E180 - AI in Banking: Innovation vs Trust - Stephan Erne

In Episode 180 of the AIAW Podcast, we’re joined by Stephan Erne, former Chief Digital Officer at Handelsbanken, to explore how AI can move from experimentation to real impact in banking and other regulated industries. Drawing on his experience leading digital transformation across the Nordics and Germany, Stephan shares how organizations can align technology, compliance, and business to turn AI into a core value driver. We discuss trust and transparency in financial services, the role of leadership and culture in scaling AI, and how emerging technologies like blockchain and quantum computing may reshape the industry. From the human role in an increasingly automated world to the long-term outlook toward AGI, this episode focuses on what it takes to build resilient, trustworthy AI in complex environments.

Follow us on youtube: https://www.youtube.com/@aiawpodcast

Transcript

Criticality Drives Sovereignty Choices

Stephan Erne 0:00

I think you have to differentiate what kind of data are we talking about, what kind of services are behind it. Uh is it a a help service like Collaboration Suite or HR? Or is it your core service like your payment services, your mortgage stuff, which is also the the the mission critical, the the the system critical services you're offering? And depending on um if they're more critical than more sovereignty. I mean, some companies they're filling up their own uh diesel tanks to have auxiliary power because they don't even trust or they they think even think about what happens if the power supply doesn't work anymore.

Henrik Göthberg 0:42

Yeah, but but this whole topic then with sovereignty, like I think step one nuanced, and then basically uh uh strategies uh fit for fit for the criticality is is is an obvious topic. The other key topic I think is how we need to have plan A continue as we have gone and then have serious continuous planning. Because sometimes also I see that I mean, like you see that you see, you know, this the starting point of the conversation is like talking to so around serenity with with the high IT leaders in different types of industries, how different we are viewing the topic. But I think the interesting topic, or the interesting follow-on question I would like to ask you is like how many of them, when you're discussing this, are highly are consciously going for a plan A, plan B strategy, or are they used pursuing whatever strategy they have?

Stephan Erne 1:38

My impression is that um I mean I think uh all the companies have kind of some kind of strateg or had some kind of strategy. How do they consume or or create IT services going forward? Uh and some have a very radical strategy that they want to move everything to the cloud, and others say, well, some things we want to stay uh keep here. And what I think is happening now is that the ones that uh maybe the ones that were pretty aggressively said we're gonna move everything there, they have a pretty hard time now because they didn't make this differentiation that I was talking about before.

Anders Arpteg 2:14

But it's also I think everyone wants to reduce the dependency to the top cloud providers in US, but there it comes at a certain cost as well. Of course, they are better than anyone in providing the best services and the best quality and the best security, etc. How do you consider that kind of gap to be?

Stephan Erne 2:32

I mean, and that's why to be honest, I don't I don't agree to the statement that everybody wants to uh decrease the dependency. Right. I think a lot of companies don't have this big problem because they understand that the the c from a cost perspective, from a functionality perspective, that all the alternatives in doing it yourself or hosting it closer to your heart or even in Europe, they have some advantages, but on the other hand, the disadvantages are so much higher that it's not a feasible option in the end. It's not really feasible. And that's why they're rather and and this is why you come, of course, to this risk risk approach. Yeah, what what what kind of risks do you have on the table? Was it only the risk that somebody uh pulls the kill switch, or is it also the risk that your services are not up to date anymore and that you're you're not able to handle your customers going forward? So and then and this coming back to the differentiation where you also have to explain that risk is not only continuity or um or um or sovereignty and then having having uh one risk vector. It's one of the vectors, yeah.

Henrik Göthberg 3:36

Yeah, but and I have a short anecdote that really maybe makes it easy to understand the fancy words we're talking when we say differentiation. So I was heading up the data platform and understanding for how we go to cloud in with data in um uh Vattenfall, I mean, many um uh some years back. And we had requirements coming in from different angles, and of course, then we had requirements from nuclear. And of course, we had like different uh security zones where the deepest security zone is terminal server, i.e., not connect not connected to anything. And then basically the the discussion when it's bipolar is like we need to be able to manage all security concerns. So all of a sudden now we are pushing the nuclear agenda to do something for all of Wattenfall. And then I basically, okay, that's gonna be quite costly, right? So, how many people are we talking about in nuclear that we want to serve with this that is critical and and um and then what became sort of millions and billions in terms of do five the strategic architecture for Wattenfall? Boil down to 25, 30 people, and I say, I think I can do a sidecar strategy for these guys that are even more safe, you know, and and this is a differentiated stuff uh in a nutshell, right?

Stephan Erne 4:56

But what's your view on this uh sovereign topic? Do you think uh different companies acting very differently, or do you think that's a very common um approach?

Anders Arpteg 5:05

I think a lot of people take it in a very subjective way and a political way rather than objective way. Yes. And I agree with what you're saying there. They need to look at the whole picture. You know, what is really the consequence if you do reduce the functionality that they do provide? I don't think I don't think people understand the level of quality that the top cloud providers do provide and what the consequences would be of getting away from them. I mean, even at uh when working in Spotify, you know, we we first went to the cloud, then went on-prem, and then we realized, you know, doing all of this on-prem takes away so much of the core you know functionality that we want to provide, and we need to focus on what the core you know USB and the functionality should be for our company, now we can't we can't operate the hard drives, we can we can't even operate the the Kubernetes engines properly, we can't even operate you know all the other services that do provide today that are you know fundamental in some sense. So if we instead can focus on the top core functionality and let the cloud providers, especially the software functionality they do provide, and they do have a functionality that is so much better than others, then we can accelerate much further. And especially if you go to AI and the quality that they have in their kind of AI and data services, if you wouldn't have that, we couldn't gain as much value as we otherwise could, I would argue.

The Nuclear Sidecar Strategy Story

Henrik Göthberg 6:30

But and and and I concur with well, I mean like I'm I'm it's not even it's even boring because we are we it's not an age. I agree, but let me add a nuance to what he's saying, which is we are also agreeing with, but which is a different angle into the topic. So let's talk about sovereign AI, not sovereign cloud, but sovereign AI. So that sounds great on principle level, and it means oh, we need to go out and we need to create a sovereign AI. So to say that is also a very, very nuanced question in terms of what is it that we really need to do to get sort of a language model, a foundation model that is culturally correct for Sweden or whatever. And and here now, once again, if you are bipolar, you say, Oh, we need to be world-class in building our own super large-scale foundation models. So we need to start the sovereignty from the lowest ground. And then and then if you look deeper into that technology, uh, the technology stack and the effort and how you can get to sovereignty that is meaningful with the least cost, then maybe okay, do we really need to invent okay? If you talk about sovereignty of the whole thing, you are really talking about a foundry for chips. No, okay, so that's off the table. Okay, what's the next level? Okay, do we you know so how can we do it then understand how we can maybe use even American uh large language models, and then we start from that base in order to fine-tune or distill out the models we need in a sovereign way for ourselves. So all of a sudden now we have taken the whole stack and compressed it to the core strategic pieces of the puzzle where we think sovereignty is key. For instance, culturally sounding advice uh in terms of uh whatever we want to, how we want the LLM to work or how we want it to work in banking or whatever. So this whole sovereignty thing, I if I take it down to enterprise level and and and and banking, of course, maybe you should work you shouldn't start from scratch, but you should maybe do it something and then you need to distill it somewhere, and then you need to do it, you know. So that so that whole the the the main word that you put on the table is differentiated or nuanced understanding for the problem to solving the right problem.

Stephan Erne 8:48

Exactly, and also for your operations, of course. Yeah, I mean, of course, if if you if you use I mean you have to keep your your business running, right? And then the more regulated you are, the more you're under pressure, and you have to prove that you have that you can keep your business running. And in the old days you said the more control I have over this stack, if I have the people in my own building, I I know who I have to call if something goes wrong. That of course gives um gives me some control, control over it, yeah, and security safety. Yeah, and at least and and this is and that's why I think it comes uh here. Yeah, like we said before, the the difference between facts and feelings of also come in. And I mean having having uh how much control do you have over your people sitting here in Sweden?

Sovereign AI Without Building Everything

Henrik Göthberg 9:44

None but you said you you were leading in also saying like it's interesting how it doesn't feel like this is a data-driven decision right now, and and what you meant with that, if I extrapolate, is really the the the the definition and the facts are out there, and still people in the same industry with the same uh regulations and risk around them, in the same market with the same innovation pressure, acts very differently on the same or answers very differently.

Stephan Erne 10:11

So it's it's more like I would say it's not fact-driven. Uh instead of it's data driven, it's not fact-driven, it's based on some bias, some political ideas, some aspirations.

Henrik Göthberg 10:21

Or it's based on the what you happen to have as knowledge at the point in time in that particular unit that is working on this.

Stephan Erne 10:30

Yeah, but I think it's also based on what do you believe will happen going forward. And this is where I think the old way of creating strategy or building strategy is not the only way or just collecting data and facts and try that you can take the right decision based on the facts that you have on the table doesn't work anymore. You have to create some scenarios because there's such a high uncertainty. I mean, I guess you've seen all these uncertainty index indices. I think we have five times more in uncertainty right now than we had during COVID. So, in this uncertainty, how do you think you can oh you can make decisions only based on the facts of today without having a common understanding how the future will look like?

Meet Stefan And His Background

Anders Arpteg 11:15

Not easy, and uh rabbit hole in itself. I added this as a topic as well. Perhaps we can get back to this topic about becoming sovereign in uh in the world of today and the need for it or not need for it. That would be super fun to add. But before that, very welcome here, Stefan Erne.

Stephan Erne 11:34

Is that the proper name? Stefan Erneople normally in Germany. It's a Düsseldorf name. You're from Dusseldorf. I'm from Düsseldorf. Yeah. Well I usually speak very very clear German, I would say. It says Hoch German. Hochdeutsch, genau Deutsch.

Anders Arpteg 11:54

Well, former CDO or chief digital officer at Handels Banken as well, and also a prominent uh advocate, I think, you know, for a number of technology techniques, which I really much looking forward to.

Henrik Göthberg 12:05

Yeah, and he was the moderator at the CIO Nordics just this week we concluded, yeah. So those kind of gigs as well.

Anders Arpteg 12:12

Not only speaking about AI perhaps tonight, uh also perhaps looking forward to speaking about quantum computing, perhaps, and uh even the cloud and other topics. Uh that would be super, super fun. But before that, perhaps you can give a quick introduction uh about really who is Stephen Stephen. Stefan Stefan. Stefan, Stefan.

Henrik Göthberg 12:33

You need to work on your German pronunciation. Stefan!

Anders Arpteg 12:37

Yes, please.

Stephan Erne 12:39

Yeah, yeah, sure. And that yeah, feel free to to stop me or to to ask more about um well I'm I'm I'm a German engineer, you could say that. So I'm I'm from Düsseldorf, then I studied in Aachen. But then I realized that maybe I should uh get some uh international experience. So I was actually writing my math CC in Stockholm at KTH. Nice nice connection. Came here for a half a year. Um and somehow my ambition, even that I studied a very technical, and I want to be, and uh I would say I I I love technical problems to solve them. So I would really say I'm I'm an engineer, and I guess most of my family and friends would agree. Um, but I want I always wanted to move somehow over to the business side because I realized that if you really want to create an impact with that, you it's not enough to be a cool techie guy. Uh you have to somehow bridge the gap and take the business side with that. The technology needs to be adopted, it needs to be used. So that's why actually I um um I wanted to um and I had the possibility to make a PhD and then as well, but I will I wanted to take a trainee program because my idea was this is the fastest path into uh into the business world. So I uh I actually found a trainee program um and I was open to do this in Sweden, but my Swedish was so bad, so nobody wanted me. Um so I and then I met my wife here as well, so I convinced her to move back to Germany. So we were living in Germany for nine years, and I started a training program at Deutsche Telekom. Deutsche Telekom. Deutsche Telekom, which by that time was I mean the incumbent, old ministry, 350,000 people, old-fashioned, really old-fashioned company, but was on the on the way of getting privatized, uh, making an IPO.

Henrik Göthberg 14:23

Um which years of this. I'm I'm trying to think. We had Rainer Deutschmann here, who was the group's uh COO of Telia, but I think he's he has a Deutsche Telekom background, maybe as well. Could be, yes. No one you know. I was just thinking if there's a connection with that was 96.

Stephan Erne 14:41

96, yeah. That's uh quite a while ago. So I was there doing the trainee program. Um, and then uh everybody told us you have to you have to understand the customers. So from the training program, I actually moved over to uh to to a B2B sales. So became uh actually account manager for Deutscher Bundestag, uh like Riksdag and then some of the big ministries, and it was um pretty interesting being like 27 and then uh getting becoming account manager for these big ministries. And this was also during the time when the um not the headquarters would say the everything was moving to Berlin, uh the capital. The capital moved from Bonn, yeah, yeah, where Deutsche Telekom everything was like uh in the five kilometer radius, uh the whole government, everything, Deutsche Telekom, Deutsche Post was also there, and then the the government moved to Berlin, and then the the question was how should they work together between Bonn and Berlin? There was a Darten Highway, there was a so um uh Darten Highway, exactly. Darten Highway between Bonn and Berlin, so that they could collaborate. But I mean the only thing they did was sending faxes to each other, so it was not uh not so advanced technology by that.

Henrik Göthberg 15:55

Darton Highway, uh we should look at it in the era of 90s, yeah, yeah.

Stephan Erne 16:01

Uh so but it's of course uh really fun. And then I actually um I became the the assistant to the CEO of Deutsche Telekom, which was extremely interesting because I mean this was a huge change in the whole industry, and the company T-Mobile was created as a subsidiary to build up the mobile network. T online for the online network was the time when uh a voice stream was bought, uh which then became T Mobile US. Uh and they also bought uh the IT subsidiary of Daimler Chrysler, which is what's called Debbie System House, like 30,000 IT people. And it was even when Daimler and Chrysler were together, so it was the IT subsidiary for the whole thing. And that was bought from Deutsche Telekom to create T systems, which what is now T-Systems, uh which and ARDS as well. Uh so which overnight became like one of the biggest IT or ICT players in in Europe because you had so much captive business. Exciting times. Yeah, it was really fun. And um, and and then I actually also moved over from from being the assistant to the CEO to T-Systems and started building up this company. And there were like four different divisions from Deutsche Telekom moved in there. There was uh this Teleco, the IT people coming together, so it was uh really interesting to see ICT convergence. Exactly, it was really the the the big bet on ICT convergence, um, with a huge captive business because I mean all companies in Germany were of course customers of Deutsche Telekom, because there was no alternative. Um, and the idea was of course to to provide IT services to them now as well, instead of only a telecom service. So it was a big uh big ambition, uh very strategic move. Um but but then of course things happened like RDSL. Yeah, I mean, uh so it was was very bumpy-wide, I would say, even if it was uh strategically a good idea, but it was was not that easy, and that it's still ups and downs.

Henrik Göthberg 18:13

And then also in the 90s and 2000s, there were so many there was much less standardization on protocols. I mean, like I you can go back, and even now we have LTE as the main protocol in in radio, but um, but if you go back, if you go back to that area, it it is so many things that is coming from different. So when you're converging communication and IT, the the and this is immature at this point in time.

Stephan Erne 18:40

And the internet world was coming up, and then you had things like WhatsApp popping up, I think. Uh during one year, all the telecom companies lost $13 billion of revenue because people stopped sending SMS.

Henrik Göthberg 18:54

Yeah, exactly. Uh so I mean there was a huge uh shift um in the industry and and and um people if you haven't been in telco and understand uh you know average revenue per user style in and and how it works, the whole era around the end of the 90s and the dot-com bubble. It has so many different applications here, and behind it all, the telco sits there and they're looking at their business model from one day to another and changing.

Telco Lessons On Industry Disruption

Stephan Erne 19:22

But still having the customers, of course. So I was and then but then once again, what's that popping up? And and then the IP business, I mean the whole IP revolution and so on. So was was really uh really cool. And then you and then I moved to Ericsson, yeah, right. Because then we moved uh also the family up here in Sweden and then I so Ericsson was your ticket into Sweden, back to Sweden.

Henrik Göthberg 19:42

Exactly. Yes.

Stephan Erne 19:43

Actually, I would say my wife was the ticket into Sweden, but I'm like, this is what did you do in Ericsson? I joined a sales marketing organization and was uh promoting mobile broadband. That was some of the first things because it was when uh all the the the countries um were giving out the 3G licenses and trying to convince mobile operators to invest into this new technology, but also especially change their business model from taking getting paid per minute to move over to the data back edges, yeah, which was not that easy. So we were actually traveling around the whole world trying to explain with different business models and creating the market for mobile broadband. Yeah, that was when all these cool stickers were uh were created for mobile broadband, which nowadays people wonder what what was that? But it was a big, big move uh from a consumer side because there was there were no devices, there was when the iPhone popped up and stuff like that. There was no real push from the consumer side, there was not really no way to make money out of it, but still they should invest into mobile broadband technology.

Henrik Göthberg 20:50

So, which part of Ericsson is this? I have my I have friends that work in the Ericsson Enterprise at the time and stuff like that. So is this B2C or B2B or the whole thing more for operators?

Stephan Erne 20:59

More yeah, it was towards operators. So there was a group function, it was called group function sales and marketing. Yeah. Probably still exists, and this was like the the support to all the countries in helping them convincing their operators to invest into mobile broadband technology to really shift the product category. Exactly. Yeah, and that's why we also created something called category marketing because we realized it's not a product itself. You have to create a category for mobile broadband. And that's when we created this category marketing, had some stories, examples from different countries. So it was uh extremely international, um, really cool. Uh and also interesting to see, of course, how this company, Ericsson, works. And then I moved uh over to something called the network society. I don't know if well, people use the word Ericsson remember that because there was the idea to explain how actually, after this infrastructure was in place, how does digitalization change different industries? So we're looking intensively into all kinds of industries, transportation,

Henrik Göthberg 22:00

At the evangelism arm of uh Ericsson, where they're basically they're having all their core, you know, radio networks, all that, and something is new is happening, but we need to start evangelizing with the customer to understand you can't look at what we are doing from your old lens. You need to you need to start understanding the scenarios of lens.

Stephan Erne 22:21

This was uh then part of this uh was I think was brand marketing. The official title was part of brand marketing. This is evangelism. And we were we were actually we had the we had business cards with the name Network Society Evangelist. Yeah, so I remember traveling around and and uh explaining, and that was not only operators, then it also moved into industries. So we would talk to mining companies, whatever.

Henrik Göthberg 22:43

In order for them to connect and make that story into their comfort zone, what does it mean for mining?

Why Banking Feels Like Time Travel

Anders Arpteg 22:50

You know, and then you moved into Hundersbanker somehow. Exactly. When you first get into Hundersbanking.

Stephan Erne 22:54

Because when I after I did this evangelism for a couple of years, I realized now actually I want to be part of it and I want to drive this transformation in a company. So I was looking for all these. And that was the time when these chief digital officer roles popped up because companies realized, well, we have to do something, but we don't know what and how. And then they took in, and then Handelsbank actually also took, they wanted somebody from the outside. So they want somebody who has been through this change before this transformation. So they um um they looked for somebody, and I think, yeah. And and and I think that also was cool because uh when I when I then joined the banking industry, it felt a little like uh time travel 10, 20 years back, yeah. Um from compared to the telco industry. So I I think I really could uh could take a lot of the learnings and experiences uh from the telco industry into banking and and really drive the transformation agenda than that at uh Handels Bank where I've been uh yeah for nine years, eight, nine years, the last eight, nine years.

Anders Arpteg 23:58

I'd love to hear more about the the work at Handels Banking as well and the banking sector in general, perhaps. Uh so that would be super fun. But you you also left there what uh recently, and and what's your main focus today?

Stephan Erne 24:11

Well, my my focus is uh to be honest, still the focus that I I would probably say I had all my life. I I want to drive this this change. I really call it um like technology-driven business transformation. And because what I think happened during the past years, there comes a new technology, it can be mobile, can be internet, can be AI now, and then uh it usually there's a big fuss about the technology, but at the end of the day, it creates some change on the business side, yeah. Business transformation. And you have to adopt your processes, you have to change your go-to-market, you maybe create totally new businesses, you can do a lot of optimization. And uh what I really like is being in this in between those two two worlds, because if you really want to create an impact and want to have a positive impact on your customers, on your employees, on your business model, or you have to make sure that this technology is not just popped in and invested. You have to you have to change a lot on the business side, and that's why I call this technology-driven business transformation.

Anders Arpteg 25:18

And perhaps that's a great segue into the first topic here, really. You know, what is really, you know, what is the problem with doing a proper business transformation? And and especially if it comes to AI, of course, is moving so fast from a technological point of view that adoption, of course, is lagging behind them. So perhaps you can elaborate a bit more. What's your new view, and perhaps base it a bit on Hannes Banken and what kind of challenges and opportunities you saw there to still, you know, to find opportunities from the tech coming.

Stephan Erne 25:50

Yeah. And I mean, if we look at the banking sector as I'm coming from from that uh angle, um, I think you have to understand what what are the driving forces and how does it really work and what makes people awake or asleep at night or not. And if you're in a business that has been extremely successful for the last hundred years, it's like an oligopole. Everybody's making tenth of millions or billions of EBIT every year. Um you have some kind of a pretty good uh competition. You your customers are happy with the digital services, they're not running away. Um you have a lot of regulations. I mean, usually I would say banks there, they're spending 80, 80% of the IT budget on keeping the lights on and and uh keeping up with the regulations. So you have a you have a machinery that is um obviously working very well. Uh so I mean there are not a lot of outages. If something happens, you really are on the first page of every newspaper. Uh you get some fines from the uh IFLAN. So um that creates, of course, some kind of culture where play not to lose is probably the the name of the game.

Henrik Göthberg 27:06

And and this is nothing we could we should not shy away from that fact, but we need to navigate that.

Stephan Erne 27:14

So yeah, and that's why I'm I'm not saying it to blame it, Aja, but I think like this. And once again, I mean the alternative that you have like uh hundred different cool banks who are popping up like crazy and they're taking your money and you cannot find it anymore. Is that better?

Henrik Göthberg 27:34

No.

Stephan Erne 27:34

So it's not, yeah. So I mean, from a society perspective, it's of course it's not that bad. You have some stable players uh which create this playing level field. And I would say still, when of compared, I mean, then compared to other countries, of course, you could say, I mean, the Swedish banking market is maybe not the most innovative, but uh mobile bank idea, swish and stuff like that. Other companies they don't understand how that works.

Henrik Göthberg 27:59

Now, Swish is one of those things which is amazing to most foreigners.

Stephan Erne 28:03

So yeah, and but it's I would say especially bank idea. Yeah, bank idea as well. Yeah, that you can log in, that everybody knows who you are, you can log into any kind of service. In Germany, it's a mess. There are different different solutions. And this is something we also have to acknowledge because I mean that the the banks agreed on creating this infrastructure and then the all the innovation that was uh possible on top, not only for the banks today, but for actually for everybody.

Henrik Göthberg 28:30

Everybody who's using bank eating.

Stephan Erne 28:32

So um I would say sometimes the the banks don't get enough credit for what kind of infrastructure for digitalization they have created.

Anders Arpteg 28:40

Yeah. Good point. I mean, interesting. And still, if you if you think about you know doing the business transformation as you want to do based on the tech as well, how how did it go about that when you started you know for a long while back now in Hannes Banking to actually do find value from tech? Can you just go through perhaps you know how how did you go about that and what were the main challenges, perhaps?

Stephan Erne 29:03

And I would say that the the good I am I really and I love my time at Hans Bank, really. And that was it was definitely the best time if my working life uh I had. And uh I think one of the reasons was that when I joined the bank, everybody knew that we have to change. Yeah, so I I didn't have to convince people that hello, wake up, uh, something is happening there. It was everybody, almost everybody I would say, uh, realized shit, things are happening out there. We have to do something.

Anders Arpteg 29:33

What was the main kind of tech change you wanted to do at that time? Was it more the uh more the internet, the the the mobile, the the cloud?

Stephan Erne 29:42

What were the main and that's interesting your question because by that when I joined Handelsbanken, 98% of all customer interactions were in digital channels. Really? 98%, but nobody knew it. Everybody still thought all the business is done in the in the branches. Yeah, and to be honest, the the real business is of course still done in the branches. So it's not that you're making a lot of new business in the digital channels, but customers experience your services 98% of times in the digital channels. So that was one of the first things I was popping up. And so I told people, well, well, it it's not that we're totally behind, it's not uh that customers don't and don't appreciate in these um SQE and qualität index uh surveys, Handelsbanken also popped up as the the best digital services. So Hannelsbanken app had the highest ranking in those services.

Henrik Göthberg 30:35

And but I think what what is interesting is that people were thinking about okay, all the contracts were signed in in the branch, it's uh so they didn't realize that the whole customer journey was 98% digital and there's just one moment uh physical.

Stephan Erne 30:52

So used that yeah, but uh to be honest, the I think the word customer journey nobody heard at the bank before. So it was not it was not a planned customer journey to be honest. It was more like that was created for transactional stuff, yeah.

Henrik Göthberg 31:06

So at this point in time, we didn't think about this in this way, but it simply used materialized in that way.

Stephan Erne 31:14

It just was like that, yeah, because of course it was much better to let people pay their bills in the app than letting them come to the to the uh to the branches. But it was, I wouldn't say that was a planned customer journey that started at one channel and then uh you had a multi-channel approach and you were able to fulfill it somewhere else. Um, it was more like and and that's what once again, uh I think it was pretty astonishing because I I could not find a strategy behind it, but still it was in place. Interesting.

Henrik Göthberg 31:44

It was just something that evolved uh uh from steps step choices.

Stephan Erne 31:49

But I think that the main driver, coming back to your question, how do you do you drive this now? Was that um I think a lot of people I wouldn't say they were afraid, but a little had a bad feeling about this branch-focused business model. How are we able to survive in the digital world?

Henrik Göthberg 32:07

Right.

Stephan Erne 32:07

And I think this was one of the main questions uh for a lot of people internally who, of course, realize that uh a lot of things are happening, but of course, also for investors. So um, that was actually one of the first things I did. I was helping investor relations a lot to create a story and explain how this digitalization is strengthening Handelsbanken's business model and not disrupting or killing it.

Henrik Göthberg 32:32

And this and maybe there's a backstory here that you maybe should fill us in on this. Because if if you're in Sweden and know a little bit about the banking sector, Handelsbanken is a phenomenon, how they have driven something uh in a way strongly centralized in terms of some of the tech and stuff like that, but then with an extreme empowered autonomy in terms of being the office of a regional uh place, which is been they've been written books about this topic, so which is quite different to how the you know Nodias or the sweat banks or this SCPS work. So, could you just give that because what you're saying now that is that has was the strength of this empowerment, this extreme empowerment. How can that work? And you did it so well, and all of a sudden now, what how what happens with that empowered uh office or regional, you know, you call it branch manager, I guess, or exactly but he was the boss, he was the you know, but the branch is the bank.

Stephan Erne 33:32

That was one of the things the branches the bank.

Henrik Göthberg 33:34

And this is something that is deeply ingrained in the Handelsbanken culture, and all of a sudden now that core fundamental competitive advantage or philosophy, how the hell does that work in in digital? And that you know, that's just my backstory to how big this topic is. Did I get it right?

Stephan Erne 33:51

That's from and exactly like you said, and the the the big name, the big word is of course decentralization.

Henrik Göthberg 33:56

Yeah.

Stephan Erne 33:57

Decentralization is one of the cornerstones of Handelsbanken, which means the branch manager is deciding, the branch should actually decide on everything. And I think there were a couple of CEOs who said the most important person in the branch is the branch in the company. The way we build the company through the branch manager. And if you see, because Handelsbank built up 200 uh offices in the UK during the last years. The way they did it, they were looking for the right person.

Henrik Göthberg 34:24

If they found the CEO of the branch.

Trust Building Beats Hit And Run

Stephan Erne 34:26

The find the the right person, then they then they started a branch, and then this person was building it up. So in this decentralized model, and that's why, of course, the people who have who understood how the bank works, they even wanted even more. How does this then work in the decentralized uh in the digital world? And uh and there were, I mean, because there was no strategy, there was there wasn't even there wasn't at Hartle Parkington, there is no strategy department. And there was when I joined, there was no strategy department, there was no marketing department, um, there was no segmentation model because each branch manager segments its customer. That's called it's called this uh church tower principle. Yeah, you build a branch, and then what what you see from the church tower, this is your area, and you know everybody, you know the people. It's it's very much built on personal relationship and uh and knowing your customers, and then you decide how to handle your customers. And based on that, you have the total PL for those people, you know how many people to hire, you know what kind of customers to take in. You have the risk on your own. There is not a lot of back office functions.

Anders Arpteg 35:34

How do you start to digitalize them in that kind of company?

Stephan Erne 35:38

Yeah. Um and once again, the good thing is that everybody was somehow a little shaken about what will happen to us going forward. So I didn't have to convince them or to drag them into the discussion. They were more uh they were really looking forward for somebody who's who's driving this discussion. And um what I then did, and I had uh I think I um uh what's like this. I think if you if you join a company like this, you have two choices. Either you go for a big hit and run, fast hit and run. You go in, you make a big fuss. You're screaming, you're you're making everybody afraid, you come with some cool solutions, and then but you and I've seen some of CDOs like this. Yeah, in one three years, tenure max. So yeah, after one, two years, you're out again. Because people, of course, it's pretty nice. Some people think you're uh you're the hero, but usually you don't have a huge impact. So what I did instead is I was building, trying to build trust with as much many as possible.

Henrik Göthberg 36:40

Alliances, alliances.

Stephan Erne 36:40

So I was I was uh in the first year, I was one of the top three travelers in the company because I was out there talking to anybody who wanted to talk to me. So I made some introduction, of course.

Henrik Göthberg 36:53

We uh said this before we created a new channel in the company talking about if we say that the most important person is a branch manager in the in this model, then you need to talk to every single branch manager in one way to have a feeling for where they are. That's your what's talking about.

Stephan Erne 37:10

And there were, of course, regional meetings with all the branch managers, and and um once again, the the good thing is people were really interested in hearing from me. I had a good story that I came with from Ericsson. Yeah, so to be honest, I could reuse a lot of the stuff that we did at Ericsson because this evangelism that I did there for the last year was. So um that's why I think it was a really, from my perspective, perfect match for me. But I would also uh say that the bank really needed someone like me by that time to was explaining what is actually happening out there, and that that even went to towards customers. So I was invited not only to talk to the branch managers, once they heard me, they invited me to talk to their customers so that I could explain to the customers where this is going. Where what is happening, where is the bank going? And usually have this what is what is digitalization, what is uh happening in the um wider industry, what is happening in the banking industry, and what what examples of what we do at Hans Banken?

Anders Arpteg 38:08

So that was an important way, the way it could actually drive change, yes, in that way. And I think it bears a lot of resemblance to the AI transformation that we have to do today as well. Would you agree?

Stephan Erne 38:19

Exactly, yeah.

Anders Arpteg 38:21

So, okay, so if you try to summarize a bit, you know, how really made did you make the the digitalization like transformation happening in Handelsbank in that time? It was not a hit and run, right?

Stephan Erne 38:31

It was building the trust, meeting the people, educating, a lot of education, yeah, a lot of education, but in a way that people can can use it, and not I'm coming as a guy, and afterwards you think, oh shit, I have no idea. No, you have to if people left the meeting and they've learned something, and they go home and say, No, I've understood more than that. I think this is the way how you build trust, how you educate in an organization, because then they also want to hear more. So I will I was hardly possible to to be at all the places where we people wanted to talk to me. And once you have this trust, I think then and and do this via this, all these conversations. I was of course also also able to find out what what is really the burn, what are the real problems? What are the real problems? And then I was of course addressing them and not some cool other stuff because that was the time when all the other banks uh came up with out with their chatbots. All the other bank had chatbots.

Henrik Göthberg 39:29

It was easy to go on that hype train, but your branch managers were pointing at something. What?

Stephan Erne 39:34

So, and and then of course, also the the the top people said This is pre-LLM chatbots. Pre-LLM chatbots. So they said, Stefan, we need a chatbot. Yeah, yeah. And I had I had a team who would have been able to do this, and we were looking into it, we were testing the other chatbots, and we came back and said, Of course we can do that, but to be honest, we don't think that we should decrease our customer satisfaction and create a bad chatbot that is is only bad for the company. And to be honest, and from my personal perspective, it's also bad for my path going forward, because if people see that there's a bad technical solution, that of course is not pushing, it's not it will not help me going forward to come with the next solution. So what we did instead, but of course, uh, that was the first stages of AI. And but of course, we said we we cannot we cannot do nothing, yeah, but then we build actually an AI uh based on natural language processing, who helped us in an area where we really had a problem, totally unvisible, uh no, but but a very good story internally to show and and prove that this technology is helping us. We did it in a way that um there was not a huge risk. Uh and consumers were not affected by this. By that time, I would say pretty bad solutions. Um, but it created a cool um proof point for us internally that we were able to deliver something with this technology uh and and also opened the door then for for for the next step of solutions.

Anders Arpteg 41:03

So speaking, if we just try to make some comparisons here, uh for the digitalization transformation and the AI transformation, um especially Hundersbank and you had a very decentralized approach, strong spokes, if you call it that, in the branch offices that you did have. Was it also that you tried to create some more centralized kind of governance of that? Or or how did you try to manage this kind of decentralized organization that you had?

Stephan Erne 41:30

Exactly. And and this is of course where um because one of the big things that I think happened um during my time was the the realization that um we weren't we are not able to deliver this handels bunken approach and feeling to to most of the customers, because the branches were of course not able to take care of all the customers in that way and meet them in this personal way.

Henrik Göthberg 41:57

So you get to the point uh that where the decentralized way where you can go complete your on your own way physically, you need to rethink how that should work when you have a digital because I mean like the alternative that each branch builds their own app that would be completely a fragmented digital experience. Yeah. So that's a tricky one, yeah.

CRM As A Digital Backbone

Stephan Erne 42:20

That's a tricky one, and this is of course also but but I think it was not that hard to to realize that we need some kind of system support, just call it system or whatever IT support. Uh IT system that helps our people, our customer-facing people, to deliver their Handelsbanken experience to more customers. And that opened the door for the whole CRM discussion because Hans Bank had no CRM system. Still, we had 98% interactions in the digital channels. We had a very good uh Hansbanken um telephone service, we had the branches, so we definitely had all these different um channels, but no system who supported it. And at the end of the day, the branch manager who should be the branches the bank responsible for all of these interactions, he was not able to do that. And that that that started based on all these decentralized uh responsibilities, I think the need and and even the branch manager said help us, give us something. It was not that they said we have to we want to build everything on our own, but we need we need uh some kind of help to handle.

Traditional ML Versus Generative AI

Anders Arpteg 43:36

I think it's funny because it uh it I think also is related to a lot of people in the AI and data space, you know, thinking about data mesh. And I think they a lot a lot of them are misinterpreting that and it's completely decentralizing too much without a centralized kind of coordination that you do need to have in some kind of hub that is supporting all the uh different spokes or the parts that you have in the organization. So it's fun to hear, you know, how you come from. That perspective as well. But perhaps if we could move to a bit more of an AI topic as well. And we started to speak about chatbot, but it's almost like you know, without AI, I guess, in that time, and it's more manually programmed, intent-based kind of chatbots at that time. But how was the more AI-related journey? When did you actually start on that?

Henrik Göthberg 44:24

If we if you jump to the later years, I mean like the last couple of years now. I mean like if we if we used pre and post Chat BPT launch, because it's something that shifts here. I mean, like we felt it so strongly on this pod as well, right? Yeah. So if we look at the last couple of years from 21, 22, I mean like when did it start?

Anders Arpteg 44:44

When did the AI start?

Stephan Erne 44:46

And and this is um when when you I think when you're looking at Gartner studies or whatever, what's the AI usage in different industries? Pre-CHPT, the financial services industry was also one of the biggest users of AI.

Henrik Göthberg 45:03

Which type of machine learning then? Uh fraud, of course, and AML. Yeah. AML.

Stephan Erne 45:08

Algorithmic trading.

Henrik Göthberg 45:10

Algorithmic trading. So we had it with model we now call traditional machine learning.

Stephan Erne 45:14

Yeah. So in traditional AI, at least from what I've seen, the banks They were on it. They were on it. Yeah.

Henrik Göthberg 45:22

And because they had a huge amount of data, they had a lot of um uh learning data, they have a lot of transactions, they have a lot of uh it's a data-driven business, so there is there's data to be made to feed the machine learning with. Exactly.

Stephan Erne 45:37

So the for in the old world there was uh and and also a pretty good um steering model um making it uh compliant. Um how to handle this old-fashioned AI. Um when I look at the status right now, um and I I've just also was on a on a on a conference on on Monday where one of the the the former uh Gartner guys was showing the adoption of generative AI in different industries. All the banks are the way behind. And and that's why when you ask me what is how does the how do you drive the the post chat GPT AI journey in a bank, uh that's something totally different than the pre-T-GPT because this is the black box, nobody really knows how it works, AI stuff, and uh in a deterministic regulatory-driven world where the whole steering model is to steer p that comes regulation comes in. Um you you put this into um direct directives, and the directives are steering the people, and then you make all kinds of education about these directives, you make some testing.

Henrik Göthberg 46:59

Are you following these rules and regular it's it's an extremely deterministic philosophy the whole way through that is ingrained in the whole model for it to work safely, and now we're entering into stochastic probabilistic technology that we then need to we need to realize what is the strength and opportunity with that, but we need to harness it. And I think you I think what you said now is so true and refreshing because sometimes I think we are bullshitting about this when we're talking at conferences and doing oh, we're doing this and this and this. Like, like actually, you were pretty fucking good at machine learning when it came to risk and fraud and traditional machine learning, but you have not find your DNA, how to do that practically with systems that are going as an antithesis to how you steered. Yeah, so it's not about that we can't use the technology, but we need to harness the technology to work in this setting. Is that a fair summary? Um I'm putting it.

AI Governance And Human In Loop

Stephan Erne 47:57

Absolutely, yes. Really good words to to to to um to express what what I was trying to say. And this is why I think this this AI governance topic is probably the big shit thing in banks. Yes. Because it's and and from my perspective, and once again, I'm I'm I'm talking to all kinds of industries, and from my perspective, it's a it's a very different thing, AI governance in banking versus AI governance in uh retail. Because in banking, nothing is allowed to go wrong. And the regulations, and of course, things go wrong and people make mistakes, whatever, but the whole system is built around how do I handle people, their decisions, their responsibilities, how do we have a four-eye principle to make sure that your wrong decision doesn't get out to the customer. And this model is so well played, yeah, because I mean that the system works pretty well, but the risk of something going wrong are of course also extreme. I mean, there have been we have been through banking crisis and so on. So the risk appetite to let's to let something in, which you nobody wouldn't say nobody, but you cannot really explain it. Uh, it's very hard to set boundaries around it, are so high that uh I think that's one of the big reasons why the adoption of generative AI in the banking sector is not really high.

Anders Arpteg 49:26

The bank sector has been a bit falling behind, perhaps. I would say a bit.

Stephan Erne 49:30

I mean, I just bank of America is not allowing any code uh system tools.

Anders Arpteg 49:35

Not even for develop development purposes.

Stephan Erne 49:37

Oh no.

Henrik Göthberg 49:38

Oh so and and the interesting thing, I mean, like you had Matthias Fros on the panel, he's a good friend, and he had he has a new role in uh as AI adoption in Odea. And I I brought him over together with me. We we were must we were hosting Derridax was hosting a master class within the CDO IQ conference. Um and he's he talks about this a little bit like when we do this stuff, we need you we need to embrace compliance. So we can't do an innovation in a bank without embracing AI and doing things by design, which which is to me, this is so obvious, but I think this is a big gap in understanding that we are using uh probabilistic and stochastic technologies or techniques because it has some major advantages, but we still need to harness it in and put the reins on it.

Anders Arpteg 50:31

But how do we do that? How do we get around you know now you know finding the value that is still there for generative AI, even in a sector like the banking industry?

Stephan Erne 50:40

And I think to be honest, the it's probably maybe sounds a little boring, but number one thing is to keep a human in the loop. And and not not maybe because it's better, but because it fulfills this old steering model and the old governance model. Because the old governance model is is based on steering and governing people.

Anders Arpteg 51:02

And even though if those people don't really do a lot, you still can point at a person who still should you know be able to review it and have the um the responsibility for for you know what's being done.

Stephan Erne 51:14

Exactly.

Anders Arpteg 51:15

Yeah.

Stephan Erne 51:16

One reason why this doesn't really work is that I would say a lot of the leaders don't trust people anymore if they get too much help.

Anders Arpteg 51:24

Too much help.

Stephan Erne 51:26

Yeah, I mean I mean if you have if you're only the human in the loop to push the button and say this is okay, this is okay, this is okay. If you push this button a thousand times a day, you will also push it the thousand first time, even if it's wrong.

Speaker 6 51:38

Yeah.

Stephan Erne 51:38

So just to have this human in the loop idea, you have to make a little smarter so that the humans who are in the loop don't get so dumb that they don't take the responsibilities.

Henrik Göthberg 51:47

But this is once again the question of nuance and understanding on a deeper level, how we design and how we then do security compliance and uh this by design. And I think this is the challenge now, because we can talk about the what about what what what is this, what is that. We need to design this. And I think that's what Matthias is doing, and that's what we're talking about here. How does this rewire our connection to compliance? How does this rewire our component? We have all these different things, how it works used to work in the bank. Yeah, they are still there and they're still needing to do their job, but we now need to flip it and re-engineer how they do their job and how that is influencing how we build systems. So it's not two worlds, but it's it's rewiring Matthias says. I like that uh rhetoric. What do you think about that?

Stephan Erne 52:36

And I think it's it's something that has to be done pretty new because I mean these cool slogans like uh compliance by design. I think companies have found a way how to do this in the old world, but there is no I haven't I have I haven't seen a good way how compliance by design works in the stochastic world.

Henrik Göthberg 52:56

No, I I I then you need to talk to me and you need to you need to talk to Anders because you need to look carefully at how SpaceX is doing this kind of stuff, right? So so of course it's doable, but it it it then it's it's you need to re-engineer the core operating model around how you do these things, but it's completely doable. I mean, like we can go into the SpaceX example, who is sort of super agile in one of the most compliance-driven I mean, like they're not.

Anders Arpteg 53:23

What do you think is interesting from the culture point of view in the bank sector, you know, they are very like governance-minded, if you call it that. And then you know, how do you go about doing that in that sector? I think I think it's a super interesting topic. Of course, we can have technical solutions for this, which ASEX and others are doing really well, uh, and also from a like a methodology point of view, but still how how should we go about doing that? Um I don't, of course, I would like to hear your thoughts here. You know, you you want to be use be able to use a technique like generative AI, even even though it's super hard to really understand. I would actually argue that even traditional machine learning is super hard to understand and easily misinterpreted. But still, how do we make the backing sector do that? Do you have any thoughts about that? You know, what what what could make them trust AI for development purposes or even in the products in the chatbots and whatnot?

Winning Compliance Over With Usefulness

Stephan Erne 54:26

And I think number one is is of course to to to make them lean into the topic. Yeah, because I think in a lot of uh compliance organizations they just hands off, they go away, and they say they just find reasons why you're not allowed to do it. So it comes back to a little what I said at the beginning. How how do you drive this change? You have to create the trust in this department as well, and maybe help them solve their problems with GenF AI. Their problem asks the what is Yeah, then the problem is how do we know what kind of regulation is coming? Because I mean they they're living in their own, they have their own challenges.

Henrik Göthberg 55:05

Which compliance which regulation do they need to comply with, and which regulation fits in which part of the bank, etc.

Stephan Erne 55:11

And how I mean, yeah, and I mean there are some places, and I'm actually there's a pretty cool guy um in the uh department in Sweden who was very open and he he was out there scanning and seeing what kind of gender AI solutions are there that help us do our business better. And how can we scan? For example, there's a new regulation coming thus. What do we have to change in our rules internally in the bank? And just to get a first scanning and and and and and and understand. Because I mean, reading all that stuff from the EU, that's not uh the easiest thing to do. So I think coming back to your question, how do you do this? I think the f one of the first steps could be to help to help the compliance people to handle their business or their task in a better way and make sure if if they can increase their accuracy and their quality with the help of Geneva, right?

Product Thinking Rewires Compliance

Henrik Göthberg 56:04

Yeah, but but I I have a different view on this, quite quite different, because I think that you are trying to find a solution within an organizational setup that was done for a different purpose in an analog world. And I I think that first you need to understand where the blind spot is, the what, the problem, and then you are realizing okay, that's the what is the problem, how do we solve it? This is where I look at this completely different to many people. I don't think you can solve the how in the old org template.

Stephan Erne 56:42

Yeah.

Henrik Göthberg 56:43

I think I think the how then brings you to a different understanding of how you organize these uh resources. Ultimately, you get to a point when when you do compliance by design, it means it's part of the product team to do compliance. And it means the role and how you execute on compliance is not done by one uh team that has tech competence, and then you have another team by business competence. So I basically think in order to solve it properly, it cannot be done as long as you have division of layer as your core uh organizational model. I think that's the main problem.

Stephan Erne 57:21

And I like it that you disagree. But but yeah, yeah, let's no, and and I and I think your approach in in an ordinary company where they dare to take the risk and make this organizational change is probably right. But in a bank, where I mean the compliance organization will not go away that fast. There's nobody coming from the top and saying, No, we have to reorganize and have the shovel them together. So I think that's why I think you have to to get them on board on that journey. Then if the end game maybe is what you said, that could be right. But if you if you have those people against you, you will not move a millimeter.

Henrik Göthberg 58:00

No, I no, so let's be sharp here, because I fully agree with you on this one. And this is a little bit like how I understand rewiring short term and then evolving into a new organism that we don't know how that looks like. So so I fully agree with you. You need to get them on board. But to get them on board is actually to them for unlearn how they looked at compliance in the old way and understanding how they do compliance in a new way. So this is a massive evangelism and education journey, and this is why it's not a flick of a switch, but rather a co-creation on what the new operating model should be like. So no one can draw this map. If McKinsey says they can draw this map now, fuck that's bullshit. But it's this co-creation on the problem and having a common understanding for how did we do compliance in the past and how should we now figure this out? I think I can combine the two thinkings here.

Anders Arpteg 58:53

Let me give a because I'm not sure. I respect your own. I think you're both right in some ways, and and I I can see both points here. One for one, I if you think of the compliance team as the hub for compliance, then if we don't get them on board and they don't understand the value of how we can actually have value, uh AI that really really helps them their purposes, it will be dead anyway. So we certainly need to get them on board, and I think that's a very good point I think you're making. But then if we take the developer team and someone else, and and traditionally you have this kind of competition between the tech and the compliance team and other like private team and whatnot, and that is hurtful as well. And we should really make the development team then have some at least some spokes, some some people that is actually responsible and actually part of the team working together with the compliance team, and in that way, yeah, having this kind of collaboration, right? Absolutely. If you do these two together, yeah, then perhaps you know you can have some kind of course.

Henrik Göthberg 59:53

And I think there's so I talked about the end game, and but there's a journey to the end game which is starting where you are. Yeah, I think that then is.

Stephan Erne 1:00:00

And I think we probably agree on the on the what has to happen. And um and I think one way of of of aligning and bringing them together and change this way of of addressing the whole compliance game is is um bringing those people together, and and because I think the compliance function itself will have a pretty hard, it's not that easy to unlearn what you've been doing for the last hundreds of years, obviously, in a way. So you you you need to bring along and you know the tech people to help them help them do this.

Henrik Göthberg 1:00:35

Let me try another one because I think we we are at one of the core ideas. If you want to talk about AI in banking and how to do that, to ignore the compliance argument and discussion is use bullshit because then you're not then it's innovation theater. And another thing that I think is what I'm talking about, it's not so much it's one thing is about how the hub and spoke thing works, but sometimes we don't speak enough about we are growing up in a bank or even in manufacturing, in what I would I would define as a process-oriented operating model, and where we are handing over in different swimming lanes where we should do things. And the the core idea here is like you can have a hub and spoke and stick in the process-oriented thinking and steering and governance. And the problem with that is that now we're getting to the point where you need to have a product-centric understanding of workflow. So if I'm working in the you know, credit or whatever, like take one core workflow, you need to be able to go 360 degrees around the actual workflows, what's the data around that, what's the algorithms around that, what's the AI agents around that. And therefore, you know, you need to start looking at that almost like this is a workflow, but we need to have a product-centric understanding to govern this workflow. And all of a sudden, now you get to a point where compliance as a function is a setup to do advisory or looking into process. And all of a sudden now the hub and spoke is not only hub and spoke, but it's also okay, how would we work as a function in compliance when we are supporting product teams and when the the spoke is the product, so to speak, the workflow is the product. So you see what I'm saying with the product-centric versus process-centric, yeah. This is another mindset shift that I think is part of that onboarding. Yeah. What do you think about that?

Stephan Erne 1:02:31

First of all, I would say that the even the process maturity is not very high. Yeah, that's the problem.

Henrik Göthberg 1:02:37

Even there, we don't skip that game then. What do you mean with that? What do you mean it's not of course I no? How can you say the bank is not maturing?

Stephan Erne 1:02:46

I was asking, I was waiting for that question. What do you mean? Yeah, of course, there are some processes like that, the credit process. And I mean, there are some processes which are very also centralized steered. Because I mean, even in a bank like Hannesburg, which is extremely decentralized, there are some processes who are like IT security, um, credit process, they are extremely centralized steered, and nobody's stepping out of that. Yeah, yeah. But a lot of the other processes, um, I would say they are not really described. It's pretty unclear who is the process owner. There are no measurements around it. So it's more humanistic than we think. Yeah, it's much more and it's also and especially in a bank where people usually stay for for all their working life, yeah. It's very much based on relationships. So in instead of having those descriptions. This is a lot of tacit knowledge. Oh, interesting. Yeah, cool.

Speaker 1 1:03:48

It's time for AI news brought to you by AI8W Podcast.

Anders Arpteg 1:03:54

Cool. So we usually take a small break in the middle of the podcast to speak about some recent AI news happening. And uh we try to keep it short. We'll usually fail, but we'll see how it goes this time. Um, do you have any favorite?

Henrik Göthberg 1:04:09

Let's start with Stefan. Do you have any do you have any things you've you that you picked up on in this sort of news feed around AI last week?

Stephan Erne 1:04:18

Or for me it's it's more the um we talked about the sovereign sovereignty before. Yeah, what I think is of course pretty pretty cool and interesting is to see what's happening in Germany about AI investments. Um when uh Deutsche Telekom, I think, announced that they're gonna invest 10 billion euros in a new AI factory in Munich. And I think the way the week after, Lidl, yeah, the Schwarzkopf said we're gonna do 100 billion in an AI factory. Little is driving. I've seen this Lidl popping up as what? So uh I think this was uh one of because I'm coming from Deutsche Telekommerce, you know, so that was one of the the the funniest things I've seen.

Henrik Göthberg 1:04:56

But what is your angle? What's the angle? What's your when you're reading this with your knowledge about the the industry and what's your interpretation of these plays? What where are they leading?

Stephan Erne 1:05:07

Um I mean it it's a very much a political thing, to be honest. I think it's a very much political thing. And because I mean the the some ministers from from the government were part of this Deutsche Telekom staff and said, Oh, this is the biggest investment. Are we gonna become independent? Ba-blah bah. And then uh once again, a week later, Little did times 10, and somebody nobody was there. So I think it was it's more at the moment, I think it's it's big enough.

Anders Arpteg 1:05:32

It's feels a little bit sovereignty play, right?

Henrik Göthberg 1:05:35

The sovereignty play is something that works right now. Yeah, exactly.

Anders Arpteg 1:05:40

That's my my mine use. Do you think they will succeed? I mean, Lidl is still a big company, right?

Stephan Erne 1:05:45

Oh yeah. I mean they have a little huge captive business. I think that the good thing about Lil is and then they're gonna make this, they do this Amazon play again, because they have so much captive business that they will probably be able to fill a lot of this capacity with their own business. And They are also there, I think they made some good uh recruitments.

Henrik Göthberg 1:06:04

Um and and but so they I think they will be able to see so you consider this it's it's a huge number, but it's still a fairly serious play in terms of actually it makes sense for them almost from an internal perspective. Exactly. So that's the Amazon angle, right? You start building a platform and then you realize shit, I could sell this platform.

Anders Arpteg 1:06:22

But it's not only that, I think also that the big cloud providers otherwise in the US, the reason they become successful is that they started for 10 years to build like AI and IT infrastructure for their own purposes and then turn it publicly. This is actually one of the few companies I would say in Europe that can do that. So it's interesting. I hope that they actually do succeed. And it's a big number.

Henrik Göthberg 1:06:45

It's a big number, and it's an interesting the what it's the right way and the right purpose to start because that purpose has been proven before. That's what I mean.

Stephan Erne 1:06:55

And I mean the T-Systems, like I said before, came from the similar perspective. Yeah, it was the captive business for Deutsche Telekom, Deutsche Post, ERDS, Daimler Kreiser. So there was a huge captive business, and they said, Well, why don't we sell this to other companies? From my experience, it's easy on paper, but not that easy on the other.

Henrik Göthberg 1:07:19

Do you believe in it? Or do you okay? What's your oh my god, this is a bad idea. This is uh because I kind of what's your take on that?

Stephan Erne 1:07:28

I think that comes back on that. We had a small uh sovereignty discussion before because I think a lot of companies want to discuss it, but I'm not so sure. I think to be honest, my my personal view, as soon as there's a new president in the US, things will move back.

Henrik Göthberg 1:07:48

Yeah, I I I I don't I don't know. That's a good comment. Yeah, anyway. Anders, you or uh Goran, do you want to start?

Goran Cvetanovski 1:07:58

I think Anders can start.

Anders Arpteg 1:08:00

Then we can or maybe I can start. Yeah, useful. If you want to take GTC, I can take another one.

Goran Cvetanovski 1:08:05

Take the GTC. No, but you can take the GTC, I can uh I can just comment on it.

Nvidia GTC And Agent Operating Systems

Anders Arpteg 1:08:10

Because I want to take another one, but I can take it a GTC as well.

Goran Cvetanovski 1:08:12

But uh take a GTC and I can comment on it. Uh what I saw there is because I spent like a week. And then you take both. You went there, right? Yes, yeah, I was there with uh around like so now we're doing GTC.

Henrik Göthberg 1:08:24

What is GTC?

Goran Cvetanovski 1:08:25

Yeah, the GTC is Nvidia's uh annual uh what is called developer uh conference. Uh I mean it was developer, now it's not developer anymore.

Anders Arpteg 1:08:35

Super, super big conference super big.

Goran Cvetanovski 1:08:37

I think it was uh they thought they they were saying that they sold out around 120,000 tickets. It's amazing. Oh what a fucked up number. I was uh had the privilege to be in the the the stadium in uh SAP stadium that they have in San Jose when he was delivering this year's um a big starting like it's like 50s like one one of the big uh 40s. So it was not the full stadium because usually, like in a concert you divide it a little bit, but I think it's it was around two-thirds. So um in my calculation, between 20 to 25,000 people. The rock star was uh Jensen, right? There was no place where to drop a needle.

Henrik Göthberg 1:09:13

Uh this is this is like a YouTube rock concert as Jensen, right?

Goran Cvetanovski 1:09:18

People were queuing from the seven o'clock in the morning, uh, and he came around 11.

Henrik Göthberg 1:09:23

So these guys are rock stars now. That's what we can summarize.

Goran Cvetanovski 1:09:25

Yeah, yeah. So this is the that is the rock stars right now. So it was a huge actually build-up towards that. And um, before that, uh, you know, uh before his presentation, there was a very, very good, powerful uh panels from uh you know Anthropic and Clove and everybody else. So it was like a very big uh build-up towards there. I was part of a Swedish um uh actually a Nordic uh delegation that went to uh to um uh to uh GTC around 250 uh people in total. So we had like activities around the whole week around GTC as well. It was uh orchestrated by Nvidia and Dell uh together with Sweden as well. So it was great. It was a lot of uh great uh memories and a good business actually around that. So it was super cool. Uh the interesting thing about it it was that uh uh they were talking about I don't know, did you hear about the five-layer cake?

Anders Arpteg 1:10:21

No, I missed that.

Goran Cvetanovski 1:10:22

No, uh I'll show you. So the five-layer cake, it was basically uh something that came up, and there was also uh side uh gigs and uh uh and um events actually around this five-layer cake, which is basically the five layers of uh uh making AI operational, starting from energy to uh chips um um to um uh what you have. So it's basically uh on the first one you have the energy, the foundation, then you have the chips, you have the infrastructure, then you have the models, and then you have applications. And Nvidia is perceived basically as a chip company, right? But with the latest move that they have been building up for the past two years, and uh keep in mind uh Jensen doesn't do anything basically from today, he has been building up this for the past two years. Now they are operational in all five layers. That is the five layers of cake, right? So you have energy, uh, chips, infrastructure models, and applications. Uh, the interesting was like how the exhibition was built, because you had the five-layer cake plus the robotics, so it was around like 200 exhibitors there. You could find the biggest ones as well. Super interesting for me because some of these companies we have never had it in uh Nordic, so I had a chance to speak with them. And then he introduced like a couple of things, which is basically in the five-layer cake, of course. So he started with the energy, that was the Vera Rubin uh systems, which is like huge basically systems that you can stack a lot of uh this NVIDIA DGX and uh uh H100s and Blackwells and whatnot. I can speak more about that. Yes, you can do that, right? So then uh there was like uh there was new basically um offerings within the H100 and Blackville GPUs. Then in the infrastructure, you have the AI factory, so it was full with people that companies that are working with data centers and neo centers. That is the AI basically inference. Inference was a lot of discussion there. Uh, then you had the models, so they basically had like the the 120 billion parameter open model, which is called like a nemotron 3 super, I think that was you you can speak a little bit more in depth about that. And the the the the the build-up, of course, it was about this Nemo Claude, which is basically on top of the um which is uh on top of the the the claw part that we open claw that we were discussing quite a lot uh now in the past years. So the idea was actually how he presented like the renaissance of AI is here. My takeaway from that it is completely right, but we need to also have a renaissance in organizational and enterprise uh application of this because as he's showing it, the technology is there, the compute is there, there is an ecosystem uh to operate AI, but we need to what we are discussing today start actually implementing these things into place because the the Nemo Cloud that he's introducing is actually an open source uh model that can uh help organizations to safely uh responsibly operate uh uh um what is called like uh agentic AI or projects into uh into the organization. However, to what we are discussing today, is that simple or not? That is a different story. And this is actually where it becomes interesting. I think that we are in a point of time where it is more clear than ever that we need to get our shit together and start implementing these things for for real because the technology is there, people are making there was a lot of startups, and these startups are leaning just more and more forward, they're just going and running at it, and we are just sitting and uh putting in the state.

Henrik Göthberg 1:14:05

But it's it's popular point that the the data AI adoption industrialization company like Daradax, founded in 2019, is the only name in town. I mean, like the it's not a technology problem anymore. Even even to do it.

Anders Arpteg 1:14:21

I I think that's just too simplistic to say.

Henrik Göthberg 1:14:23

No, but uh no, the I I'm saying to get value out of AI with the technology that is around today.

Anders Arpteg 1:14:30

But I think there are a lot of shifts, and if I just were to give some views of the trend shifts, so to speak, that were given in GTC. I think there are a lot of shifts in terms of actually technology as well, which is interesting. And one, of course, is that it moves from the traditional discussion about oh, we have Opus or we have GPT or we have you know a new Gemini model to instead speak about systems. So moving from discussion about you know how is this model performing to actually having more encompassing systems around it. And of course, we're speaking about open claw here, and then how uh Nvidia has embraced that and built a system around that for enterprise purposes to actually do find value in an enterprise, to actually do add the additional you know functionality about authorization and logging and uh and supervision and security that you need to be able to use this kind of system in an enterprise setting. So but still you know, starting to speak about how AI is moving from just being the type of reactive chatbot to actually being a system that is proactive and always on as OpenClaw and Nemo Claw now that Nvidia has is of course a big thing. So he even speaks about you know open claw being an operating system for agents, um, which could be an interesting way to speak about it. And I think that's a big trend shift. Of course, building on an open core trend has been but that's a big shift, it's a shift it it is a system he's talking about.

Goran Cvetanovski 1:15:59

There was no discussion anymore about AI uh as uh one or another, it was actually an ecosystem, an operating model that you need to actually put.

Henrik Göthberg 1:16:07

But but this is a shift in rhetoric and how the industry is trying to sell shit. No, it's actually a shift how you use AI. I talked about AI compound system, Barclay. I mean, I had the system view. I had the system view. That's that's the wrong view.

Goran Cvetanovski 1:16:24

I think I must agree with Anders in this case because I think that the ecosystem that we were sticking there, in terms of uh of uh just basically by going to the exhibitors, you can see that right now it's everything coming together. Yes, you are correct. Uh you know, in the ways how how we could have implemented this, but you can see that the whole ecosystem was there. There was like uh data centers, AI centers, cooling, energy suppliers, uh utility companies. Everybody now is in the same rates. Before that, it was divided rates.

Henrik Göthberg 1:16:56

Yes, and now I get your point because my point of view, which is the enterprise point of view, how you navigate to safety uh scalable AI, has always been a system perspective. And we had a hard time because we needed to figure out how to pull the pieces together to create a system. And and what you are saying now is the industry, the industry was going in these like pockets, and now the industry is telling the system story and they're building things more as a system.

Anders Arpteg 1:17:25

We have technological technological support that we haven't had in the past.

Henrik Göthberg 1:17:29

No, you need you needed to be smart. No, you you fixed it, but you needed to stitch that yourself together.

Anders Arpteg 1:17:34

Of course, we still need to do the organizational change, that's always been there, but it's a big shift how now we can have these kind of proactive systems and they actually storing memory in a way with the souls and the skills and everything that they're gonna be able to do.

Henrik Göthberg 1:17:47

What I'm trying to get to, Anders, if I'm if I'm understanding you right or not. I think the smart people who understood this from the past has gone the systemic or system way, but it was a pain in the ass to stitch it all together. And now it becomes more. And now it becomes more doable.

Anders Arpteg 1:18:12

It's been it was actually surprisingly, you know. If you just take OpenClaw as an example, as Jensen said himself, this is the biggest open source project ever. It's never been a product that had this kind of increase in popularity ever, compared even to the Linux system or to uh Databricks or to Spark or whatever. I mean, it is a big change how we actually had a system that now can operate as an operating system for agents. That was a big revolution. So that it's been, you know, we wanted to do this for a long time, but we haven't been able to, and now it's suddenly it was a big shift. So I think we need to recognize the shift at least.

Henrik Göthberg 1:18:54

Because I I thought this was something that the best of the best were doing, but it was a pain in the ass to stitch it all together. But you are saying actually to some degrees we were missing components that was making it really hard.

Speaker 6 1:19:06

Yeah.

Henrik Göthberg 1:19:07

So this is what I'm is that what you said? I mean, like, are we yeah, yeah, because I was okay.

Goran Cvetanovski 1:19:12

Well I'm I'm with you on that. That I thought I didn't think I didn't think that uh the open claw moment was so big. But when uh you landed in San Jose, in Palo Alto, we were everywhere there, it's like uh you know in Silicon Valley and visiting uh that is the the one revolutionary moment for decades that have actually that is changing something. I didn't understand that, but when you landed there, you knew this is it. Everybody was talking about that.

Anders Arpteg 1:19:41

So everyone will copy this. I mean, uh now Google would do their own version of this, Amazon would do their own version of this, everyone will go in this direction. It's a new type of proactive always on AI system that will be copied everywhere. It's I I'm very sure about that.

Stephan Erne 1:19:57

Yeah, and I mean if if I'm um allowed to pop in here, please all the people that I'm talking to, they have no clue what's happening. Yeah, yeah, no, nothing. Who who knows about open clue in the enterprise market? Nobody knows how to, and that's why nobody's I agree to you. I think this was one a big moment in the industry, especially if you if you realize that this is one guy from Austria who created all that stuff. Yeah, yeah, exactly. So I mean even more impressive, you don't need to have billions. So and if this is possible for one guy from Austria to do, what will be possible in five years for corporation like Google or whatever to do? So I think the for my my why I believe this is uh a pivotal moment is not that he created OpenClaw, but if one guy can do something like this, what will a whole organization will be able to do in five years?

Anders Arpteg 1:20:57

And it's possible for one single person to do it.

Goran Cvetanovski 1:20:59

But coming back to my first point, Jensen said, like I pointed on this architecture two years ago. That's of course what it's like a Schmittuberg.

Stephan Erne 1:21:08

Coming back to my because if if this podcast is more about the enterprise perspective of it, yeah. All the enterprise people I've talked to so far, they they they they know what they don't think.

Henrik Göthberg 1:21:20

But this is happening. This is the enterprise divide is massive because we have completely different audiences.

Anders Arpteg 1:21:25

But it is, I think we I I think we hope we can agree that this is a big shift, and most people haven't really understood it yet. I think Jensen has, and I think we'll see it in in a lot of enterprises, in all the cloud providers, they will do a copy of this and have their own version of it and sell it like crazy, and it will change uh value possibilities.

Henrik Göthberg 1:21:43

But what it means is also then that even if Handelsbanken or the other enterprises for Scania doesn't have a clue about what we're talking about, we can tell them what they're gonna be sold in two years' time. This is the argument that they can start preparing for. Is that true? But the question is who should they buy it from?

Anders Arpteg 1:22:03

The cloud providers, or they can do it themselves. Yeah, who knows?

Goran Cvetanovski 1:22:05

But now you also have open source. I mean, like this is the open source.

Anders Arpteg 1:22:11

If you want to go serenity, so you need to have the enterprise functionality. So this is really what NMID is doing. This is what the cloud providers will be doing. So I think the open source.

Goran Cvetanovski 1:22:24

Of course.

Anders Arpteg 1:22:25

You know, Amazon will just share its part, build their own version. It would be better, but it won't.

Henrik Göthberg 1:22:30

Because this is what I said. For me, uh it's it on many sides, he's been doing very large smart things. But to say that we need an AI compound system, we need to figure out what memory is and what all these things as thinking uh pieces of it has been around for a couple of years already.

Anders Arpteg 1:22:50

No, it's it was a big shift. That is a big shift. We haven't seen this possibility. We wanted to have it, but we haven't seen how to do it on the yeah.

Henrik Göthberg 1:22:59

Yeah, you're right. It's to to dream about it versus someone building it is a little bit different.

Anders Arpteg 1:23:04

Okay, so just two other shifts, I think, on GTC as well.

Henrik Göthberg 1:23:07

This was the best uh news that we have pulled out this year. I think because it's a new thing. And we are giving a trajectory.

Goran Cvetanovski 1:23:15

Just to finish this, uh I I'm uh I'm really grateful for this journey because we had the opportunity to visit a lot of the Silicon Valley Giants and etc. So it was super good, and I'm looking forward to their planning already for next year, and etc. And many people can join, so it's gonna be super cool.

Henrik Göthberg 1:23:31

I want to join next year, yeah.

Goran Cvetanovski 1:23:32

I think it's gonna be uh it's it's it's sounded so much fun. It was very well organized. There was a lot of uh so Wallenbergs were there, the the private uh the Slotner was there. We had like a reception, so there was a Nordic reception with the hackathons. We promoted a lot of uh what is called Swedish and no Nordic uh startups there. Um, there was quite a lot of cool things that were happening around that. There was like a really, really big influence, like hey, the Nordics are here, and etc. It was it felt just getting into a conference room and then you hear like a hundred and fifty Swedish people. That's funny. That's really really cool. I think it was really nice now. Uh go for it then.

Anders Arpteg 1:24:12

Okay, so we spent far too much time now on this, but I'm I'm trying to be really brief here. Uh but it's a big moment again.

Goran Cvetanovski 1:24:18

We will go to three hours.

Henrik Göthberg 1:24:19

Uh you know, we have we have hit three hours. It's fun, no problem for me.

Goran Cvetanovski 1:24:26

But you mentioned Virarubin, and I think it's just pause on two hours I allow it.

Anders Arpteg 1:24:32

Toilet break, you know. Okay, but just to I think you know Virarubin is cool, and uh uh it's not really for the reason Jensen mentioned, but so Virarubin is basically the successor of the Grace Blackwell. Is an next run from the Hulk Hobber and and the other kind of GPUs that uh that Nvidia has. But what's special about Vera Rubin, so Vera is a CPU, Rubin is the names of technology for GP. You know, we have Grace Blackwell. Grace was a CPU, blackwell was a GPU before, now it's the next version of that. But what's different here is actually going a bit neuromorphic. And now you see what is thinking about it. Oh, let's go neomorphic, not comp not quantum, neomorphic, yes. And and uh you haven't uh been listening perhaps before, Stefan, about this, but yeah, I'm a big fan of neuromorphic, and basically we have a problem with computers that have memory and CPU separated, and all the power that the current kind of computers and machines needs today is just to move the memory back and forth all the time on flipping the switch. So neuromorphic is about moving them together and even having it in the same as memory sous out. But Rubin is doing that, so they are moving memory not in the same transistor as proper neuromorphic, but very much closer to basically the same die, meaning they can go from you know, the the transfer speed, the bandwidth of moving memory uh was in Grace Blackwell, I think like eight uh terabits or something. Now it's 22. So by just moving closer the memory ship to the same die, they had it's insanely die in a chip.

Henrik Göthberg 1:26:13

What is die? It die is the the wafer basically. The wafer is the die, right?

Goran Cvetanovski 1:26:17

Yeah, you remember Kage Lupescu when he was.

Henrik Göthberg 1:26:20

Yeah, he has a major the Cebedas uh wafer is massive, right? So so what you are saying now is like you're moving the memory and storage into the same into the same uh wafer die, right? And and that then then we can 20 times making the uh transfer speed faster.

Anders Arpteg 1:26:37

At least four, four or five times, but still a massive improvement by making the problem with the memory uh bandwidth problem that we said.

Henrik Göthberg 1:26:44

So it's like it's also it's like uh von Neumann with distance to von Neumann with small distance, and then if you go even further, you can get a neomorphic. Yeah.

Anders Arpteg 1:26:55

So, and then of course, a big problem with in the future that we have so much uh inference um demand, and uh the training demand will be like very, very small compared to the inference demand here. So, you know, Nvidia also has a deal with Cerebras.

Henrik Göthberg 1:27:10

Yeah, but Cerebras is the kings of inference. We know that, right? So Cerebras is uh the ones that said instead of having all these GPUs in many in many different chips, why don't we do one big wafer and have everything in one as one mega ship, which then up obviously is like 20 times more faster inference, right? Which is the the the number one use case is the speed, is speed run inference.

Anders Arpteg 1:27:35

So it means Cerebrus is also very neuromorphic-ish.

Henrik Göthberg 1:27:38

So it's it's it's going in that direction, but still, you know, but it means sort of all of them perplexity, open AI, everybody have done on their premiums, on their premium offers, you know, they are you they are they are running on Cerebrus machines in order to get speed of inference, which means it translates to the whole When you do the more advanced uh thinking you're doing and reasoning, if you're taking the premium, you know, stop talking now. Jesus Christ.

Anders Arpteg 1:28:08

We have to cut this short, right? It's so cool. It's so cool. But just I want to have a Neilon story as well in here with a terraf, and perhaps you can find a segue to this topic.

Henrik Göthberg 1:28:20

So you're not that you're talking, but there needs to be a place where Elon Musk in this car is okay.

Anders Arpteg 1:28:28

So moving into the automotive, self-driving cars, physics AI kind of world. And of course, um, Nvidia is playing big here. They have the big Alpha Mayo, the open source version of self-driving car kind of functionality. But they also had a big deal with all the automakers in trying to find you know they can be the orchestrator of self-driving cars and trying to to by bringing I think 12-13 of these kind of companies together, they can at least build this best kind of FSD in through the world. And of course, it's a big play against Elon because Elon is having the best solution here, uh, I would argue. And then they are saying that we we can at least you know starting to find an alternative, which I think is super good because I'm really scared about you know the concentration on power that's but who is in this uh consortium Mercedes and BMW?

Henrik Göthberg 1:29:25

And what's it called again? Terra.

Anders Arpteg 1:29:27

No, no, okay, but Terra is another thing. That's enough. I'm coming to that soon.

Henrik Göthberg 1:29:31

Okay.

Anders Arpteg 1:29:33

So you know, during the GTC, they actually did this kind of I didn't I don't recall exactly what it was called, but they did a launch where they they had a partnership with a lot of automotive companies that are working together to build up this kind of you know uh yeah, autonomous briding kind of functionality, right? Because they can't do it themselves, then they but then they have the partnership, and with Nvidia they should rule the world, but I don't think they will. So, well I hope I hope I think it's super good because we need to find an answer. So this was actually discussed on the GTC.

Goran Cvetanovski 1:30:05

I actually don't agree, okay, because I think that Google is actually leading the way. Ah okay, just because you drove away moment. So in one week, I think we we had like uh I don't know, at least like a 50 of those drives. And I don't understand if people you know, like if people didn't believe that this is actually like this is working, it's working. Go to San Francisco, you will find more autonomous vehicles on the the the the of course, yeah. So like and that is Google, and then you have China, which is uh not China, it's something in Him or something like that I don't know what's called which is Amazon No no it's Amazon and Google and Google. I don't know, wait. What are you talking about? Like the Tesla's like two times the actual rabbit told is a different thing.

Anders Arpteg 1:30:49

It's not the different see if I can convince you then.

Goran Cvetanovski 1:30:55

Yes, I mean, but uh at least I found uh I I didn't talk to my shot. We are moving the podcast, so let's go back to it.

Elon Terafab And Moon Computing

Anders Arpteg 1:31:07

Okay. We take this afterwards. But finish it off now, Andosh. You have a story, that's sorry. So I just want to make the segue to Terafab. Okay, so Elon just made this amazing story about the biggest kind of investments ever, and they're going to build their own factory for ships. Now we know that the the uh ship manufacturing is very much a monopoly now for TSMC and Nvidia and Nvidia. Together they, you know, Nvidia is designing the ships and then TSMC is manufacturing them, and then we have the European ASML who's creating the machine, the lithography machine that is creating it. And Elon, of course, um wants to reduce the dependency at a huge margin. You know, Nvidia is taking like 70-80% uh uh margin on every kind of ship, it's insane. So, of course, he wants to get rid of that. So he's launching this new uh super, super, super big uh terafab uh announcement. Now, what he's saying with terafab is basically saying it's going to produce a terawatt of compute per year. Now, today, if you combine what uh TSMC and the rest of the world is doing, it's about 20 gigawatts.

Henrik Göthberg 1:32:23

It's a 50th terra.

Anders Arpteg 1:32:27

So it's he's going to say that with Terrafab, they're going to produce 50 times more than we're doing today in compute. And when? By what by what time? Time frame? 2030-ish.

Henrik Göthberg 1:32:41

Is it going to ramp up to have that capacity in 2030?

Anders Arpteg 1:32:45

And he's going to do it on the moon.

Henrik Göthberg 1:32:52

This is cool stuff, yeah. Isn't this amazing? It's amazing. I love it, but I mean, like so. I'm learning from these guys every time. Make your claim the most outrageous, and then you get media, and then you sell some other shit.

Stephan Erne 1:33:13

Even people in Sweden will talk about it, huh? Yeah, but it's cool, man.

Goran Cvetanovski 1:33:17

By the way, I bought the plot next to the factory so I can.

Anders Arpteg 1:33:21

But is it this bullshit or is how can we how can we look at this as any of course the starships, you know, that the sea is building, that's going to be the enabler for him to be able to build the manufacturing on the moon. He's going to he's already starting this year with the manufacturing Texas Austin for the advanced um ship manufacturing plant or whatever. It's just more or less a prototype. And then he's going to start to build up um the production on the moon as well in the end. And the reason is, of course, doing the data centers in space. So by building the data centers in space and having the ships there, they have a huge amount of more energy directly from the sun, much more than ever can do on Earth. And if we just can get the cost of bringing the material and the necessary conditions into the moon, then the manufacturing there will be much cheaper. The launch cost, since the gravity is so much less than the moon, is six times less, they can they don't need a rocket, they don't need rockets to launch the satellites anymore. So as soon as you come to the point where it's cheaper to launch to just you know get the material from Earth to the Moon, it will be much cheaper to build and run satellites and data set centers in space there. But it's not just Elon doing this. This is a comment on everyone's.

Henrik Göthberg 1:34:45

So the logic, the logic of the physics that this is a good idea eventually. I I think it holds. Uh, but are we in the point of view where we can actually industrialize to get there in use four years?

Anders Arpteg 1:34:58

You know, he's been running the Starship now for a long time. They more or less need to have it reusable, meaning. Yeah, yeah.

Henrik Göthberg 1:35:05

But I think this is part of the journey to Mars, of course.

Anders Arpteg 1:35:08

Or is it? I think it is because I think you know, can you see that the economical kind of uh value that he will gain from this is insane? You but this is it. The point is he doesn't need to go to the Mars to do it, it's sufficient to go to the Mars.

Henrik Göthberg 1:35:23

You put a bold game like Mars to go to Mars, and with that bold goal, it it unlocks key things that drastically reduces how we do things. We launch uh Starlink, we launch Latlice way cheaper than anyone else, and we own that market.

Anders Arpteg 1:35:40

We we we we we need to be able to launch things out of you know from low gravity space, so we need to have a lunar base, part of the journey to Mars, but at the same time, an insane um productivity frontier uh increase that he gets when he gets and if you get cons uh like a conspiracy kind of mindset, you can say that you know the Mars thing is from uh more of a um let's say not really the main goal that he's having, but he's doing that from more uh um building the enthusiasm up, but in reality, if he just gets to space, if he can take you know build the start to to exploit the moon, and what he can do if you start to reuse the material on the moon, Jesus Christ. I I'm actually you know both excited about this, but I'm super scared about this.

Henrik Göthberg 1:36:33

No, so so this is the uh the downside of things. Who owns land on the moon?

Goran Cvetanovski 1:36:39

I I actually have that work. You know, there was this website that was selling uh that was selling like space on the moon. I actually have a diploma. I bought one supposed to be a little bit more than a corporate owns your uh land. But what what what what is it? Yes, but let's go back to how to implement Chat GPT.

Anders Arpteg 1:37:03

Yes, yes, okay. Let's stop this. Anyway, but at least you know this is big news.

Henrik Göthberg 1:37:09

I think you know, more than is it news or is it just an uh uh expose play? What do you call it? An an oot spiel. Is it is it news or is it just uh we'll see in a couple of years, but I I I think it will happen.

Stephan Erne 1:37:26

It creates it creates headlines at least.

Henrik Göthberg 1:37:28

It creates headlines. What was your view? Uh listening to this uh whole it's it's very entertaining to be honest.

Stephan Erne 1:37:35

Yeah, it's really that's pretty cool. That's the first thing I have to say.

Henrik Göthberg 1:37:37

That's all we we all that's all we care about. That's all we care about.

Stephan Erne 1:37:41

Yeah, but I mean you shouldn't underestimate. I I agree to your point, you shouldn't underestimate uh the power of having extremely bold visions and how this is motivating people around you. And even if you don't reach the vision all the way, or only the steps is in between are massive. So from that perspective, and it's of course, I think to be honest, it's impressive but also scary how single people are driving to do this stuff like this. This is very philosophical question, of course. Maybe not the right thing for this podcast here, but no, it is, it is, it is the right thing. This is exactly you're looking for rabbit holes, yeah? Yeah, we heard that before.

Henrik Göthberg 1:38:23

You know, what is the we we started it based on the AI divide, and this is exactly at the AI divide. What happens when one guy and not the society is doing this?

Stephan Erne 1:38:33

So, and of course, and unfortunately uh from my personal perspective, unfortunately, we see more and more proof points of this that single guys outside of democratic rules or whatever, what we think is should be right or wrong, are having visions, are able to align people, money, yeah, whatever it takes behind them to get going. And and and even if you don't believe in the vision and you think they never get that, they will they have such a huge impact on society, on yeah, because technology, that's technology-driven business transformation. That's a little my if technology changes stepwise, it will also have sooner or later a stepwise impact on society and on businesses.

Henrik Göthberg 1:39:21

Yeah, and I made a post that where I'm trying to explain where the problem is, and I was I was doing the metaphor. Elon Musk and some other guys are driving race cars around the track and going 100 miles an hour, and then they are with so much money, so they are able to go off the track, crash it, take get back on track, and and keep moving. And at one time, in one way, it's sort of all the way improving society, improving all this. The problem is the guys responsible for investing in the race car, right next to the racetracks stands billions of people, the spectators, looking at the race, spectators of the race. And every time he crashes off the rails, millions of people die as an impact that he's not really oh, they're spectators. So that's the tricky one when it's like uh one man race is driving this and it's impressive, but they're not picking up the bill, or they they don't really in the way the race is working, they're not responsible for the spectators. That's tricky, yeah. Do you see what I mean? Yeah, absolutely. On the other hand, if they wouldn't race if they wouldn't race, would that be better? I'm not sure. Anyway, thank you for that uh comment. Nice one. Back to back to the real story of uh digitalization in enterprise.

Germany Versus Nordics Change Culture

Anders Arpteg 1:40:46

Cool. Uh let's try to bring it down back to that. But um back to Earth. Back to Earth, literally, Stefan. Yes. I mean, I I think it's interesting. You you have a deep understanding of both the German market but also then the Nordic Martic uh market. And I would love to try to understand a bit more. I mean, you both have also the understanding of the telco and the banking industry, so it would be fun to just try to understand a bit, you know, what the differences are there, uh in terms of culture or you know, maturity, or or or really how how the differences in terms of driving change.

Stephan Erne 1:41:24

If you if you start with Germany versus the Nordics, I'm trying to take this from a also of course from an AI perspective. Yeah, but I think it's of course I think it's good to understand the basic drivers or differences to also understand how the um adoption of different technologies now, for example, AI, I from my perspective, uh working. And uh I think the the biggest difference that I've ex that I've experienced, and that doesn't mean that it's right or wrong, but at least from my experience, but between the the German way of driving a company and and and the Swedish way is that in Germany things are much, much more top-down. Much, much more top. Yeah. So if you have it, if you have a problem, um you potentially take in one of the big consultant companies, they make an assessment, and they make an implementation plan, and then they leave you. And then the the the top leadership thinks the problem is solved. We have a we have an assessment, uh, we have an implementation plan, and then the thing is top-down to to uh to drive the plan. On the other hand, the organization is trying to find all kinds of reasons why this plan is shit and doesn't work. So you have somehow the whole organization against you because uh they are it's fear they're gonna lose power, whatever, and then they don't trust that stuff. And of course, and in a lot of cases, I mean I've been working on some you have some uh 20-year-old Porsche guys coming in and explaining to you need to explain workers' counsel, yeah, yeah, I can do this as well, yeah. Which I think is this has has some advantages as well, yeah, because it's somehow um regulatory force, coming back to regulation. But um I think this top-down approach is is one I think the the thing that's happening, and in the US maybe even more, but even in Germany, and and that means that also the um from a cultural perspective, people are waiting for someone to tell them what to do.

Speaker 6 1:43:26

Right.

Stephan Erne 1:43:27

And this in an organization with uh 10, 20, 100,000 people makes, of course, a lot of people waiting for orders, which also, I mean, I think the the hierarchy is is a totally different thing. As a as a as a if you have a high position in Germany, people expect you to tell them what to do.

Henrik Göthberg 1:43:46

Is that a good or perfect?

Stephan Erne 1:43:48

Hey, or start with contrasting. Wait, start with contrasting. What's the Swedish way? Exactly. So I'm coming back to you what is good and bad about the one or the other thing. When when I moved to Sweden and I joined Ericsson, I tried to understand what is happening here and how does the company work. I had all kinds of questions and nobody was able to answer them. And then I I organized we organized some meetings up again. There were 10, 20, 50 people joining that because nobody really knew how things are working. And then consensus-based. And then somehow it was this consensus, and people agree, everybody knows each other, so much less process-driven from my perspective than it was in Germany. But on the other hand, people are much more they wanted to contribute. So everybody had a totally diff from my perspective, very different.

Henrik Göthberg 1:44:38

Um the way you tackle problems becomes very different when you're not waiting for orders, but you're trying to debate and discuss from bottom up in a way, and then and then the decision is what the majority is doing.

Stephan Erne 1:44:52

So yeah, which makes this process, of course, much longer. Yeah, but then at least it's it's much wider committed, committed, and then it works in the execution works faster. But the the way to find the solutions and to to agree on what is happening is endless. Yeah. So and this coming back to your question: what is better, what is worse. I think um in a deterministic way, world where a McKinsey, Bain, BCG, whatever can come in and make an assessment, fact-based, data-based, and say, This is what this is our assessment, these are the options, this is what we believe is the right thing. Um it works pretty well as long as they have an operational excellence and a way to make people execute it.

Henrik Göthberg 1:45:41

This works in a stable, slow-moving context. And now you put this in a VUCA context when you have a scenario and uncertainty five times we said the index said, does that work anymore?

Stephan Erne 1:45:53

Yeah, yeah. And and that's of course interesting because if if you you usually it it doesn't work because top-down assessment could go very wrong. But if you wait for the bottom-up approach, because it's also fast, it's not only VUCA, it's also extremely fast.

Speaker 6 1:46:10

Exactly.

Stephan Erne 1:46:11

If you if you let organizations um work on it for one or two years to agree on what is the best way, that doesn't work either. So unfortunately, I don't have a clear answer to your question. What is good or what is wrong?

Henrik Göthberg 1:46:28

And and and actually, what you what you're saying is you are discarding both models. So both you're discarding both models and looking for a third. Because the top-down approach where you have we have a fact-based, but you have limited understanding, we don't have all the tentacles is is one way. The other one where you have endless debate is uh is another way, and what you're looking for is something that has way more agency to act within its uh approaches. Well, this is a different ball game.

Stephan Erne 1:47:03

Exactly.

Henrik Göthberg 1:47:03

And and and this is and I think that this new model is a third model we're talking about that we need.

Stephan Erne 1:47:08

The new model has somehow to fit into the the because you if you have 100,000 people who are waiting for instructions, doesn't work. You cannot move this into now you have 100,000 people who are waking up every morning and think, how can do my work better and and are pushing bottom up. It doesn't work either. That doesn't work, yeah. So I think it's um this is a third model. Do you see what I mean? Yeah, yeah, exactly. And I think the the the prerequisites to create this third model, from my perspective, are better in the Swedish context than the German ones, because here at least you have a lot of people thinking about it and having a driving force and not awaiting for stuff.

Henrik Göthberg 1:47:52

And now but now it's a bit about organizing decision flow smarter in that.

Stephan Erne 1:47:57

So you have to to somehow streamline this consensus.

Henrik Göthberg 1:48:00

The consensus is clear, but you need to then think about how you comportment like that so people can act and move on their thinking, and you end up in a thinking game and consensus game of five instead of fifty. So there so there so you're thinking about uh agentic ah, I have I have a new terminology for me. Distributed agentic intelligence reform. Think about that.

Stephan Erne 1:48:29

And I don't know if you have to you can stop me if I'm talking too much, but I think this the whole uh agile transformation was one proof point for that because I was talking to people from Lidl in Germany, how that worked to add to introduce agile. Like not at all. And I think in Sweden worked pretty well because people were used to take responsibility about them creating these smaller teams, taking responsible for stuff like so. I think um coming back to the cultural difference between the different countries, I think in Sweden it's easier to make those changes because I I I feel it's a higher maturity in the organizations as well, overall. And if you if you come with a new model, if if you can explain it in the right way, you will also get people along.

Henrik Göthberg 1:49:16

It's tricky to say maturity, I think it's simply the cultural fit is naturally higher for that angle. So that in this sense, the yes, they start from a higher maturity point of view if you go in this decentralized direction.

Stephan Erne 1:49:31

And I mean I had a lot of I got a lot of questions from from my friends in in Germany. How can it be that Sweden with like a tenth of the population has so many international global companies?

Henrik Göthberg 1:49:44

Exactly.

Stephan Erne 1:49:45

So I think this is w w one part of the of of the the reason for that.

Anders Arpteg 1:49:52

I think it's interesting if you compare to some of the tech companies in in the US as well. Perhaps even Tesla could be a good example to compare with here. Because I think it has a bit of both of some German sense but also some Swedish sense in it. Meaning, for one, you know, Yelen, of course, is almost like a dictator that in the Tesla competition and he says something, that is what goes, right? Nothing can be said against that. But on the other side, it's extremely um democratize is the wrong word, but they they have a way where anyone is supposed to supposed to really come up with what you should do. Um and the people that do not you know contribute to these are a good way to just optimize this is a water pump in a Tesla engine, then um they're going to be you know reduced in in value. So in some sense, it's very, very much a bottom-up approach in Tesla companies, I would say, and in other companies. And uh and then you know, you're giving a lot of authority as a small, small team, like three, four people, to make change decisions about uh regulation, about budget, and they have more or less unlimited budget, and and they can just you know do what you do, better to make a decision and go with it than be wrong than not to make a decision. So, in some ways it's extremely uh trusted kind of an environment, but also almost like a dictatorship in some sense.

Henrik Göthberg 1:51:28

So I think you know it's a dictatomy because it you know it because what they have what he has created as a culture, I think SpaceX is a good example of this, maybe even better than Tesla, is is this very clear top-down that we can move from that direction, it's clear where we're going, but then the very much autonomy in the smaller teams and making that work at the same time, which is very different. I mean, like we are not making that work. Sweden are going all the way on the on the bipolar spectrum in one end, and then we have the other top-down hierarchical approach. I don't even think Germany is the worst uh in the world, but and then we have the some of these super companies is finding a way to balance the extremes. Do you agree? I don't know.

Anders Arpteg 1:52:12

I mean, I think we should do more research on this because obviously it's found a recipe that works, you know. Every company that uh Elon owns is operating in the same way and it's being very secretive about you know how the way working there works. You can you can see some videos on it on online, and I actually been studying it quite a lot in recent years.

Henrik Göthberg 1:52:28

Because it is interesting as a research phenomenon. I mean, like you can think whatever you want about Elon and if he's an idiot or not, or is he crazy or not? But if you used to look at the mechanics of how things and the results, it's quite interesting.

Anders Arpteg 1:52:41

And he's moving at a speed that you know, if you just take Germany and I take the automotive company there, Jesus Christ, I'm scared about them. I mean, I wish them the best, but it's super tough to be able to do that.

Henrik Göthberg 1:52:51

But you can also look at if you look at the if you take automotive industry, it's very interesting, and automotive in electric vehicles and the speed of light that the automotive industry is moving in China, which is also some sort of dictatorship authoritarian, but goes really hard there. But then how it works in that speed, of course, it's not sent every every single with that speed they have in China in terms of things popping up. Of course, it's not controlled top-down. It's it's it's a direction which is dictatorship, but then it's very fluid. This is also very interesting.

Stephan Erne 1:53:27

Yeah, and I'm a little astonished that this operating models are structural or things I'm not discussed more. Exactly. Like you said, I think I think they are so crucial for success and not success.

Henrik Göthberg 1:53:40

This is the real design, and this is where I'm sometimes getting a little bit fed up with the the big uh management consultancies where I don't think they are they are not deep enough in how the technology works and how the innovation speed works. So they are there are there are I don't know. It's something here that I don't think the management consultants is picking up. I mean like they're trying to parrot the shit, but they they are not really able to repackage this. And and and I I and I mean like we have Daniel Lankilde here, who is very, very he works with uh test data and how to uh you know uh annotate data, especially for the uh autonomy, you know, for self-driving cars and all that kind of stuff. And he's been doing a lot of posts around the European car ecosystem, manufacturing ecosystem versus China and like. And we we need to we need to wake up. There's a big wake-up call in some industries here. And and I think if you think about what you said now to research and understand these operating models, when you look at the patterns what they're doing, I think they're transferable to banking or whatever. Yeah, yeah. Um you need to apply them in your context, yada yada yada. You need to figure them out, but you have some fundamental pattern shifts here.

Stephan Erne 1:54:58

Or yeah, but at least you need you need a driving force that wants to achieve a moonshot. Yeah, that's moonshots.

Henrik Göthberg 1:55:06

Yeah, moonshot, moonshot a bit of a sortie, very interesting.

Blockchain And The Libra Near Disruption

Anders Arpteg 1:55:12

Okay, time is flying away, and I'd love to just get into some bit about your also other areas of interest, which could be quantum computing, it could be blockchain, it could be other areas. Um but you mentioned that they could be disruptive. Yeah. Uh if we start, you can choose, but if we go with quantum computing or blockchain, yeah, blockchain is maybe a good example.

Stephan Erne 1:55:35

Okay, because we were looking um, and you know, I had a uh I had a team that was uh scanning extremely well what is happening technology-wise mainly, and then we were trying to to see what implications this potentially could have on the banking industry and on Hannes Banken, of course. And uh I think one of the coolest things that happened uh during this time was when um uh Facebook announced to create uh Libra. Libra, yeah, this uh new cryptocurrency where they created this Libra foundation in Switzerland. I don't know how much you followed that or not. But was it was also like um I wouldn't say moon, it was not a really moonshow, but the the the vision was there should be as easy to send money as it is to send a picture in WhatsApp. So exactly and that's vision is of course not uh it's it's not that far-fledged, but um that was the the the the driver behind it, and I think they and as we assess at least uh they created and and I mean of course they have the market power to to put another button in WhatsApp, uh distribution, like you said. They have uh billions of of potential customers who could use it from day one. Um, they had a lot of thinking behind it about uh financial inclusion. Uh so it was maybe from a Swedish perspective, say we have Swiss, what's the problem? But of course, for from uh from other in uh parts of the the um the world, the world where this is not possible, financial inclusion, even in the US, sending money is not that easy if you uh as you know. They still have checks, right? Exactly. So I think that there was a big um big need um in pain point, and they created a model or uh they based it is on the and then we talked about blockchain, then of course you can discuss how much real blockchain was behind, it was a more centralized model, but I would say the whole blockchain um hype was was driving this, and they created a model where you could you could um exchange your crowns or yours, whatever, into those Libras. All this real money, the fiat currencies, as they're called, they would be collected in uh uh Libra consortium or whatever in in Switzerland. So they what could collect all those money, they had partners from uh Spotify, Visa, MasterCard, 20, 30 partners, because they also realized if this money, whatever you want to call it, is not you cannot use it because it doesn't make sense. Blah blah blah. Uh um and the the whole thing, at least from our perspective, was so well thought through they had the right partners on board, they had the safety somehow of this uh Swiss uh Lieber Foundation. Um that this would have disrupted the financial industry like crazy, and not only the financial industry, but even, and I think this is what at the end of the day stopped it the power of the US dollar.

Anders Arpteg 1:58:41

Interesting point.

Henrik Göthberg 1:58:42

So, what happened in the end? So, how did this play out? Uh, from guess the Libra story. It was a couple of years back.

Stephan Erne 1:58:47

Yeah, from what we've seen is that um Zuckerberg got an invite from the Senate, and they somehow told him if you don't stop this, we're gonna kill your company. Yeah.

Henrik Göthberg 1:59:01

Interesting.

Stephan Erne 1:59:02

Because they were so and and and once again, we assessed this. We had a actual pretty bitty, uh pretty good big group in in the bank, because I can not really not judge what what kind of impactations that would have. But we had some people from our finance department, so um pretty high uh a big bunch of people in this process to assess what this potentially could mean. And uh, I mean, of course, the disruption for the financial sector would have been good, but big, but even for the central banks, because they would have left, they would have losen the power over the financial policies of uh control with the inflation and employment rates and if if Facebook now is deciding and leaving Libra, Facebook is deciding Libra. What are the the rates uh for saving Libra?

Henrik Göthberg 1:59:53

And would did you so uh tell me where when was this playing out? Is it a couple of years back, or when it was it? Yeah still playing out.

Stephan Erne 2:00:02

No, no, no, no, they shut it down there. I think it was if first it was called Libra, then they they changed it to the DM. It was that they changed the name to DM because this Libra thing is.

Henrik Göthberg 2:00:12

Is it five years ago or something or even more?

Stephan Erne 2:00:16

DM um right.

Anders Arpteg 2:00:20

DM and this is and and and this is also, I mean okay, but interesting. But the your point then is you know, blockchain as a currency at least could have a big disruption power potentially, right? And of course we know uh Bitcoin has you know worked rather well, right? Uh so in that way, you know, blockchain can reduce the the need for the middleman, which could be the banks in this case. Yes, but yeah, I just like to ask, okay, so if we we can see for currency, it really could be disruptive. But have you seen that much disruption from blockchain as a technology otherwise outside of currencies?

Stephan Erne 2:01:05

I would say no. Um, and I think this has a lot to do with the fluctuation of the value. Yeah, that's why Bitcoin, for example, or these other values, uh whatever. I don't know if I don't know if you will call them a currency or an asset or whatever, but that the fluctuation is so high that you you cannot really use them as a payment method. That's why these stable coins came in now, uh during the last years, where a lot of people believe that this is uh somehow taking the advantages of the stable currencies but putting them on new payment rails.

Anders Arpteg 2:01:42

But blockchain could be used for so many more things, and uh now I'm going to you know enforce or talk about my biases here a bit. So please disagree with me. But you could use it just to have some way to let's say sign legal documents between companies to have some other company that doesn't have to be involved, you can simply have the blockchain that and for some kind of you know uh uh insurance that it's it's accurate and it's immutable as the blockchain is, right? But it doesn't happen, it doesn't work. And um, I mean, even in Spotify days, we try to have a blockchain because we wanted to, in some way, to ensure that people artists could trust the numbers of views that you had and how much we should pay to the artists, you know, depending on the users, but it failed also completely because of a simple reason. In that case, we still have Spotify as a middleman, and Spotify had a business model which you know made it positive for them to be the middleman, and the blockchain would remove that possibility, and simply it destroys the business model of any company that tries to use the blockchain, I would argue. So that's why we have more or less not seen any company finding a positive use as a business model to use blockchain. Yeah, I haven't seen one, so I would really try to.

Stephan Erne 2:03:02

I agree, and I think one and there were some other first use cases when the whole thing popped up, was in global trade. If you sell something to India, whatever, how many middlemen are in the chain to tick the box that something has happened? And of course, you could put this on a blockchain, then it gets much easier. But I think here the big problem is that these oracles, so the the the institution who is signing off thing, you still need them. You have somebody has to sign off that uh something has arrived in China, whatever.

Henrik Göthberg 2:03:36

I mean, like so maybe that's the point, right? You wanted to remove that fundamental mechanism with the blockchain, but as a society, we weren't really ready for that, or it it it it it doesn't really fit, it's not the product market fit for how society works. So those institutions, those middlemen, sometimes they're middlemen, sometimes sometimes there are certification bodies, and and and I the argument is of course that there's a lot of waste in there, so that's why we want to do it in a different way. But you but it's always like you end up with well, but who is certifying the blockchain? Exactly. It's a little bit like you you you didn't remove the problem, you shifted the problem to another power. Is that better?

Stephan Erne 2:04:17

If you look at the process, I think the the things that are hindering or slowing down the process will not be removed by a blockchain. They are still there, they just have to sign on, sign up on another technology layer. So I think that you have you would have you you would need to redo the whole process. And this is maybe what takes us to the AI thing as well, because just put a little AI in the old process.

Henrik Göthberg 2:04:42

Sugarcoating with, I mean, like sugarcoating with blockchain won't work. Sugarcoating with AI won't really do the whole trick either. And this is maybe the big, big chance challenge, you know, how do we how to really reform or re-engineer a bank is really, really hard then because at the same time you are not there to be this, you know, to have failures. We want to reform at the fundamental level, but we can't have failures. That's a really hard uh equation.

Quantum Computing Reality And Benchmarks

Anders Arpteg 2:05:10

Moving to another technology that failed? No, sorry. Uh okay. I I we if we move to quantum computing, uh it's a national, you know, it's a it's a topic of mine that I like to discuss uh and and uh and I have some strong opinions there. But please, if if you would, what do you think about quantum computing?

Stephan Erne 2:05:29

Um I think it's also one of the technologies that has a huge potential that probably has a pretty steep curve in how how it is uh developing. So I mean, judging quantum computing now compared to five years from today, and which probably has one of the most disruptive powers in in different perspectives, and that's why we also actually worked with it like four or five years ago. We as Hans Bunken, uh, we took in a lot of people from university to try to get a grasp on this. Yeah, to get and and and both and and we also do this from from two angles. The one is to really understand the technology and and let some people test and try different solutions that were on the market, everything from IBM, from D-Wave, annealing qubits, and so on. So we we we ask them to go out and really assess what kind of different technologies are there, what is the path, how are they um progressing, and on the other hand, of course, run around in the bank and see what kind of potential problems could we solve with it or not. Um and then finding people on on the on the on this business side uh that were willing to to work with us together to to understand and and assess uh how that works or not. And um, it's what we at least realize is that uh, and we we did a lot of test cases and also we're running those our today's problems on these D-Wave annealing qubits, for example, to see are they solving it better or not. And the idea was really to have some test cases ready and test them like all the time against the new technology, because then we would know when is the tipping point to look really into this, right?

Henrik Göthberg 2:07:14

Exactly.

Stephan Erne 2:07:15

Yeah, but if you if you don't prepare for that, if you haven't made your homework there, as your test cases that you want to benchmark against.

Henrik Göthberg 2:07:21

Exactly.

Stephan Erne 2:07:22

So that that's what we did in in some of the um mostly in um uh risk uh assessment and uh capital markets. We had two two or three cases that are already there to benchmark the test. But the one of the biggest questions, of course, in in uh crypto. How what we do with our crypto crypto keys and how do we make them uh quantum safe going forward? So these are the I think are the the the two areas that we were looking into where I see potential disruption happening.

Henrik Göthberg 2:07:54

But but this is now such a nuanced topic, and I think um we can go into specific quantum techniques that might are working well together today, and and and and that is is is good, and then we get into fundamentals of getting quantum computing to scale.

Anders Arpteg 2:08:14

Uh, and I think this is also where quantum computing technology is working well today.

Henrik Göthberg 2:08:19

No, I'm not the expert, but for me it's a little bit like what we are using uh I mean, like I'm trying to um paraphrase Salah Francians who are who are trying to make a case between there are there are quantum technology or quantum techniques. You're thinking about quantum mechanics, uh that's the fundamental of quantum mechanics can be applied in specific use cases and work really well today. That is not has nothing to do with the field of quantum computing, it's two different things. So, what I'm when I say quantum mechanics or quantum techniques, I mean technologies are quantum mechanical based, but it's not the same as saying that all we have a quantum computer, it's two different stories, and that's what I was trying to convey. I'm not I'm not technical enough under so you take take it over.

Anders Arpteg 2:09:05

So it's not a rabbit hole. I can speak for half an hour about this. But do you see what I'm saying? I was actually a better user of D-Wave 12 years ago or something. So I tried out the the aneline qubits at the time in a simulated kind of a computer. But no, but uh, you're right, Henrik. So, of course, quantum mechanics is something that exists and works extremely well. For certain use cases, we couldn't have any kind of mobile phone, computer, or anything, or even a transistor wouldn't work without the theory behind quantum mechanics that we do have today. So, quantum mechanics is there, very useful, and um, you know, not very little would work today without the theory behind quantum mechanics. Quantum computers is a different thing, of course. It's the the idea of supercomputing entangled qubits that we have. And uh and if anyone claims that there is a working quantum computer that provides value better than a classic computer, I would say they're very much wrong. So there is not a single quantum computer today that is more powerful than a classical computer.

Henrik Göthberg 2:10:05

Yeah, and just to put that into context, what we talked about. When we talk about quantum crypto, there's these techniques working on quantum mechanics that is quite smart, right? It has nothing to do with quantum computing, it's quantum mechanics used in the crypto context. Am I right? No, no, no, you're wrong, actually. Oh, okay. This is how I understood it. Explain it because there are quantum techniques in crypto that I think is working. You don't think so? I don't I know so.

Anders Arpteg 2:10:34

No, no, but okay, so quantum computers could potentially decrypt messages that RSA or these kind of public key encryption techniques could do if we could have a quantum computer that could scale. There is no quantum mechanics techniques used in crypto technologies today.

Henrik Göthberg 2:10:58

What are the cases where we had we talk about quantum keys? So quantum keys.

Anders Arpteg 2:11:02

Are you thinking about the quant uh quantum key distribution? Perhaps that's exactly something we used in my previous project.

Henrik Göthberg 2:11:08

This is what I mean. So this is quantum mechanics used in key distribution.

Anders Arpteg 2:11:11

Yes.

Henrik Göthberg 2:11:12

That that that that means that is what I'm pointing at. There's quantum mechanics techniques used for certain specific cases within security, like key distribution. It doesn't that work? That works.

unknown 2:11:25

Yes.

Henrik Göthberg 2:11:26

You used it.

Anders Arpteg 2:11:27

But it's not an encryption technique, it's a technique to to send symmetric keys across long distances in a way that's impossible to intercept.

Henrik Göthberg 2:11:37

Yeah. But isn't that what I'm trying to say? There are there are quantum mechanical techniques that can be applied in certain use cases within how you how you build your uh encryption uh approach. And this particular case was keys. Isn't that part of encryption? Isn't keys part of encryption? I I I I am right, but We are talking about different levels. So if you go back down to in in crypto, one of the key aspects is keys. Keys can do be done with quantum mechanics. Then I'm right. Don't say I'm wrong. I'm right, but I'm right in a very narrow niche. We can say he's right and then we go on. And then we take the discussion later on. It's okay. Yes. Oh, yeah. I was just trying to make the once again. The nuance is the point. As long as we are not nuanced when we're talking about sovereignty, cloud, AI, blah, blah, blah, or quantum, it's really hard. Sorry.

Anders Arpteg 2:12:35

Okay. Anyway, I think you know, quantum computing is a very interesting field, and it's fun to see what Google is doing there with uh the Willow ship that they are building, and uh Microsoft is doing a lot of progress there as well with the Myorama ship. Um, but it's super hard to scale. And uh I don't want to go too far into it. I'm I would have been happy to speak much more about this, but the problem is that we do have working quantum computers, but they are in super small scale. And the problem is we have not seen them being able to scale beyond like a thousand qubits. What we need to have to have them do anything practically useful, which is like decrypting RS RSA case of like say a thousand or two thousand bits, we would need like a million physical qubits, and we only have a thousand of them today. And I would argue that it's exponentially harder to scale uh as we grow, and uh we haven't seen it for 40 years uh than being able to scale.

Stephan Erne 2:13:34

So you don't expect it to scale pretty in the new future either.

Anders Arpteg 2:13:37

No, but uh but I I'm yeah, probably wrong.

Stephan Erne 2:13:43

Hopefully you're wrong, yeah.

Anders Arpteg 2:13:44

Hopefully, because it would be kind of dangerous, but sometimes yes, because quantum computer could open up for a lot of very exciting areas of of possibilities, especially in AI. So one of the things I did with D-Wave was actually using quantum computers with machine learning, uh, because you can have the quantum computer doing the training more or less, and you just send the supervised training data to the quantum machine and then it trains in an exponentially faster way if it would work. Um, so it would be super fun, uh, but also a bit scary, of course, because you could potentially decrypt uh all the public key encryption techniques that we have today.

Henrik Göthberg 2:14:21

But one of the things that we have talked about off and on on this pod has been like, okay, we we put all this effort into understanding quantum computing and it's a quite complex problem to solve. At the same time, new morphic computing seems a little bit closer to the von Neumann and what we are doing today. So but it's it's it's a storytelling that very few is listening to, or or it doesn't get as much uh you know hype or media coverage, but it's much more feasible. So it's it's so if if you really want to understand the journey, you are you kind of need to understand these different approaches and where they fit. And I think that's interesting how little we're talking about neomorphic compared to quantum.

Anders Arpteg 2:15:05

Yeah, or photonic AI is a super interesting field, so it's more much more realistic.

Stephan Erne 2:15:10

But I would say to come back to the enterprise angle, yeah. You you need to find some use cases where this is applicable. Yeah. Because people are usually not that interested in this uh back-end uh technology development stuff or whatever. Yeah, but you need some people who are looking at that to understand which tracks are going on and what kind of companies should you work together with. All these techniques, but what should we use them for becomes the real question. So I think that the the most important part for us, at least, was to find some potential use cases for those technologies so that we can test them and also then prove internally that this is really making attention to scan techniques that can potentially disrupt the whole bank is part of the game, right?

Henrik Göthberg 2:15:51

Exactly. But then but then to understand what is just invention and what is actually real that becomes adoptable, this is two different stories, of course.

Stephan Erne 2:16:01

Yeah, but I mean the the value you only create for the company if if it's adopted. Adopted, exactly, if it makes sense. So I mean, even something in the in the background and you you cannot get it in, or it doesn't make sense. So I think it's it's important to to find people, and that's why once again we talked about that before. It's important to have people on even on the business side who are interested in that and think this is a cool cool thing. They're curious, they want to learn, and then they also help you too. And because if they start scanning the market and see what their other competitors are doing and so on, that that really helps you.

Henrik Göthberg 2:16:34

But this is a philosophical or sort of macro tricky question. But in some of the enterprises, especially in a regulated business like industry, the one of the challenges I think that we are needing to crack is that you have this technology know-how or AI divide, even within the company, that you have some part of the organization that are trying to figure this out and they are they have a responsibility to look into this, like you had. But at the same time, you have an ongoing business with the P core PL people, so to speak, who are maybe less interested in this because they're used busy doing surviving and making the business work. So you get this sort of on the one hand side, the bank is very ambitious, but on the other side, not ambitious enough in trying to interpret what it's doing. What do you think about that? This is a dictatonomy in the again, a challenge.

Stephan Erne 2:17:27

To be honest, I think you maybe you shouldn't go to the top-level people in the in the in the organization. No. Try to find uh somebody in this organization, uh, it doesn't matter in which hierarchy who is interested in that stuff and just help them and understand it, work together with them.

Henrik Göthberg 2:17:43

That's this is back to the community building, the evangelism building, and that is part of the design. If you want to innovate, you need to design for those arenas, maybe.

Anders Arpteg 2:17:53

Time is flying away here, and I would love to move a bit more into more philosophical topics. You know, after one beer and another, it's easier to start to speculate a bit more freebie, perhaps. Um, you mean? Even more. Even more, even more. Yes. Uh perhaps we need a corner computer to do it. But we we'll try anyway with our measly human brains to see what we can do. But of course, you know, AI is continuing to improve their functionality, and we're seeing AI being able to not only do more memory, you know, intensive tasks, but actually start to reason a bit and take actions more and more and start to hand a little more and more of the cognitive tasks and perhaps even physical tasks soon. What's your thinking here? What if you were to talk to your kids? You had a 19-year-old or yeah?

Stephan Erne 2:18:44

And two more, yes. And two more.

Anders Arpteg 2:18:47

Older or younger? Older. Older. Okay. What would you advise them to do? You know, what do you think about the future? How how what should we really ask people to go in to type of business? Easy question. Easy question.

Stephan Erne 2:19:05

Um yeah, I think um, I mean, of course, it's uh I probably need more than one beer to speculate about that, to be honest.

Anders Arpteg 2:19:13

But uh we can fix that.

Can Machines Deliver Real Empathy

Stephan Erne 2:19:17

I as a starting point, maybe you could uh I was thinking when we were creating ourselves uh these lists of on one PowerPoint slide where we were listing things that AI is good at and people are good at. And um, there was uh on on the AI side was usually crunching a lot of data and uh all this data, data focused stuff. Yeah, um, and on the other side was like creativity, empathy, this human to action, and so on. Um, and from from that oh from that perspective, of course, you should uh advise them to go into a human interaction business. But there was one phrase which I pretty soon started to uh to mistrust in this was about empathy. Because uh, in a lot of discussion, people always had uh, especially after ChatGPT and so popped up, and people were interacting and were astonished how good that works and what kind of stuff is popping up. They said, but empathy is a human thing. And uh then I uh started calling some call centers in in Sweden and in Germany, whatever, and uh I didn't feel a lot of empathy from those people. Exactly. From the humans, you mean? From the humans, yeah, because they didn't give a shit about my problem. They didn't help me, they obviously they didn't understand me this. So I I we started to think about is empathy really the thing for humans or not? Because when I was interacting with ChatG2B or Claude or whatever, yeah, at least I get the impression, and this they're very empathic.

Henrik Göthberg 2:20:56

They decide they're designed to listen in and follow you even.

Anders Arpteg 2:21:03

But but I think you know psychopathic is different from an empathic. And I think they actually are a bit empathic.

Stephan Erne 2:21:09

So and that's when we actually also started because I mean if you're a relationship bank, where the the basis is that you have people, you have educated people, you have hired people, you're educating people, you're incentiving people on being good people to your customers, and always make sure that this customer is um leaving this interaction more positively than he came in. So that's a lot about empathy. How do you even even if I would say the the word was not really used in that context, but I would say it's it's a lot about this empathic way of interacting with people that is um is one of the key cornerstones of Handelsbanken. And I would say, even still, if if I call a call call setup person from Handelsbanken, they somehow give me the feeling that they care about my problem. So and and then we were um looking at the way that uh chatbots uh even from from the beginning were interacting with this, and probably they were there they were trained to to do this, but uh um we we we actually did a lot of we were going into and say what what is actually empathy? Is it just that I I hear what you're saying or that I somehow give you some feedback? Ah, I experience the same stuff. So from from a more cognitive understanding for what what you're saying and trying to understand your feelings, mirroring maybe your feelings, all the way to would I at the end of the day be willing to act based on what I think your feelings are and the your problem is, which is a I would say a pretty big bandwidth because at the lowest end you can say I feel a pity for you or what happened to you. But yeah, I don't give a shit.

Henrik Göthberg 2:23:03

It doesn't change what I'm acting on.

Stephan Erne 2:23:05

All the way to I I leave everything behind and I'm coming to help you. And this um and this is of course also I think it gets even more um interesting if we go into agentic now, where chatbots are not only there to help you verbally, but to act.

Henrik Göthberg 2:23:27

Can they act with empathy? What does that mean in terms of agency to uh uh uh do another decision or action on a certain thing? Uh it's super interesting.

Stephan Erne 2:23:36

So and and and uh I think there are also there are some proof points, at least I think, why uh where we see that chatbots are obviously perceived as empathic. If if you look at some studies, what what are what is chat TPT used for? It's a lot about life coaching, finding meaning in life, all kinds of stuff where you if I mean if I read these studies, I I I always feel I feel very sorry for the people using the chatbots in that way because obviously they don't have humans to turn to. Which would make us even more concerned about this question: what is empathy really like? Are humans delivering empathy or not, or are maybe those chatbots better in delivering empathy in the interactions?

Henrik Göthberg 2:24:25

Yeah, but and and if you think about this now, maybe we can get a much better empathic one-to-one relationship in a unique setting. But if we want to do empathy at scale in a society, and we realize how many lonely people have where they don't get empathy the human way, so that so their empathy is zero in relation to how they act. And then and then empathy at scale becomes a different ball game. It's a very interesting angle on this. So maybe they're not maybe the ideal case is better in human to human, but if if the for many people there there is no human, there's no alternative, yeah.

Stephan Erne 2:25:12

And and then of course, uh like you said, I mean, that you can go all the way. How do you how do you what do you do with the elderly people that rather go into prison than staying at home? Because they have some some companionship. But but also to be honest, the the more maybe easier question, well, what do you do with your customer service? How do you want to put chatbots in your customer service and how do you make sure that these agents, or whatever you want to call them, are acting in the same empathic way like the people you had before?

Henrik Göthberg 2:25:43

I would say if you're a company who have very unempathic customer service agents, but you're not basically but but your spectrum conversation is so important then because to have empathy that I hear you, but I still rule the actions in without empathy, is that empathy? Or is that just you know, versus having empathetic agency where you can act differently based on an empathic feeling in how you use look at a loan or a credit or whatever. Of course.

Stephan Erne 2:26:14

And I mean there are some studies that show that if you if you if you call a call center, you have a problem, and the person at least gives the impression that he cares about what your problem is. He's pretending whatever I mean to I did everything I could to solve your problem, but I'm so so sorry I couldn't do that. You leave as a customer much more happy than if you have somebody who says, Well, that's not my problem, go somewhere else. I mean, at the end of the day, the result is the same. But the that's why I think it's a lot of perception. What is the perceived empathy that that you have as a human?

Anders Arpteg 2:26:51

I think I saw some study before, and and that was in the medical sector, and it's basically about the patient kind of interactions. Um and uh in that case, it they were forced to reply by text, like in email form or something, and they compared the perception from the patient um from both the humans and the AI, and and uh they basically felt that the empathic kind of response that they got from AI was significantly higher than the human. Now, what you could simply argue is that the humans simply didn't have the time to write this kind of lengthy reply, you know, describing how you feel, you understand them. I really you know you do this and understand that. And the humans were pressed on time, and they can't simply write text as fast as AI can, so therefore they weren't really fairly compared, perhaps. But still, it was the result that if you just have the textual domain, the AI were able to reply in a way that made the patients feel happier. And yes, of course, AI is able to do it faster, so and and it can at least phrase itself you know in a way that is perceived as being empathic.

Henrik Göthberg 2:28:08

So then but but I but I think this is a very, very good example because one of the key ideas I think empathy takes time, cost money, because empathy means you need to listen and you need to uh uh communicate in a way that makes it clear that you feel it. And if if the fundamental here is time, it means that they don't have time to be empathic. You know, in this sense now, with with with the speed of light, AI can be more empathic. I I I don't I don't I actually think what if you look at the hard fundamental of perception, then there is no excuse, it's just a a reason why the human they could have done more, they could have been more empathic, but they but they could have woulda should have their outcome was less empathic because they were they didn't have the time to be empathic. And to be honest, I think a lot of they just don't care.

Stephan Erne 2:29:16

If you call a call center agent who sits somewhere in the world, how much does he care about a suite calling and have a has a problem with this flight?

Henrik Göthberg 2:29:22

And and how to leave because that becomes they should care, but that requires leadership. Once again, time to be empathic. Time to be empathic by the person doing it or time to focus on empathy as a leadership steering.

Stephan Erne 2:29:40

Exactly. What does it mean? And and this is maybe we were talking about the kid dry self-driving cars and stuff like that. And there's I think there's some similarities because the the target is not to be 100% empathic. I think the target is only to be better than the humans, like like with the cars. I mean, I don't know how many people are dying by car crashes. It's a lot, at least. It's a lot, it's a lot, it's a lot, but obviously we have we think that's part of the life. And so and I think the the bar of um also and going also think about AI going forward, yeah, is there the bar that AI has to be 100% perfect all the time? That's what a lot of people expect. And they somehow forget that humans are not perfect either. They're not 100% empathic all the time, they're not taking care, they're not uh paying attention when they're driving cars all the time. They're drunk, whatever, all kinds of stuff is happening. But from a self-driving car, you expect that this should be 100% correct.

Henrik Göthberg 2:30:40

But but that does that tell us something on how we regulate or look at these topics in a more nuanced way again? Like the word is differentiated or nuanced, like that.

Stephan Erne 2:30:50

And responsibility. Yeah, because I mean in the human, as long as humans do mistakes, it's the human who's responsible. And we have a society or company system to take care of the humans that don't follow the rules.

Anders Arpteg 2:31:06

But coming back to your question here, and you know, okay, we can see that for some tasks, it likes actually being empathic and even driving a car or writing a reply to customer service case or even uh yeah, patient diagnosis. An AI potentially is better. What should you what will you advise your kids to do?

Stephan Erne 2:31:28

Come, you don't you didn't forget that one, yeah? No, sorry. My advice would be to to stay in an area where because my my and and this is of course based on my belief that for a while humans still wait still will take a lot of decisions. And um like I said before, I'm I don't think that humans are hundred percent open for fact or databased uh decisions, so they need people to sell it to them. So my my my advice would actually be to be to be in a in a position or in a role where you have to face humans. Okay because I think this need, and especially thinking about the the the long back hole of people who have haven't ever used chat GPD or stuff like that and don't know what don't really know what's going on here. And that's I think this this will continue for a long, long while. And if you don't if uh we as a society you don't want to leave those people behind totally, we have to take care of them. And that's why I think there's roles where you somehow have a technological understanding, but are interacting with people to help them take the right decisions. I think this is the spot where there will be a need for a very, very long time.

Henrik Göthberg 2:33:07

Yeah, you and you can extrapolate on that and think and and talk about roles where you need judgment, roles where you need coordination, roles where you need alignment, uh those to stitch things together.

Stephan Erne 2:33:23

Because I think taking decisions going forward will get harder and harder.

Henrik Göthberg 2:33:28

Yeah, so the decisions do is done operationally super fast, but someone needs to apply judgment, and that won't go away. It might take other uh actually might be more roles that it's judgment oriented and task oriented, if you think about it. So how do it develop develop your ability to lean into judgment and be able to train on a certain new topic in order to apply judgment? That's that that's a fundamental trait, I think.

Stephan Erne 2:33:58

Yeah, I like it. There's human empathy, and that's also I think because I think we will we will get delivered a lot of AI empathy going forward. But I hope that there will be a difference between the AI delivered and the human delivered empathy. And I think this is also one reason why why I believe this is a muscle you should train.

Henrik Göthberg 2:34:21

But I like the way we got into this, that we almost didn't start with empathy, and then you lean into judgment because I think judgment is more clear-cut that this will be around empathy. I don't think it's so clear-cut what is AI empathy versus human empathy. We have already good examples here that on the on the basic task level empathy is too costly, right? So maybe then human what is then human empathy that the machine can't do better, then we need to be nuanced again, right? It's a different style of empathy, which is maybe much more complex. I don't know. I haven't thought about it. I was like, because I think we have cases where this machine empathy is better already.

The Real Risk Is Bad Actors

Anders Arpteg 2:35:02

So thinking even longer ahead, and uh perhaps we can move to a point for your kids where we could see a very positive future. Perhaps if we think about you know, AGI actually do happen. I'm not sure. Do you have any thinking if AGI will happen or not? Or what do you think?

Stephan Erne 2:35:21

I think it's pretty for me personally, it's a very hard term to grasp. Yeah, what does it really mean to have AGI? Because I think in in some areas, I think AI already is superior than humans.

Anders Arpteg 2:35:37

So the so the but in general, if you take it literally, like being generally better than than humans, um, you know, like some ultimate definition of an AI should be better than an average level human coworker. And that means handling a lot of tasks in a very general sense. I think we're a bit far off, but still it could happen at some point. I mean, for one, you can think you can think about the two extremes here. Uh, one, of course, is the super positive one, and uh super positive could be that we move into a world of abundance, meaning AI have solved all the problems we have, it can cure cancer, it can fix the energy crisis, fix the climate crisis, and basically makes any kind of products and services cost go to zero. Meaning you don't really have to work. Your kids then if they choose to pursue a future in arch or gaming or whatever, they can choose to do so, and that's their choice. That could be a very positive potential future, or it could be another extreme, which is that the machines will try to kill us all. The Terminator, the Matrix, and uh we will have you know the Iranians and the Russians and the Chinese and perhaps the Americans and the rest of the world sending nukes on us and killing us all. Decided by AI potentially. Uh, and that could be kind of bad. Where do you think we will end up on this kind of spectrum, you know?

Stephan Erne 2:37:10

Um I'm actually more afraid that we end up in a in a in a third angle, and that is that uh pretty crazy people will use this technology. Okay, okay, please elaborate to uh to do very big harm or to keep their own power or to suppress people and so on. So um I'm not that afraid that AI somehow will take over and we come to this Terminator um future where they all by themselves somehow decide we should get rid of all the people. Um I'm more afraid of the um of the angle that some people because I think the potential of doing good or doing harm will increase exponentially going forward, because they are empowered by AI, yeah. And and I think the the the the the power of AI or the the potential what you could do with it will definitely increase. I think that that's what I'm very sure about. The question is only how will it be used? And uh based on the what I see is happening in the world right now, I'm not too confident that this power will end up in the hands of the white people.

Speaker 6 2:38:30

Right.

Henrik Göthberg 2:38:31

I think this is so in there's so many angles on this because what you're saying is we've said this before, you're more afraid on the path from artificial narrow intelligence or very powerful AI, but not AGI, being used in the wrong way by the wrong people, and the journey, you know, if we if we manage to get to AGI and then maybe the overlords can run us better. So this is the problem. The second dimension of that problem, and I I think this is a really real, this is the real threat to be discussed instead of a philosophical threat. Another key challenge of that kind of threat that you're talking about, it doesn't really matter if we have the AI Act or not. Because the people who want to do bad they don't fucking care at all, right? So it's it's a different understanding for how we deal with those threats than that we regulate. You can't regulate against that part anyway. How you do you agree with that? Yes, I agree.

Stephan Erne 2:39:33

And to be honest, I I don't think that the AI Act is a hinder at the moment for adoption of AI and companies because so they are so far away from that, anyhow. Yeah, but I agree. I I also think that the the threat is not fixed by regulation anyway.

Anders Arpteg 2:39:50

Super cool. And uh we I hope you can stay on for some more discussions after we turn off the cameras, uh Stefan. But it's been a true pleasure to have you here. And uh so many interesting topics, and uh I'm looking looking forward now to continue the discussion when we turn off the cameras. But before that, thank you so much for coming here to the AI Have to Work Podcast.

Stephan Erne 2:40:12

Yeah, thanks for having me, and uh really interesting to see your dynamic here also.

Henrik Göthberg 2:40:16

Stefan, yeah, I get I guess maybe not for the guests, but uh it's a good uh grade from us two hours and 40 minutes. We couldn't shut up talking to you. That's cool, man.

Stephan Erne 2:40:28

Oh, okay. Thank you for being with us. Thank you very much.