And I think the really good part of that, what you're describing, is that you 436 00:27:04,380 --> 00:27:07,800 are also starting with your own thinking, which means that you're not handing 437 00:27:07,840 --> 00:27:10,000 over the thinking from the get-go to the machine. 438 00:27:10,040 --> 00:27:12,510 And I think we should also talk about the anti-patterns. 439 00:27:12,760 --> 00:27:15,800 When I talk to people about AI, one of the things I really try to say, there are 440 00:27:15,840 --> 00:27:17,490 ways in which you should not use it. 441 00:27:17,920 --> 00:27:21,080 So I there are a few simple tips that I always come back to. 442 00:27:21,120 --> 00:27:26,160 One is use more than one AI. Ask it to take the opposing stances. 443 00:27:26,200 --> 00:27:28,560 So one is pro something, one is con something. 444 00:27:28,600 --> 00:27:31,310 That's a really good way to make sure that you are in the driver's seat. 445 00:27:31,640 --> 00:27:35,140 It's a judgment-centric setup. The other thing that I think is really important 446 00:27:35,180 --> 00:27:40,520 is start with human thinking. I'm allergic to this notion that the AI can give 447 00:27:40,560 --> 00:27:42,360 you a draft and then you can expand on it. 448 00:27:42,400 --> 00:27:44,370 That's exactly the wrong way around. 449 00:27:44,879 --> 00:27:48,840 I sit down now, I write longhand when I do a talk or an, you know, an essay or 450 00:27:48,880 --> 00:27:52,740 something. I write longhand for a couple of hours sometimes, and then I take 451 00:27:52,780 --> 00:27:57,160 photos of my longhand writing. I digitize it, and then I ask for comments and 452 00:27:57,200 --> 00:28:00,510 ideas and, you know, things I missed to your point about the keynote. 453 00:28:00,940 --> 00:28:05,250 But if I would start with asking the AI, I'm giving a keynote on this. 454 00:28:05,380 --> 00:28:09,620 give me a few ideas, that's n- at that point I don't think I am necessary 455 00:28:09,660 --> 00:28:11,530 anymore. And then I think also- 456 00:28:11,860 --> 00:28:17,130 at that point we're missing human originality and human creativity, which still, 457 00:28:17,280 --> 00:28:20,330 I believe, is greater than that you can get from machines. 458 00:28:20,660 --> 00:28:24,540 Also because the creativity in the machines is derived from human creativity. 459 00:28:24,580 --> 00:28:29,080 We are c- we are capable of creating truly novel things, so we should start with 460 00:28:29,120 --> 00:28:34,120 the human. So judgment centric, always start with the human are things that I 461 00:28:34,160 --> 00:28:37,310 think are so important when it comes to the use of AI. 462 00:28:37,600 --> 00:28:41,400 Otherwise you end up in what's sort of now popularly called cognitive offloading 463 00:28:41,440 --> 00:28:44,940 or cognitive surrender where you're essentially saying, Oh, there's this 464 00:28:44,980 --> 00:28:50,340 thinking machine here. let's have it think for me because that seems like okay, 465 00:28:50,380 --> 00:28:52,990 and I don't have the energy for thinking right now. 466 00:28:53,400 --> 00:28:55,990 And that's a path we should absolutely avoid, I think. 467 00:28:57,180 --> 00:29:02,770 That I think is one of the key learnings from the last half hour- 468 00:29:03,260 --> 00:29:09,200 if I were to put it that way, is that always start by thinking 469 00:29:09,240 --> 00:29:10,210 yourself first. 470 00:29:11,260 --> 00:29:14,460 And I think that is something that it's like if there's nothing else you take 471 00:29:14,500 --> 00:29:15,320 out of this- 472 00:29:15,580 --> 00:29:19,270 I would say that. And this is something also that I have been very much 473 00:29:19,420 --> 00:29:25,280 emphasizing at the business school is that we need to start, always start 474 00:29:25,320 --> 00:29:29,890 thinking first as well as I think business schools need to teach thinking. 475 00:29:30,440 --> 00:29:30,880 Yes. 476 00:29:30,920 --> 00:29:36,830 they fa- teach concepts and teach thinking because through that we can then 477 00:29:37,040 --> 00:29:40,830 use these assistants in driving different things. 478 00:29:42,320 --> 00:29:45,250 what other things would you not hand over to AI today? 479 00:29:46,980 --> 00:29:50,760 So I... The most important thing I think is what we just talked about. 480 00:29:50,800 --> 00:29:55,010 I wouldn't hand over the first draft to an AI at all. 481 00:29:55,380 --> 00:29:59,790 I wouldn't, I also wouldn't hand over reading to an AI. 482 00:30:00,060 --> 00:30:03,700 So I actually tend to believe that reading is really important, and I like to 483 00:30:03,740 --> 00:30:08,900 read myself. I don't want just to have an AI sum up what an article says or sum 484 00:30:08,940 --> 00:30:12,920 up what a book says. Sometimes if it's in the periphery, it can be fine to just 485 00:30:12,960 --> 00:30:15,030 know what the main argument in a book is. 486 00:30:15,300 --> 00:30:18,860 But if I'm really working on something, then I want to really get into it 487 00:30:18,900 --> 00:30:22,510 myself, and the experience of reading a book changes you. 488 00:30:22,820 --> 00:30:26,960 And that means that next time you sit down and write, you have new things to 489 00:30:27,000 --> 00:30:29,810 write about. So I don't hand off reading. 490 00:30:29,920 --> 00:30:32,970 I don't hand off the first draft. I don't hand off judgment. 491 00:30:33,260 --> 00:30:37,400 Those are the key things that I try to make sure remain v- very solidly within 492 00:30:37,440 --> 00:30:39,090 my p- purview. 493 00:30:40,340 --> 00:30:43,010 and I think those are the key things. 494 00:30:44,220 --> 00:30:47,860 well, what I also have found is I need to be very clear when handing off 495 00:30:47,900 --> 00:30:51,440 scheduling because I was at some point scheduled to have 20 meetings in a day 496 00:30:51,480 --> 00:30:56,000 because the AI was going for efficiency and not necessarily for just human 497 00:30:56,040 --> 00:31:01,040 survival at this point. so there are a few things where, you know, you have to 498 00:31:01,080 --> 00:31:03,350 be very clear about your delegation as well. 499 00:31:04,080 --> 00:31:05,310 and I think that's important. 500 00:31:05,660 --> 00:31:11,520 And maybe to offer a more a more complex point, I think y- what you do 501 00:31:11,560 --> 00:31:16,810 not want to do is to hand over to the AI to decide your ontology for you. 502 00:31:17,260 --> 00:31:21,930 An ontology is like a very, it's like a fancy word of the stuff that exists. 503 00:31:22,360 --> 00:31:26,160 If you, if you sort of take your world description from your AI, then you will 504 00:31:26,200 --> 00:31:30,660 be limited in the way you think. You should own the way you describe the world, 505 00:31:30,700 --> 00:31:33,970 its ontology, the stuff that exists, how it interrelates. 506 00:31:34,440 --> 00:31:36,670 That is something that i- belongs to you. 507 00:31:36,900 --> 00:31:42,810 Being able to specify, describe, invent ontologies or world models, 508 00:31:43,160 --> 00:31:47,290 I think is a key to the long-term use of complex AI as well. 509 00:31:48,920 --> 00:31:50,730 What about you? What do you not hand over? 510 00:31:52,280 --> 00:31:58,320 Well, we mentioned already about the thinking and 511 00:31:58,360 --> 00:32:03,670 the beginning of that process. otherwise you'll get the averages of averages, 512 00:32:04,040 --> 00:32:07,050 and you don't get the thinking of, so to say 513 00:32:09,180 --> 00:32:13,050 an unorthodox thinking as well. so you're missing that. 514 00:32:14,800 --> 00:32:19,780 there are th- areas of data, for example financial data, that I'm very hesitant 515 00:32:19,820 --> 00:32:23,690 with. There are health data today where I'm very hesitant with. 516 00:32:23,740 --> 00:32:28,440 And so there are areas like this, and I don't know should I be that hesitant or 517 00:32:28,480 --> 00:32:30,410 not, but I am very hesitant with those. 518 00:32:30,840 --> 00:32:36,360 and the question always obviously is that do you trust this not to use your 519 00:32:36,400 --> 00:32:42,340 data in the, in the training? do you trust them not to leak your data and so 520 00:32:42,380 --> 00:32:47,300 on? And I don't have the answer to that right now and but it makes me a bit 521 00:32:47,340 --> 00:32:52,830 hesitant. And sometimes I feel like I could probably be more efficient 522 00:32:54,160 --> 00:32:59,740 if I would trust them more, so to say. The question is that do I dare to trust 523 00:32:59,780 --> 00:33:01,990 them more? Do you trust this? 524 00:33:02,220 --> 00:33:04,720 Well, no, I ma- I think this is a good point. 525 00:33:04,760 --> 00:33:07,240 I think it's not necessarily about the handing over the data. 526 00:33:07,280 --> 00:33:09,960 What sort of concerns me is the use and ownership of the data. 527 00:33:10,000 --> 00:33:13,690 But so to me, this is actually an area for massive innovation. 528 00:33:14,180 --> 00:33:17,600 I would- be happy to hand over my health data to a trust, and then have- 529 00:33:17,680 --> 00:33:21,460 that trust anonymize and aggregate the data in such a way that AI companies 530 00:33:21,500 --> 00:33:23,280 could do research on it. I would actually- 531 00:33:23,360 --> 00:33:26,580 quite I would love that because that would mean that I could be- have better 532 00:33:26,620 --> 00:33:28,180 health outcomes in different ways. 533 00:33:28,220 --> 00:33:28,640 Correct. 534 00:33:28,680 --> 00:33:32,790 So for me that would be good. But what we lack here is institutional innovation. 535 00:33:33,000 --> 00:33:36,260 And the kind of institutional innovation that's needed is essentially innovation 536 00:33:36,300 --> 00:33:40,770 that allows for individual trust and technological learning. 537 00:33:40,940 --> 00:33:45,280 so how do you enable the maximum amount of learning on the side of the AI 538 00:33:45,320 --> 00:33:48,850 companies and the maximum amount of trust on the side of the users? 539 00:33:48,900 --> 00:33:54,170 I think the, it's not inconceivable to me that there could be like a third-party 540 00:33:54,300 --> 00:33:58,970 institutional solution to that would be quite interesting- 541 00:33:59,260 --> 00:34:02,560 because I would love to, I would love to have an AI look at my health data and 542 00:34:02,600 --> 00:34:02,820 say- 543 00:34:02,860 --> 00:34:03,180 Agree 544 00:34:03,220 --> 00:34:06,710 You do this and this and this, and you'll be, you know, feeling much better. 545 00:34:07,080 --> 00:34:09,230 Agree. Finan- finance is the same thing. 546 00:34:09,560 --> 00:34:12,920 Yeah, totally. If, you know, if you could hand over your finances, you could 547 00:34:12,960 --> 00:34:16,600 hand over... I think finances is interesting because we're all trying to get an 548 00:34:16,640 --> 00:34:20,070 individual edge. There's like a really interesting risk being discussed by the, 549 00:34:20,160 --> 00:34:23,980 by the central banks right now, which is that may- people actually do hand over 550 00:34:24,020 --> 00:34:27,660 their finances to ChatGPT. So they will ask ChatGPT, What should my portfolio 551 00:34:27,700 --> 00:34:28,780 look like? And- 552 00:34:28,940 --> 00:34:30,880 then you get a portfolio recommendation. 553 00:34:30,920 --> 00:34:34,180 But then what happens if 100 million people get the same portfolio 554 00:34:34,220 --> 00:34:38,640 recommendation and ChatGPT one morning wakes up and says, You should sell 555 00:34:38,680 --> 00:34:41,180 Nvidia. And the 100 million people managing- 556 00:34:41,240 --> 00:34:43,330 the portfolios through ChatGPT does that. 557 00:34:43,440 --> 00:34:45,580 The volatility that then results. 558 00:34:45,620 --> 00:34:49,170 But now you outsource your thinking. 559 00:34:49,680 --> 00:34:49,900 Yes. 560 00:34:49,940 --> 00:34:53,410 And we said, we just said earlier that we should not outsource the thinking. 561 00:34:53,720 --> 00:34:57,840 What I would use it for more or less is that, for exam- sample say that what's 562 00:34:57,880 --> 00:35:00,950 my exposure to hyperscalers? 563 00:35:01,600 --> 00:35:06,200 Because like you have ETFs, funds. You might have even bonds from them. 564 00:35:06,240 --> 00:35:09,210 You might have, you know, stocks from them and so on. 565 00:35:10,140 --> 00:35:14,950 very rarely can anybody tell me today is that what's my w- and through that, 566 00:35:15,040 --> 00:35:18,350 what's my the, w- what's the risk that I'm taking? 567 00:35:19,080 --> 00:35:21,840 Because like probably these are all moving in the same direction. 568 00:35:21,880 --> 00:35:27,840 and so I would, I would use my 100 assistants to dig into 569 00:35:27,880 --> 00:35:33,160 these, to these instruments and then do calculations of how do they correlate 570 00:35:33,200 --> 00:35:36,440 with each other and how they don't correlate with each other and what's the risk 571 00:35:36,480 --> 00:35:39,100 there and is there a beta risk there and so on and so on. 572 00:35:39,140 --> 00:35:44,024 so I'm just saying that I think I would be willing to do that Which would help 573 00:35:44,064 --> 00:35:45,614 me to make better decisions. 574 00:35:46,044 --> 00:35:46,364 Yes. 575 00:35:46,404 --> 00:35:49,914 But I would not ask it to make that decision for me, so to say. 576 00:35:50,404 --> 00:35:50,704 Mm-hmm. 577 00:35:50,744 --> 00:35:54,454 And so that's, I think those are those areas where I would really be 578 00:35:56,544 --> 00:36:02,384 I would love it to work. I would also be probably willing to do it if, and I 579 00:36:02,424 --> 00:36:08,124 probably need to do some learning here, but so that I could run those locally on 580 00:36:08,164 --> 00:36:09,644 m- on my, on my own laptops. 581 00:36:09,684 --> 00:36:10,064 Yeah. 582 00:36:10,104 --> 00:36:14,994 And then I would be able to do those a- analysis. 583 00:36:15,284 --> 00:36:17,144 that's probably could be also one solution. 584 00:36:17,184 --> 00:36:20,924 And it's quite possible to do today. I mean, you can run some of the open weight 585 00:36:20,964 --> 00:36:23,264 models on your local MacBook if you want to. 586 00:36:23,304 --> 00:36:23,624 So it's- 587 00:36:23,664 --> 00:36:24,164 Correct. Yeah 588 00:36:24,204 --> 00:36:26,544 it is, it is probably going to more and more common. 589 00:36:26,584 --> 00:36:31,304 And you can imagine that kind of federated edge computing for AI being a thing 590 00:36:31,344 --> 00:36:35,144 for a lot of things that you want to do for, you know, photos, et cetera, et 591 00:36:35,184 --> 00:36:37,104 cetera, that you don't want to upload to the cloud. 592 00:36:37,144 --> 00:36:40,404 So I think that will probably come. But there's s- there's something important 593 00:36:40,444 --> 00:36:44,284 about what you said about finances and you can sort of sum that up as a method, 594 00:36:44,324 --> 00:36:48,734 which is that what you do with finances is that you start with a hypothesis. 595 00:36:48,864 --> 00:36:52,704 You have a hypothesis about what good finance m- portfolio management looks 596 00:36:52,744 --> 00:36:53,304 like. 597 00:36:53,364 --> 00:36:56,504 And then you ask the machine to test the hypothesis. 598 00:36:56,544 --> 00:37:00,064 And I think that's a, that's another way of framing think first, is saying you 599 00:37:00,104 --> 00:37:03,184 should have a hypothesis about what you think the world looks like or what you 600 00:37:03,224 --> 00:37:06,264 want to do or what you think the solution to the problem is, then use the 601 00:37:06,304 --> 00:37:09,804 machine to test the hypothesis. Never allow the machine to give you the 602 00:37:09,844 --> 00:37:13,864 hypothesis. I think that's a, that's another way of sort of casting the problem. 603 00:37:13,904 --> 00:37:16,964 I think you're hitting on a, on a, on a, on a really important note there, and I 604 00:37:17,004 --> 00:37:20,104 think this is one of those things that we need to cover in one of our podcasts 605 00:37:20,144 --> 00:37:20,544 later is- 606 00:37:20,584 --> 00:37:20,904 Yes. 607 00:37:20,944 --> 00:37:22,094 for example, board work. 608 00:37:22,784 --> 00:37:26,314 And how do you build strategies? How do you test strategies? 609 00:37:26,644 --> 00:37:31,204 How do you look at... 'Cause one of the most important w- jobs of a, of a, of a 610 00:37:31,244 --> 00:37:35,844 board is obviously oversight and see that everything goes well, but also is to 611 00:37:35,884 --> 00:37:37,374 evaluate strategies. 612 00:37:37,904 --> 00:37:41,444 And what you can do with these things are absolutely amazing. 613 00:37:41,484 --> 00:37:45,764 And for example, I've built my own supervisory boards in order to make those, 614 00:37:45,804 --> 00:37:50,584 some of those decisions. We'll get back to that in another episode. 615 00:37:50,644 --> 00:37:54,034 But I think that I think you're hitting one of the key things here. 616 00:37:54,384 --> 00:37:58,194 I have two quick questions for you, and then let's wrap up. 617 00:37:59,564 --> 00:38:04,964 one for me is that if there would be one habit for all the listeners to start 618 00:38:05,004 --> 00:38:06,974 with this week, and we're now 619 00:38:08,684 --> 00:38:11,454 middle of the week, so what would that be? 620 00:38:12,704 --> 00:38:17,424 Habit of asking the AI to teach you something about how you can use AI in order 621 00:38:17,464 --> 00:38:20,504 to become more, in order to learn faster. 622 00:38:20,544 --> 00:38:23,794 That's a habit that I think is really good to acquire quickly. 623 00:38:24,824 --> 00:38:25,974 What would your choice be? 624 00:38:26,664 --> 00:38:32,404 Oh, I like, I like that. I think one of the things really is to dig into 625 00:38:32,444 --> 00:38:33,754 these tools. 626 00:38:35,824 --> 00:38:41,524 I've been saying quite often that my imagination is the limit for what I can do 627 00:38:41,564 --> 00:38:42,464 with these things. 628 00:38:42,724 --> 00:38:46,524 So I think the first question I always should ask is that, so could this be 629 00:38:46,564 --> 00:38:47,464 automated? 630 00:38:47,764 --> 00:38:49,844 What I'm now doing, could it be automated? 631 00:38:49,884 --> 00:38:50,834 And if it could 632 00:38:52,864 --> 00:38:57,174 aut- automate it. I have one big question coming up. 633 00:38:57,584 --> 00:38:58,484 Uh-oh. 634 00:38:59,044 --> 00:39:01,194 What does the world look like a year from now? 635 00:39:03,104 --> 00:39:07,304 What does your crystal ball say? I know you have it in front of you, and you've 636 00:39:07,344 --> 00:39:09,764 been thinking a lot about this, and I know w- well, you have one of those 637 00:39:09,804 --> 00:39:14,264 companies that are also looking into more as forecasting, foresight, and so on. 638 00:39:14,324 --> 00:39:16,564 Yeah, we do foresight. We do forecasting, et cetera. 639 00:39:16,604 --> 00:39:21,664 I think, I think it's really interesting because to me I think the evolution of 640 00:39:21,704 --> 00:39:23,684 AI is going much faster than we recognize. 641 00:39:23,724 --> 00:39:27,374 So I think that the world might be very different in a year than it is now. 642 00:39:27,764 --> 00:39:31,664 So I think, for example, that we will see agents fully integrated with 643 00:39:31,704 --> 00:39:33,134 everything that you can do on a screen. 644 00:39:33,564 --> 00:39:35,704 I think that is going to be a difference. 645 00:39:35,744 --> 00:39:39,704 So you can ask an AI, and you can do it in voice mode or any other mode, to do 646 00:39:39,744 --> 00:39:41,694 anything that you can currently do on a screen. 647 00:39:41,904 --> 00:39:45,764 And if you think about what that means, it means that, you know, the screens and 648 00:39:45,804 --> 00:39:48,424 the keyboards might no longer actually be necessary. 649 00:39:48,464 --> 00:39:53,244 So one of my sort of soft predictions would be that we probably will see new 650 00:39:53,284 --> 00:39:57,984 form factors for AI within the year because nobody has cracked what the device 651 00:39:58,024 --> 00:40:00,704 looks like that encapsulates AI in a great way. 652 00:40:00,744 --> 00:40:02,843 We're still using phones. We're still using computers. 653 00:40:02,883 --> 00:40:06,254 We're still using keyboards, and I think that might change. 654 00:40:06,324 --> 00:40:09,904 So I will be really interested to see what the form factor is, and I do think 655 00:40:09,944 --> 00:40:15,194 that we will be delegating a whole lot more to our collection of agents. 656 00:40:15,424 --> 00:40:21,384 And then I think that within a year we will also have experienced a lot more in 657 00:40:21,424 --> 00:40:24,544 terms of security incidents, and that has just started. 658 00:40:24,584 --> 00:40:25,084 Mm-hmm. 659 00:40:25,124 --> 00:40:27,434 So our systems will become less connected. 660 00:40:27,904 --> 00:40:32,044 I predict, and this is not going out on a limb, that freezers will not be 661 00:40:32,084 --> 00:40:34,194 connected to the internet in a year from now. 662 00:40:34,504 --> 00:40:37,534 Yes. Yeah. I think that's 663 00:40:38,964 --> 00:40:42,424 I, I'd love to come back on that in a year's time to see what people- 664 00:40:42,464 --> 00:40:45,834 yes. You should hold me accountable. You should hold me accountable 665 00:40:46,284 --> 00:40:47,934 what happened. 666 00:40:49,024 --> 00:40:54,044 topics we want to cover in this going forward, I mentioned already we need to be 667 00:40:54,084 --> 00:40:59,964 looking into how should AI sit on the boards table- board table, and how should 668 00:41:00,004 --> 00:41:04,244 actually every single board member in the future and already today have, so to 669 00:41:04,284 --> 00:41:06,784 say, their own agents to do the, to do research. 670 00:41:06,824 --> 00:41:12,344 I think that is a fundamental change that we will see in my view, how 671 00:41:12,384 --> 00:41:14,174 companies are managed. 672 00:41:15,464 --> 00:41:16,364 data centers. 673 00:41:16,764 --> 00:41:17,664 Data centers. 674 00:41:18,064 --> 00:41:22,934 Well, I think we should absolutely look into that topic. 675 00:41:24,344 --> 00:41:28,994 you mentioned the data security and all of this area. 676 00:41:29,744 --> 00:41:31,874 I think that should be one of those areas. 677 00:41:32,204 --> 00:41:34,114 I would love to also for us to discuss 678 00:41:35,184 --> 00:41:36,084 a little bit 679 00:41:37,004 --> 00:41:41,774 how do we trust those tools, and can those tools be trusted, and can we... 680 00:41:41,864 --> 00:41:45,264 what should companies... Like I see so many companies today, they say, Okay, you 681 00:41:45,304 --> 00:41:48,474 know what? We have this license with Microsoft Copilot. 682 00:41:49,044 --> 00:41:50,804 you are not allowed to use anything else. 683 00:41:50,844 --> 00:41:55,944 And then the employees or the students, they use all kind of other tools, so to 684 00:41:55,984 --> 00:41:56,644 say, widely. 685 00:41:56,684 --> 00:41:58,844 Oh, there's huge shadow use. I think you're right. 686 00:41:58,884 --> 00:41:59,184 Exactly. 687 00:41:59,404 --> 00:42:04,104 Exactly. and what we can do with that. Any other topics you think we should be 688 00:42:04,144 --> 00:42:04,964 covering? 689 00:42:05,004 --> 00:42:07,044 Yeah, I think, well, I think we talked about this before. 690 00:42:07,084 --> 00:42:09,664 One, there are a couple of things that I think we should add to your excellent 691 00:42:09,704 --> 00:42:12,204 list, and one of them is we should talk about AI in the labor market. 692 00:42:12,244 --> 00:42:12,984 What will actually happen- 693 00:42:13,164 --> 00:42:13,784 to your job? 694 00:42:13,824 --> 00:42:14,184 Yes. 695 00:42:14,224 --> 00:42:18,314 We should also talk about AI and universities and higher education. 696 00:42:18,604 --> 00:42:21,954 How do you think about AI when everyone can write an excellent essay- 697 00:42:22,284 --> 00:42:22,624 Yes 698 00:42:22,664 --> 00:42:26,804 with, you know, pushing a few buttons? How do you reorganize higher education 699 00:42:26,844 --> 00:42:31,674 and research around that? And then I think there is, there is there is, there is 700 00:42:31,744 --> 00:42:36,334 also I think at some point we should just tackle the geopolitics of the thing. 701 00:42:36,424 --> 00:42:39,054 The US, China, Europe sovereignty. 702 00:42:39,124 --> 00:42:39,424 Absolutely. 703 00:42:39,464 --> 00:42:42,584 Who does what, who, and, you know, and what do we need to be able to do in 704 00:42:42,624 --> 00:42:48,534 Europe, and how should we put some clout behind European regulation and demands? 705 00:42:48,984 --> 00:42:51,364 And should we, or do we risk sliding behind? 706 00:42:51,404 --> 00:42:53,944 So I think those are things I would add to your list, and I would be- I'm 707 00:42:53,984 --> 00:42:55,334 looking forward to discussing them. 708 00:42:55,824 --> 00:42:58,734 Fantastic. So am I. Thank you very much, Nicklas. 709 00:42:58,824 --> 00:43:00,564 It was great catching up with you as always. 710 00:43:00,604 --> 00:43:02,284 Likewise, Jussi. 711 00:43:02,324 --> 00:43:05,194 Enjoy the bright fall in the archipelago- 712 00:43:05,364 --> 00:43:06,444 I definitely will 713 00:43:06,484 --> 00:43:09,774 in Sweden, and we will be catching up very soon. 714 00:43:10,064 --> 00:43:11,174 We will. Thank you. 715 00:43:13,900 --> 00:43:16,400 You've been listening to You are on mute! 716 00:43:16,500 --> 00:43:19,800 Follow the show so you don't miss the next episode.