E113 - AI In public sector - David Magård and Patrick Eckemo

Don't miss out on Episode 113 of the AI After Work Podcast, a must-listen for anyone interested in the burgeoning role of AI in the public sector. This episode brings together two of Sweden's most influential AI figures, David Magård and Patrick Eckemo, for an insightful exploration of AI's impact on government and public services. David, a Senior Advisor at Bolagsverket and an AI policy expert, will delve into Sweden's stance on the upcoming AI Act and the creation of trust models for AI in public services. Patrick, Chief Digital and Innovation Officer at Bolagsverket and AI Swede of the year nominee in 2021, brings his deep expertise in digitization and digital transformation to the table, along with insights on AI and blockchain technologies in governance. Get ready for a riveting discussion on AI integration in public services, the latest news on the AI Act, and a glimpse into the future of AI policy and innovation in Sweden. Tune in for a journey through the evolving landscape of AI and its transformative potential in the public sector. 

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

Chapters

• 0:00 European Digital Identity Wallet Consortium
• 12:13 Digital Innovation and Company Registration
• 27:16 AI Strategy in Public Administration
• 40:09 The Challenges of Digitalization in Sweden
• 54:16 AI Commission and AI Act Updates
• 1:07:49 Guitar Drawing and AI Demonstrations
• 1:18:22 European AI Regulation and Negotiations
• 1:26:03 Regulating AI Usage
• 1:37:21 Regulatory Sandboxes in the EU AI Act
• 1:49:19 Investing in AI for Sweden's Future
• 2:05:57 Innovation Strategies and Technology Testing
• 2:10:24 AGI and Innovation in Public Sector

Transcript

European Digital Identity Wallet Consortium

David Magård 0:00

100% into the smart card Systems. Really, yeah, they're not on the forefront in this area. They have something else.

Henrik Göthberg 0:11

They have their work now.

David Magård 0:12

They are working. As it's a physical card that you bring with you, this is something that should be digital, so in your wallet, but in your, as an app in your phone, but it can also be online, so web flows.

Anders Arpteg 0:26

They can at least do elections immediately since 2005 or something.

David Magård 0:31

They can do, but it kind of comes down to the security level of that right. So you have the smart card. It ensures that you are at the highest possible security level when it comes to identifying you as a person. Of course, in Sweden we work now with Thank you dear, yeah, the assignment on developing the digital state EID. But Okay, so we have this regulation. It's under negotiation, but the commission also does. That is, they have an expert group consisting of experts from the different countries, member states, and then they have these large-scale pilots, which are consortias four of us actually.

David Magård 1:17

We are a coordinator of one of those that gathers a lot of participants that are to pilot the possible wallet, how it would look like, and then see kind of, oh what do we need to change? Is this working? Is it not working? And the consortia that we are coordinating have three main use cases. One of them is travel.

David Magård 1:41

So you travel, you need to have you travel data boarding card, passport information, maybe Check into a hotel and so on. Get the receipts through your digital wallet. Then there is a payment use case. So that's, of course, how you pay with your digital wallet. And then there's what we call ODI, organization of Digital Identities, so how you as an organization identify yourself and that's also very neat when it comes to business travel, of course or a manette of power you know full-makted Swedish but how you represent a company or organization. And also, of course, for organizations to speak to each other digitally and ensuring that you speak to the correct organization, but also a natural person to a legal person, like is this organization really the organization that they say it is? And the company registration comes in there, of course, because we hold the registers of the kind of confirmed companies in the different countries and you say we.

Anders Arpteg 2:42

I guess this is the.

David Magård 2:43

EU level. This is the EU level. Yes, and the whole of the is coordinating with you. It's coordinating one of these consortia and it's quite. It's 20 million euros for two years. We have about 80 participants. Some of them were bigs like Visa, bosch, sikpa, amadeus is a huge in travel industry. And then eight business registers. We have BankID from Sweden and Nets from Denmark. They have a similar solution that we have.

Henrik Göthberg 3:15

Another public domain type parties as well.

David Magård 3:19

Yeah, exactly.

Anders Arpteg 3:20

From different countries. And what's the goal of the consortium? Is it to like standardize everything?

David Magård 3:25

Not everything but to some degree. Yes, to standardize, but that of course happens in the standardization organization. But we provide them with information. So it's like this way of signing a credential works in this kind of flow From a technical point of view.

Patrick Eckemo 3:44

Also, of course, from a legal point of view. So specifications for that, and we run one of its four consortia in total.

David Magård 3:53

In Europe, in Europe.

Patrick Eckemo 3:55

And 20 million each of them.

David Magård 3:57

The biggest one is actually 80 million Are they all connected to the EWC consortia.

Henrik Göthberg 4:06

EWC consortia of one out of four. Yeah, yeah, so we're talking about the EU Digital Identity Wallet Consortium, shortened EWC, and that is one of the consortium out of four. And what are the other three? What are them?

David Magård 4:22

It's basically divided into use cases. So the big one is called potential. It's driven by mainly German and French authorities. Pretty much consists of authorities as well. Our consortia has a private public 50-50, but the other ones are driven by the public sector. They focus on driver license and some other things that come along with signing and signing into public services and so on. No bid it's Norwegian-led consortia Focus on payments as well, but in a kind of different angle than we do it. And then there's a Spanish-led consortia, dc Free EU, also a lot of Swedish and Finnish actors in that, and they focus on social security and education diplomas. So it's different angles. Our use cases are kind of Our idea with having travel and payments. Is that's something that you do quite often, payments every day pretty much. I travel quite a lot, at least in some countries, you know, go cross-border for working and so on, and that's good because that would probably drive adoption of the European digital currency wallet.

Anders Arpteg 5:37

But the wallet is basically what you are driving from Bologs, vaget and throughout.

David Magård 5:41

Europe, everyone is. I mean, that's the kind of purpose for everyone to develop the wallet. Then of course, the member states. So in our case Sweden will have the final kind of mandate on what wallet to use and who will be the authority to provide that, or if it should be come from the private sector, but on the mandate of the country.

Henrik Göthberg 6:04

So the ultimate goal here is to have, in a way, how should we put this digital movement in a secure way? And doing different things, so we can basically do the things that we do as individuals, as organizations across Europe.

David Magård 6:21

Exactly, and it comes. I mean, if you take even a step further back and you look on the kind of targets from the EU and the Commission, it goes into the digital inner market. So we need to move and we need to. If I'm in another country, I want to identify myself. I should be able to use the things that I use in Sweden, in Spain, and also, of course, online. And then there's also the digital sovereignty part of that. I mean you, of course, remember the discussions with Facebook and Meta, their coin, that they were supposed to provide.

David Magård 6:59

If you look at that in my mind this is my personal opinion that's actually an identity kind of scheme they're driving. They also want to have a second wallet, that, and kind of have your identity there. Of course the payment sector would be one thing. Then they kind of gather all the information as you and you. They are now the kind of provider of your digital identity and there are kind of an uprage from many countries in Europe about that, because identity is basically one of the founding factors of our state.

Henrik Göthberg 7:32

And now can we take our identities back to the individual in these things?

David Magård 7:37

And there's many kind of policy things in there, so it's a lot of focus on having the individual in control of their data as well. The whole.

Anders Arpteg 7:47

Meta movement into that coin. What's it called Libra or something Libra, I think it was first.

David Magård 7:54

Then it changed to something when they moved back to the US. I'm not sure what the name is now.

Henrik Göthberg 8:01

So in this consortium you now had an assembly last week, so you had one of the major meetups, or how we put it, 80.

David Magård 8:09

So I think it was 120 participants, of course, from all of Europe the commission, a lot of people from the states and Paris companies and we went basically through what we have done or achieved during these seven months that we've been active, and I think it's all kind of impressive actually, because it's a bit of a tricky situation Regulation isn't done, we don't have the architecture reference framework that we were expecting from the commission, so you had some gaps on the video.

David Magård 8:44

Yeah a lot of gaps that you can navigate Horrible many gaps. And then this completely new technology in some aspects that need to be integrated to existing technology and infrastructures Not everywhere, but we kind of scoped a lot. We see that everyone's agreeing on what are the fundamental pieces into that. So, for organization, what is an organization that differs between different countries? How do you identify an organization and what kind of semantics should we use here?

Anders Arpteg 9:17

What's the next step? When do you think we will have some kind of this under that?

David Magård 9:21

this Every member states. The relation is up to vote now, in January in the parliament. I think the council is a bit later, but when they're done, which they most likely will be, it's to 2026. Every citizen and organization should have the availability to have a digital identity vote.

Anders Arpteg 9:46

That sounds surprisingly fast.

David Magård 9:48

Yeah, it's very fast.

Patrick Eckemo 9:52

Well, let's hope that actually the plan adds in it's also very interesting when it comes to digital identity votes, when it comes to adding another level of documents and proof and things like that, Since today you need to go into each country's e-services to get that type of information. Here we put a common infrastructure in place which you can share that type of proof among each other using the same basic infrastructure underneath.

Henrik Göthberg 10:27

I can use to think about so many use cases where you reduce a lot of friction. I'm thinking about buying a house in Spain, I'm thinking about getting a job over there, or I'm thinking about using my insurance in a certain way, or whatever.

David Magård 10:41

Exactly. There should be a lot of benefits. We did some trials before from Bola Svartky, together with some Finnish and also some Norwegian participants. That's why I don't know if the reasons that we did the consortium because we had actual experience from before. We can see that we can automate a lot of things If you have the wallet and you have these attestations that you called. This is basically secured data. It's something that you can assign and trustworthy and you can automate a lot of things. You can also ensure that there's a focus point for the individual. You get the attestations on many different things from different authorities. You don't have to go back and ask them all the time, because the actual proof of whatever would be in your wallet instead. Then you can share it to the others.

Henrik Göthberg 11:35

Sounds like a good thing to strive for, but with that, let's introduce the guests.

Anders Arpteg 11:41

So very welcome here. Thank you, I'm a big ecomo and David Maghord. I'm not sure how to say your last name in English, but Magarde is usually good for.

Henrik Göthberg 11:50

Magarde.

Anders Arpteg 11:53

I know you, patrick, for a number of years I'm not sure how long, really, but like five years or something. But I think you are very famous for partly the 140 billion crown article from DIG.

Henrik Göthberg 12:05

I think when you were presenting on stage, when you were still at Posten.

Patrick Eckemo 12:09

Yeah, I used to work for Postenward some years back.

Anders Arpteg 12:13

But now working as the let's see if I get this right the chief digital and innovation officer at the Swedish companies and registration office at the Bullock Circuit. That's correct. Did I get that right? Yeah, awesome. So many things to speak more about, so looking forward to that. But also, david, you're a senior advisor at the same company as Swedish companies and registration office at Bullock Circuit. What is your title? Strategist.

Henrik Göthberg 12:38

Yeah, it's really a senior advisor it just sounds a bit off.

David Magård 12:43

Senior advisor. Yeah, government of 40, public authority.

Anders Arpteg 12:47

I would say then the coordinator of the EU Digital Bullock Consortium that we just spoke about. But also been part of the Swedish team working with the AI Act, something that is very newsworthy, and we probably will speak more about that as well.

Henrik Göthberg 13:01

Going back into your work at the Registration Council to work on the directives from AI. At the Registration Council we had a guest, Johan Howard, here in the first season and I think you all more or less was you worked on the same stuff after him.

David Magård 13:18

Yeah, pretty much.

Anders Arpteg 13:20

So a lot of knowledge at this table and a lot of movement forward for the AI Act as well.

Henrik Göthberg 13:28

There's a very timely podcast today, guests, so we're happy for that.

Patrick Eckemo 13:34

And I can just add that I used to work for the agency for digital government and by that, at the same time as David was working for the Hans Kansley.

Henrik Göthberg 13:44

Yeah, so we were very tightly by that. Yeah, he told me about that. He was receiving your. I'm not going to go into details, but I heard that the report, yeah, the report Many reports, but the report, the report as well. We will, of course, going to talk about the report, but, patrick, why don't you give us your short breakdown? Who are you as a person and how did you end up with your how?

Patrick Eckemo 14:11

would you?

Henrik Göthberg 14:12

describe yourself.

Patrick Eckemo 14:13

That's a good question. I grew up basically at SAS Institute for 11 years, and so I spent most of my time as a consultant and developer and for a couple of years in analytics, BI and advanced analytics helping clients both in Gothenburg and Stockholm and even working the global teams in the end of my time at SAS.

Patrick Eckemo 14:43

When I left SAS in the end I can't remember 2008 or something like that I was manager or director for pre-sales, technical, pre-sales and technical strategy so deeply involved with the customers and helping them out with the right architecture and type of use, make the best use of analytics and SAS, of course, and trying to help them to increase their maturity in the area.

Patrick Eckemo 15:13

And of that, I moved into post-Nord in the margin between Sweden and Denmark. Really interesting journey. A lot of really really good people both in Denmark and Sweden working with analytics and I tried to build a team around analytics at post-Nord. So try to combine people working with SAP, for example, and SAS and other type of business objects by that time and the more advanced analytics people as well and architecture as well. So I worked with like a certain enterprise information architect by then and the director for the center of excellence for analytics and the information management at post-Nord. So I did that for six years and then I moved into the agency for digital government where I've been at in the executive team for some time and a senior advisor as well and I worked a lot with official government assignments, like the report you just mentioned and very interesting agency working at the EU level and the national level, and then I moved to the post-Nord.

Anders Arpteg 16:28

That was seen as Swedish as well, right.

Patrick Eckemo 16:30

Yeah, yeah, I've been in Sonsval since I left post-Nord.

Henrik Göthberg 16:33

Because SAS and post-Nord, that was your Stockholm career. And then you said, let's go back.

Patrick Eckemo 16:41

Yeah, I moved back, since I have a cottage up in Jämtland.

Henrik Göthberg 16:46

Tell me the truth. It was the skiing, right, yeah, skiing of course.

Anders Arpteg 16:52

And then you moved to Borlaxverket. Yeah, how long ago was that?

Patrick Eckemo 16:56

It was two years ago, and the major reason why I started at Borlaxverket is actually that I, everything I do, I do it actually for Sweden, it's not for a specific agency. So I go every time, every day I'm going to work, I do it for Sweden. I think, and I think Borlaxverket is a medium-sized agency where we can do things for real, I think. So my vision is to make Borlaxverket as a role model how to digitalize agencies in Sweden and as a use case we can do like this and to save money, increase efficiencies and productivity, make use of new technologies and try to pivoting the business model we have today.

Anders Arpteg 17:45

So that's my purpose Sounds great and I think we really need that, but if you were for people that don't know really what Borlaxverket do, how would you describe their?

Patrick Eckemo 17:54

business. We manage company information in Sweden basically. So we are more or less a broker. All the companies need to register information in our databases, of course, like annual reports and things like that. If you are going to start a company, you need to contact us, of course, for names and things like that. Everything ends up in our registers after decisions that, for example, can have that name and so on, and then we sell the information. So we get information and then we sell information. So we are in an information broker.

Henrik Göthberg 18:32

Could you make that into a profitable department?

Patrick Eckemo 18:39

We are financed by fees, basically, but we are looking into more, of course. Since EU is focusing a lot on open data and open up the data resources within EU, it's a huge amount of benefits by doing that. I think it's up to 300 billion euros a year or something like that. I think it's something like that. So it's huge value. But at the same time, we have a lock-in, since we have our finance model connected to our businesses, so we need to operate Otherwise. We need to have money from the government in another way.

Anders Arpteg 19:19

Who are the main customers, so to speak, of?

Patrick Eckemo 19:21

the information. It's a lot of different types of companies. We have everything from ERPs to system vendors using the data, but we also have other types of companies trying to enrich the data because it's part of their services, so they combine it with a lot of other data that they sell to other parties as well.

Henrik Göthberg 19:44

So in some ways Dun and Brad Street and the likes become an example of a type company that enriches data.

Patrick Eckemo 19:52

And the data that comes from us is trustworthy, of course.

Henrik Göthberg 19:56

I'm awesome. And then let's move to David. So what's your story?

David Magård 20:09

Well, I'm a lawyer by training. I started at Swedish Land Redistry and, speaking of AI, there I was quite soon into digitalization and worked actually not in the legal department, but in the digitalization team innovation teams as a legal counsel, and we did some things, so we automated decision making in some parts you did that early. Yeah, we did it quite early, I think this is 2014 or something.

Henrik Göthberg 20:43

So this is something that we look back at. That is quite early.

David Magård 20:47

Yeah, I think we were one of the first agencies to actually do that.

Henrik Göthberg 20:50

What type of decisions was it.

David Magård 20:53

It's kind of when you take up an mortgage on your house. It's a very, very easy decision, but still you know you need to do a lot of technical work, some legal work, so on. So I worked with that and also kind of easy unit, so digital identity now then and so on. And also we had the IBM Watson chatbot there, a small rabbit hole. What was that?

Henrik Göthberg 21:17

all about.

David Magård 21:20

It's a fine experience to kind of look into and kind of have an idea of what the future would look like with the chatbots and digital assistants. That's now kind of the main topic. What were you?

Henrik Göthberg 21:33

trying to do? Was it like a chatbot?

David Magård 21:35

Yeah, yeah, chatbots to help customers or I mean citizens when they navigate our web page, basically to provide information so that they didn't need to call in, or so it worked alright, but I mean, this was some time ago, so today we wouldn't see it, it's almost like a different life. Yeah, it's not many years, but it's a different life. Then I worked at the Swedish Ministry of Justice and Finance and Infrastructure with basically new technology. What was my kind of portfolio? So AI, but also blockchain and some other things digital identity, of course.

Henrik Göthberg 22:20

And what was your angle?

David Magård 22:22

My angle was what the politicians wanted Good answer but of course I helped out in kind of informing what's happening and maybe this is something that we can do and this is something that we maybe should get.

Anders Arpteg 22:38

I have to ask as well. Speaking about blockchain, I have my opinions about that. But if we just go back to the European Wallet consortium, is blockchain a part of that strategy?

David Magård 22:49

It's not a big part. The EU kind of wants us to have inclusive blockchain. It might be interesting in a very small layers of the kind of infrastructure and that is basically key management and to make that kind of easier to understand. That's, if you need to manage your public and private keys which is how you sign and get to the data should that be at a central authority or organization or should that be decentralized? If you want to go for decentralization, you might use blockchains, might and of course, there's a definition here. So it's a blockchain or it's a distributed latest technology. It's not a blockchain, it's a DLT, if you go for that definition. So that's what we're kind of looking at and that could be interesting also for trust registers. So who can actually be part of the ecosystem without going through only one central authority?

Patrick Eckemo 23:47

Yeah, but it's the ledger idea here we're really talking about, so we're not confusing and we tried that when it comes to our own proof of business Together with Finland, so we had a blockchain in the Like for BULLOX verket.

Henrik Göthberg 24:00

Yeah, and so the whole idea then is because you're not having a logical central authority. Because if you had that, you would probably put it there.

David Magård 24:11

Yeah, but you need to kind of consider that this is the EU right? So who would that be?

Anders Arpteg 24:16

That's a good point, isn't it still BULLOX verket, the central place of management? No, not really.

David Magård 24:20

I mean, we could be that some parts, especially when it comes to like authentic source, but then when the data is kind of out of our repository, then how do we manage that and how do people keep track of their data? Then if we want to have it like their data, and how can others trust it and you know it's, you have to have other actually being part of the chain as well, but I really like the topic here, that we are talking about what is actual problem and use case, and that is what we are focusing on and not technology A or B.

David Magård 24:57

Yeah, technology. I mean there's some benefits and some downsides of any technology to use. And, like what's being explored at the EU level with blockchain is basically the idea that, okay, we don't want to have one central authority that keeps track of everything, so how do we do it? And maybe we don't trust each other to 100% between the member states and different organizations, so how do we do that in a technical way?

Patrick Eckemo 25:23

We can do it in a legal way, but then and then blockchain can be one of the use cases.

Anders Arpteg 25:29

Can we believe that will be the case or do you think another solution will?

David Magård 25:32

if you were to make a prediction, I think it will be part of the future ecosystem in Europe. I think.

Anders Arpteg 25:39

I hope so, but I don't believe so.

David Magård 25:42

But then again, it's not like a blockchain, that's like Bitcoin or Ethereum.

Anders Arpteg 25:48

It's a hybrid version where it's why should there be part of it? And if you can't mine anything, because it's mandated. You force people to do it.

David Magård 25:57

Yeah, we force the foretists to do it.

Henrik Göthberg 25:59

This is another angle, but the government right.

David Magård 26:04

No, but it's something that you need to consider.

Henrik Göthberg 26:07

You're jumping from here to like the regency council, yet what's the English?

David Magård 26:10

word Government offices.

Henrik Göthberg 26:12

Government offices. What was from an AI perspective? What do you think was your main achievements during your time there?

David Magård 26:19

Main achievements mine, I mean it's a collaborative and it's a politicians' achievements, but I think we were, when I was working there, we sent out some assignments to dig and some others so that they could provide with a lot of information like what is out there, what do we need to do, what would be the benefits, and so on. So I think we did some interesting stuff there together. The trust model that was developed and still is under development is one part of that, so it's a model for how you are transparent when you use AI in the public sector. So that's part of it. Of course, the numbers here. Personally, I think I would have hoped that we would have come further and also been a bit more prepared for AI Act, but you know it's.

Henrik Göthberg 27:16

But when you're three years, kansli if I'm doing a little bit like a fly on the wall scenario you know what was the discussion on, what was the main topics that was during your period. What were you trying to achieve? We need to know better, we need to have reports coming in. How was that sort of conversation going?

David Magård 27:40

I mean, the political idea was basically to work on with the Swedish AI strategy which is not really a strategy but kind of a direction and to implement that, and that said a lot of things. So we should have common infrastructures, we should be more involved in the EU work, we should share data in a way that's suitable for AI systems, and so on. That was the kind of main focus areas around the EU.

Anders Arpteg 28:14

And then you came to Bologsverket, somehow, yeah.

David Magård 28:18

What did that happen? Yeah, there was GD. Gold Some people maybe, and there was a job opening there that sounded interesting, so, yeah, I applied for that.

Anders Arpteg 28:31

And if you were to just very, very briefly describe your role?

David Magård 28:35

Yeah, I'm a senior advisor. Basically, I do things as I work with this consortia, so that's more tangible, but I also provide, I take part in expert roles. I'm the Swedish expert in the European Commission's working group on AI regulatory sandboxes and also in European blockchain parties.

Anders Arpteg 28:59

Could you add that to the list? Yes, yes, yes, I was thinking the same.

David Magård 29:05

And I take part in some kind of regulatory work, and then I do some kind of foresighting as well and help out with some strategic things Foresighting, foresighting, onverse analys, onverse analys.

Henrik Göthberg 29:16

The Swedish word is better.

David Magård 29:18

Yeah, it's much better. That's the only word that I've been able to find. But you know, to see like, okay, what's happening and mostly focused on the EU level for my purposes, of course and then kind of take that back to our organization, say like I think this is happening and I think this is where we should go and when you work with.

Henrik Göthberg 29:36

As I understand it, as a senior advisor, you're looking into the future. Are we three years, five years, ten years? What is your kind of frame when you talk strategy?

David Magård 29:50

I would say three to five maybe, and Patrick is ten.

Henrik Göthberg 29:56

Ten years. Yeah he's, he wants to be. At least he's up in the air. I still have my feet on the ground here.

David Magård 30:03

Yeah, exactly, I mean, of course, the management level sets the kind of this is where we want to go, and then I try to see like, okay, but this will happen. I think this will happen when it comes to regulation and this will happen when it comes to policy and tech. Of course I'm a lawyer, so I wouldn't be. I wouldn't want to go too deep in that, but I can ask people, but in this strategy role now because, as you highlighted, you were counseling and there was a technology dimension.

Henrik Göthberg 30:35

How big is when you look at three to five years ahead strategy? Where are we going to? How will a bull-logs market look like in the future? How big portion of your work would you sort of roughly estimate this technology innovation? I would say quite a bit of that.

David Magård 30:54

I would say digitalization, because that's broader than technology. But one thing that me and Patrick has been speaking about a lot internally but also externally, is that the main bulk of what we're doing is digitalization.

Henrik Göthberg 31:10

Because that's what's going to change.

David Magård 31:14

Yeah, that's how we become more efficient, that's how we prepare services, that's how we connect with the ecosystem, the companies and so on. And then the other part is that Digitalization and kind of ensuring that we have something to offer in the future, because that's not clear, Then I flip the question what of your strategic work does not have a connection to digitalization right here, or technology?

David Magård 31:49

I'm not sure, maybe Patrick. But I will say also one thing. I lost a bit here, but it's digitalization and also that EU is the main driver of all the regulations basically in our domain and mostly for Sweden as well, which makes us we try to have more kind of outreaching activities than many of my assumption as Mission, many public authorities that they kind of look internally and then to maybe the government offices, but we kind of look at EU quite a lot, because you need to look broader to understand where you will end up as a company, as an organization, as a government organization, of course, but of course the government decides.

David Magård 32:34

But a lot of the regulations come from EU. It comes both on subject matter, so how companies should be defined, but also on data.

Anders Arpteg 32:48

I think we should move to. We have a short. We have to finish before seven, I guess. But okay, let's move to the next question, perhaps the infamous or the famous, sorry, the famous.

Henrik Göthberg 32:59

Infamous Dig article when did it turn infamous. I love that, by the way.

Anders Arpteg 33:06

Yeah, but it's been spoken about and has so much media attention and, I think, for good reasons.

Henrik Göthberg 33:12

For very good reasons.

Anders Arpteg 33:13

And it's super interesting to hear. But basically, one of the core messages at least, is we can make substantial savings benefits. I'm not sure what have we framed it if we made the proper use of AI, but please, if you can please, give us some description of the origin and how that report.

Henrik Göthberg 33:31

What was the name of the actual report?

Patrick Eckemo 33:33

Promote AI within the public administration to make the best use of it actually, and how did it get started? It started with a lot of dialogues, with Rian's conflict around the benefit that we need to do something to start promoting AI within the public service administration, and we've been talking about open data for some time and working on the data layers, but not that much in analytics and AI. So after that, we received this government assignment and we're really happy for it. We spent a lot of time in this report, both when it comes to making analysis of what we are doing in Sweden right now, but also comparing with what other countries are doing in AI right now.

Anders Arpteg 34:26

And it started in 2020? Yeah, I think in 2019, 2020.

Patrick Eckemo 34:31

I think it was finished in 2019, right yeah, something like that Maybe published, maybe in the beginning of January or 2020, something I can't remember. So we made most of the job. Work was done by 2018-2019. So we went to a couple of other countries, like the US and Canada and all that, and met government agencies over there as well to have a discussion on how they use using AI, what type of strategies and thinking and financing steering from the government and so forth. So that ended up in some recommendations that we can, which is basically in line, I would say, with a guideline from 2018. Yeah, johan Horvath, what about?

Henrik Göthberg 35:23

In my opinion and correct me if I'm wrong. This document has sort of, in my opinion, some of the more deeper, and maybe do we have an AI strategy in Sweden or not, I'm not sure, but at least now for the public sector, from an AI point of view, I think this is one of the most comprehensive, or the most comprehensive report that sort of looks at this that you can almost take as a TDIG document, because it has clear recommendations.

Henrik Göthberg 35:53

It has a dollar value or Swedish kronor value and it has clear recommendations on where to put your money, so to speak.

Patrick Eckemo 36:03

Yeah, and that's what I mean. We wanted to estimate the value of using AI, since it's very hard to just make, come up with recommendations, do that, do that, but what I mean we want to make some money of it as well, saving for some higher values for the citizens and companies, and that's one of the reasons why we looked around in Sweden for use cases. We couldn't find in it Actually, it was very few of them by then and we even looked in Europe and I was part of the GRC and Europe by then AI Watch as part of.

Patrick Eckemo 36:41

I can export from the government by then and when we tried to analyze use cases within Europe. But that wasn't enough either. So we turn to McKinsey, since they come up with a report for a couple years earlier when it comes to the famous 30 billion or trillion. Yeah, they came up with a report global report and for each country as well, and based from then they had a huge database of use cases globally. So we had a discussion about is it possible to use that their use cases and try to see as a reference.

Patrick Eckemo 37:22

I mean because we didn't have any use cases otherwise and we went through them. A couple of hundreds of use cases I think roughly around 100, was applicable for public administration in Sweden and that's the base. We tried to scale it from that in full usage within the public administration Simplify some use cases that you looked at.

Patrick Eckemo 37:47

For example, like in planning, public administration, a lot of planning of course activities. That would one part how it could when it comes to budgeting or something like that, or for preventing things like in pipelines maintenance and things like that. So we don't wait until they break down In healthcare prevention, of course. Huge value in that.

Anders Arpteg 38:15

Or it actually has some preventive care in some way, exactly.

Patrick Eckemo 38:20

So a lot of different use cases. That builds up that, the number, the number, the famous number. We come up with 140 billion.

Anders Arpteg 38:28

What was the majority of the contributions to the number 140? Was it healthcare or was it?

Patrick Eckemo 38:34

Actually it was. I brought some papers with me here. It's actually pretty fairly distributed, I would say, across domains. We used the co-fog standard model to distribute it over a couple of different types of sectors and we could clearly see that, when it comes to common processes within the public administration, it's a huge value. So we had the highest value in that perspective.

David Magård 39:03

And common processes would be like HR and.

Patrick Eckemo 39:09

It would be HR, but also, like in production, we have island hunting, errands, management and things like that, and we can clearly see now moving into generative AI. It's probably much higher value than we came up with by then. So if we look at across the main healthcare, education, labor market, how to optimize the fit between matchmaking and social security Because we have a lot of frauds and wrong payments in the systems and how to prohibit that type of stuff, so in the end, to sum it up, increased productivity, higher revenue and less fraud and wrong payments.

The Challenges of Digitalization in Sweden

Henrik Göthberg 39:55

Those were the value levers, so to speak, and how was it received, in your opinion, at the time when youbecause I think there was quite a lot of bugs around it, because I was not. This was something that we in the community, we knew about this and we read the report and all that, but what was your view on this from the inside, I think?

Patrick Eckemo 40:19

it's been fairly referenced and discussed and all that. The main problem I would say in Sweden is actually we don't have any debate about digitalization at all, Since it's notno parties actually sees as their own area or topic, so they never talk about it because it's not an issue.

Anders Arpteg 40:46

If we have that big potential by digitalization using AI? Why wasn't it more? Because it was veryI think you underestimate the impact it had in media and other places, but we would love to see some more actions being taken given this right, yeah, but at the same time you don't win elections, or?

Patrick Eckemo 41:08

talk about digitalization, you win elections, talking about more teachers, more nurses and things like what we've been talking about for a hundred years, and that's the problem. We're still talking about all the recipes to solve today's problems, and this is a means to an end. I mean, digitalization is just a means to an end.

Henrik Göthberg 41:25

Okay, I think you're spot on and we cannot get into the political debate. We have never really seen them argue for the election on the digitalization topics, and they should, and I think you could position yourself. I think they are making a miscalculation if you do it in a good way. But why aren't we doing that? What is your? Is it too obvious and simple? What is missing to make this part of the debate?

Patrick Eckemo 41:55

I think we are. That's just my personal opinion. I think we are trying to solve symptoms in the system all the time, and by I mean, we can clearly see we need more people in the healthcare, we need more people in the schools. So we solve it by more teachers, more doctors. But the problem is we still have the same way of working as we have been doing for 100 years, no differences at all. So when it comes to digitalization, we are really really good at optimizing the current business model.

Anders Arpteg 42:31

Yeah.

Patrick Eckemo 42:32

Not trying to change the way we work, and if we're going to make use of the full potential of digitalization, including AI, we should start first looking at how to do the right things. How to reinvent First and then optimize the things we're doing.

Henrik Göthberg 42:51

Yeah, do the right things. Yeah, I think it's so common for non-technical people.

Anders Arpteg 42:58

They see a problem. We need better healthcare and better education and whatnot. And what is the normal solution? Well, you add more people to it and for technical people you at least can get some discussion on. Perhaps we can have tech that actually helps if make that process more efficient. So that is super hard to get through and the traditional solution is just to add more people to it.

Patrick Eckemo 43:21

And in Sweden we have an autonomous, I would say, system. We have an extremely stable public administration one of the best in the world, I would say, and it serves well the last couple of hundred years, but it's optimized for specific tasks. And digitalization is a horizontal thing and today we need to start. We've been always looking from inside, outside, delivering our own services, our own customers in our own channels as I speak, instead of looking at from who are we going to serve?

Patrick Eckemo 43:56

Citizens, companies and so forth, and they want to see this journey. It should be as easy as possible, without any hassle, and we should try to remove all the unnecessary demand on the public administration, but we're still focusing on each organization and even if the politicians really want to change something, it's really, really hard, since it's like self playing piano.

Anders Arpteg 44:22

It's like going horizontal instead of going vertical. Yes, you're trying to optimize the vertical, but in reality, if you did optimize the horizontal which is hard to speak about because you can't really be that concrete but if you did focus on the horizontal improvement, you would gain benefits in all the vertical at the same time Exactly.

Patrick Eckemo 44:42

And just as an example here, in the latest report we brought about AI and one of the things we came up with was we are badly needing a common infrastructure for AI within the public administration, since it's a huge value by having common resources instead of everyone is trying to do it their best. So if we look at just as an example for savings on the pure technology part, not the value creation so if the four larger institutions in the public administration would try to establish the basic AI infrastructure in place, it would roughly cost around half a billion, something like that, and if it depends on how you calculate, it could be 300 million as well, but something between 300 and 500 million. And if you do it centrally, it costs 50 million. But the challenge here is, if you do it centrally, you need to do it from the ends, can sleep and you need to do it. They need to do it. If they don't do it, then the public administration need to manage it anyway and then it will cost roughly 500 millions.

Patrick Eckemo 46:07

And so a proactive decision is 50 million, non proactive decision is 500 million. Then we need to manage it. But the challenge here is then just the four bigger public administration, like agencies and maybe some regions and municipalities. They will use AI as a means to enhance their operations. All the rest will be left out. Basically, I would say, of course they will have it included in the Microsoft tools and all that like common, but they won't be able to use it by themselves. So that's the reason why we push so much for the common infrastructure.

Henrik Göthberg 46:51

But the root problem here, some of the root problems here, in my opinion, exactly the same as you go to the corporate sector. So we are talking about the fundamentals of how we are steered, organized and how the mandates look like. And in order to fix some of this stuff, you need to fix the organization, you need to fix the mandates, you need to fix the steering and, if you're a little bit technical, we can talk about Conway's law or doing an inverse Conway maneuver. I mean, like in software engineering, we learn since the 60s that the systems you design seems to follow the organizational communication structures of your organizations. So ultimately, an inverse Conway maneuver is basically how do we architect the organization that allows us to then be able to budget and steer good systems? Do you agree with this kind of logic? Because I think you can't fix this if you're not at the same time fixing the organizational mandates. And it needs to fit. They need to be cohesive to two dimensions.

Patrick Eckemo 47:59

I mean you need to change. I mean I would say this is a really challenge when it comes to the governance model we have in the public administration. I mean, it's been for so long and everyone's like don't touch it. So even if we have politicians that clearly want to do something and I met a lot of them it's very hard for them to do it Because of these things.

Patrick Eckemo 48:23

Yes, and since they have a lot of burning issues, I mean we have increased our, I mean we need to invest a lot of money in the military capabilities right now, and so on. So how can we motivate to steer money?

Anders Arpteg 48:40

to this area. I think Bulexvar can be a role model here in trying to build up something that can be reused throughout different public sector domains.

Patrick Eckemo 48:49

I really hope so and I think we have a really, I would say our vision is I wouldn't say that we are going to take a self-altered loop. We are talking about information that should flow freely within the society, especially when it comes to company data. So how are we going to move, transform our business to that, to make possible by using AI and new technologies and things like that. So and but it's not that easy to say we are going to use new tech In the. I would say it's more of a question how should we change the way we work as we do today in the future? So we're looking very much into change to how we operate our business in like five, ten years.

Henrik Göthberg 49:45

And then you can kind of take that conversational question to the macro level. That's the hard question how should we operate?

Patrick Eckemo 49:53

Sweden.

Henrik Göthberg 49:54

How should we operate the departments?

Anders Arpteg 49:56

ultimately I said big one- yes, I'm eager to go to the friendly initiative etc. But maybe we should.

Henrik Göthberg 50:07

Let's go there. I mean like so, if you wrap up the dig report we got into rabbit hole. But what was the after play a couple of years later?

Patrick Eckemo 50:16

It's the after place where we came up with some more reports connecting to it. So one is trying new technologies and Laatmati Riet was in lead of that report. We came up with the trust model, which is really interesting and David is really deep into it and, as you can talk about you, can take.

Henrik Göthberg 50:37

The trust model is a good topic.

Patrick Eckemo 50:39

Yeah, we can dig into that, I think. And then we came up with new report, also when we dig in even further than when it comes to the trust model, but also an AI guide for the public administration and some recommendation for common.

Anders Arpteg 50:55

So this was another government assignment based? Yes, what was the name in English Is it called?

David Magård 51:02

Promote AI in public administration. It's almost the same as the last, yeah.

Henrik Göthberg 51:07

If I'm a little bit cynical. Now to move to present day, like a week ago, two weeks ago, we have now a new AI commission, that by swanberry. Is it the same old, same old. Are we adding value now or are you chewing on this? I'm not to be a little bit.

David Magård 51:24

Yeah, so my personal opinion is that we pretty much do. I mean, we, patrick and I, have been involved in some of the reports, but there have been many reports from different agencies, from AI again an AI, a gender of AI for Sweden, because it's companies as well and academia.

Patrick Eckemo 51:42

S&B, for example, been over one S&B, so Sturzson.

David Magård 51:45

Yeah, a lot of them. Everyone kind of says the same things, which I think they can only do strategy for so long. Yeah, they kind of. They kind of took that into the Commission's directive quite neatly, and it also has the possibility to give proposals on regulation, which is different this time.

Henrik Göthberg 52:05

That's a good thing.

David Magård 52:07

Yeah, but of course in I think, for many of us who have been working in this for quite some years, we do expect that many of the outcomes, most of the outcomes will be something that's already written to a large extent, and that, I think, is a bit sad.

Henrik Göthberg 52:25

Can we help it? Can we? Can we? I mean like so they have the way, an opportunity to shape. You know what they do, so how can we help?

Anders Arpteg 52:32

Let's just start with just describing what it is. So we have two different assignments here. One is the offently AI, to promote AI in public sector. Then we have the AI Commission. Can someone just give some background, describe what the purpose is, and something.

David Magård 52:49

It's a bit different things, but the promote of the AI in public administration that that's something that's been reported. It's finished. There were some expectations that it would kind of go on. We'll see about that. Still, there is this beta portal called offentlyse. We're all of the things that we delivered are, so you can watch them there.

David Magård 53:19

Use cases, guide, trust models and so on and I hope what still is, I think, to kind of develop that, because we see that a lot from the reports that's already written, that people need guidance. They need each guidance and they need kind of then they need to go to the deeper levels, but they need these guys guidance and examples at first to kind of start out.

Anders Arpteg 53:40

I think it's funny. We had another big assignment called the speech AI agenda and they basically at least what I tried to contribute to was actually a use case library basically which you basically did in a friendly AI, exactly. I think what you said is very true. You know you redo a lot of things from the past.

David Magård 53:59

But the reason why that was part of the assignment was because of the agenda and others who already? Said like we need this. Okay, who will do it?

Henrik Göthberg 54:09

let's we need to put it in this assignment, so it's actually a positive action. Yeah, it has a connection.

Anders Arpteg 54:15

Definitely.

David Magård 54:16

The air commission, then it's something that was announced a few days ago just by the. I wish you'd taken on the news. Yeah, we were already in the news. Okay, yeah, but so it's in Swedish. It's U-training, so it will have the kind of normal boundaries, which is that someone who's the chair and there's a secretary who will write a report and several reports maybe, and then there's some experts into this. They come from different parts, so they come from WASP, the Wallenberg autonomous program.

Anders Arpteg 54:59

AI system program.

David Magård 55:00

Exactly From academia and from private sector, of course, as well.

Henrik Göthberg 55:08

We have some guests from the podcast on the yeah, yes, sweden is there as well.

David Magård 55:17

Yeah, so a lot of people there and the idea is that they would provide provide suggestions to the government on how Sweden should work with the AI, and it's segmented in four columns, basically. So how should we act? In kind of international ecosystem, so EU mostly, but also the other, ocd and so on. It's a focus on public sector. It's focused on education and on private sector.

Henrik Göthberg 55:49

This is a slightly different tweak or angle on it compared to before. Is it similar to what we've done before?

David Magård 55:55

I would say it's very similar to the Swedish strategy, but the Swedish AS didn't have a report before it, really, so it's a bit different. When it comes to the assignment, the assignments that we were part of, that, they are, of course, directed to the public administration, and this is a bigger, broader commission, so that that's, that's something at least.

Patrick Eckemo 56:22

And the good thing, I think, is that it may be like an hopefully an endpoint of the analysis. So maybe we they, since it's a broad spectrum of good people within the commission, so hopefully it can end up with like a summarized summarized of all the things we've been doing for a long time now and add some of the new aspects and dimensions David mentioned we have a huge development with the AI Act and Generty AI, that those are two major new drivers.

Henrik Göthberg 56:52

I would argue that that makes this kind of worthwhile. Now, patrick and David, I ask you if you could dream what would be the added value that they could do in this commission. Now, put an endpoint, I think, was a really nice way of putting it. What else? How would you frame what the added value can really be at this point?

David Magård 57:13

If you say endpoints, I would say targets that are financed.

Henrik Göthberg 57:19

Financed targets. I think there's a big point that is missing.

Patrick Eckemo 57:22

Another thing, except financing, is really, really important. You can argue that we have money in the system today, but you need to steer the money so and so steering governance is a huge part of what will be some type of success within this area, since we are not governing AI at all within Sweden and we need to do that. So hopefully they will end up in both governance and financing.

Henrik Göthberg 57:54

I love it because, to do a report, this is the big pink elephant in the room right. So either it's a report like you have done before, with has not so many teeth, to be honest or you have the dimensions of governance and finance, and then you have something that is very different to what we have before I would argue and a regulation to some parts.

David Magård 58:16

So it's okay, there are probably not that many obstacles when it comes to legislation. Actually, there are a lot of reports on that as well in Sweden, of course there is. You can argue that you know we have different public administrations that sit on a lot of data in silos, but you have also have to remember that that it's the way it is for purpose. You know, because if we gather all the data, who actually owns the data?

Henrik Göthberg 58:42

then yeah, you pretty much have a you know 1984 state right, but I think Patrick's summary is profound because if you can add regulation, governance, like organization views and finance to whatever we have done before, that is what is missing A clear target in these dimensions, that is, that someone can vote on to be honest, and can you also combine it with, like some type of missions?

Patrick Eckemo 59:10

We have a lot of like societal challenges. That needs to be solved really badly, and if we can have some of those in place and using digitalization and AS means to solve these issues, that would be really, really good.

Henrik Göthberg 59:24

So and then it's, then it's completely worthwhile because we haven't been able to sort of tie the knot on these topics and this could tie the knot. But these are the things. If you don't do them, I don't see so much added value, to be honest, because we have done great work. But that's a really good summary.

Patrick Eckemo 59:42

And we see a lot of other comparable countries like eminent neighbors and even in Europe Norway came out with like a billion yeah, they're spending a lot of money and we can see I mean, I was, I mean it's a long time. They spent five point five billions and hopefully that the government should add a. I mean meet it somehow, like we did before in Sweden when we built Sweden Hopefully we'll see that more in the future as well that we can together both in the government and the business, and the public administration can work together.

Henrik Göthberg 1:00:17

Actually, I have a topic that we can take on more philosophical level. If it was your money, how would you say, let's take that when you have one more beer?

Anders Arpteg 1:00:27

If it were your money? Should we move to the AI act kind of thing, or should we be?

Henrik Göthberg 1:00:41

it's time for AI news very cheesy AI.

Anders Arpteg 1:00:46

It's very good.

Henrik Göthberg 1:00:47

Do it again, do it again, do it again.

Anders Arpteg 1:00:56

So in the middle of the podcast we have a short break or not too short, we'll see speaking about some recent news that have occurred in the last week, or something that is of personal interest for all of us.

Henrik Göthberg 1:01:08

I mean, the AI commission was actually weekly news, but we there was too long, so we covered as a topic good.

Anders Arpteg 1:01:15

Yes, yes. So okay, it's been happening a lot of stuff in in our last week, as usual. Anyone that wants to go first speaking about, you know, the follow-ups from Geni my or the AI act, which I think, of course, is super interesting as well, why don't we start with the AI act angle and I'm looking at David.

David Magård 1:01:36

Is that news, or should we take it afterwards?

Henrik Göthberg 1:01:47

This is news, the AI news, will now come, not in actual, but in updragging.

David Magård 1:01:53

We can say that there's a provisional agreement on the act.

Henrik Göthberg 1:01:56

Okay, that was the head of news later. Okay, good one, well saved.

Anders Arpteg 1:02:03

Okay, anyone else should we go?

Henrik Göthberg 1:02:08

I'm gonna, I'm gonna, I'm gonna persevere to take topics that is not LLM or Genitive AI oriented. I'm gonna push this, but I don't want to start. I want to start with the. You know what is the other news? That Mistral released.

David Magård 1:02:24

Mistral exactly.

Anders Arpteg 1:02:27

Let's go there.

David Magård 1:02:29

Yeah, that's basically what I know, because they released it as a, as a as a torrent and just put it up on on X and I think it was well as like here, here you go, guys, which, which is quite interesting and also very different from German AI when it comes to kind of marketing.

David Magård 1:02:46

I, for one, was very I was struck I was struck by lighting when I saw the video from from Google on Gemini, but then I really understood that they all did a lot of cutting and editing in that one, so let's continue with the mixture.

Anders Arpteg 1:03:03

I think it's interesting from a technical point of view in many ways.

Anders Arpteg 1:03:07

So, mistral, you know they released another, you know 7B model, 7 billion model, before that was really amazing and it could beat the performance of a lot of other much bigger models as well, and it came from France, european model. That's awesome and very great to see. Now they released another mixture of expert model, the MixTroll, which is eight models at once and and this is, as at least speculated what also GPT4 is. So GPT4, they say, is also a mixture of expert model, that is, potentially eight or sixteen different models that have different experts in them. And now finally and I've been waiting for this for some time when will the first mixture of expert open source model come out? And MixTroll was the first one to do it. So I think this is amazing, and I think this, of course, is the future to have this kind of not as simplistic as just more transformer blocks, but actually having a set of experts where each expert focus on different topics and in that way it's much faster to train and, especially, it's much faster to do inference on.

Henrik Göthberg 1:04:17

And the energy consumption is, and the whole consumption of how many, how you use the whole neural network to have what you need to light up is a very different architecture. Yes, so could you elaborate a little bit? I was trying to read up to understand the differences, but it's quite cool yeah.

Anders Arpteg 1:04:33

So I mean basically each token as it's go through the network. You know this kind of autoregressive it takes one token at a time and when it has one token it takes the next token. But depending on the token it goes to different paths in this huge network of in this case, not seven billion but, like forty ish billion, forty five billion I think it was, but it doesn't go through every parameter.

Anders Arpteg 1:04:56

So this kind of traditional dense models have to take all, use all the parameters for every token. Now it can just take like a one eighth of it. It's not really true, because it's not every part of the transformer block that is duplicated is just the feed forward part, which is the one that has the most parameter in it. So that is the most important one, but still a part of each block. As it goes from layer to layer to layer, as all these kind of deep learning networks do, they are routed depending on the token and that is much more efficient to do, more efficient to train and, especially, much more efficient to influence on.

Henrik Göthberg 1:05:34

So the interesting thing is there's not only there's a new open source model from Europe is actually it's a first in building. Expert models like this, like the architecture, make sure expert has been around since 2014.

Anders Arpteg 1:05:45

So actually, I think it was Ilya Citruskever that is still, you know, driving open AI at first wrote a paper on this.

Henrik Göthberg 1:05:53

I think, in 2004. No, no, but to put it out in the wild, so to speak.

Anders Arpteg 1:05:58

Yeah, the big open source one. And, and of course, ilya is now working in open AI and probably GPT-4 is a mixture of experts, since he first published about it, to my knowledge. So it's super cool that we actually have an open source version that is in potentially the same type of projection.

Henrik Göthberg 1:06:19

It's not the bigger what the build 7B. It's 8 times 7B. It's a slightly different architecture. It's cool.

Anders Arpteg 1:06:25

It's bigger, but still faster, smarter, bigger and faster and more efficient.

Henrik Göthberg 1:06:30

Okay, that's, that's one news. Then we almost slipped into the Gemini news and the Gemini news. We talked about Gemini last week, but there is a twist now. It was a little bit faked. What did you hear, david?

David Magård 1:06:45

No, it's so on. On following the news and saw that they actually edited that quite extensively. They haven't shown, as far as I know, how much it is, but it wasn't that real time interaction.

Henrik Göthberg 1:07:04

We were blown away by the video. We were blown away by the video. I texted Patrick like what is this?

David Magård 1:07:11

Now? We're on a GI, but yeah, it's a bit disappointed, of course, but as far as I understand it, anders, you're the real expert here, probably, but still a better model than Chachi Petit.

Anders Arpteg 1:07:28

In some sense we can speak about the MMLU benchmark, which is to be heavily discussed and questioned, but which is the one that they beat on potentially. But leave that aside. I think you know I feel cheated, I feel annoyed because we spoke last week about Gemini and I spoke specifically about.

Anders Arpteg 1:07:49

They are just drawing an image of a guitar and now they have an audio, and then they were writing or drawing other stuff and suddenly they have an electric guitar and suddenly they have drums and it was fake. It looked like it was real time. Now it did produce the things, but instead of what the video suggests, which is that it's just drawing and it's in real time doing stuff, it was not at all in real time and they produced like text prompts as well, and not just drawings. And they had this kind of other demo which was this kind of duck which they drew.

Anders Arpteg 1:08:22

Exactly dead ones. Yeah and that looked like it in real time as you just added some blue paint on it, but that was also like not in real time and they had, you know, textual prompts.

Henrik Göthberg 1:08:34

Why would you do that at this point in time? And Google has done this before. You know if you do it.

Anders Arpteg 1:08:39

Duplex thing. You know it's a standard.

Henrik Göthberg 1:08:43

We all got to know about it and it's used to destroy trust.

Anders Arpteg 1:08:47

You know, I think this I know one person.

Henrik Göthberg 1:08:50

It destroys trust if you get caught.

Anders Arpteg 1:08:53

But they get caught, they get caught and they got caught before with the duplex thing, you know they get caught, steve Jobs?

Henrik Göthberg 1:08:57

He didn't get caught.

Anders Arpteg 1:08:59

Yeah, well, no one you know who likes Apple, I do. But anyway, we have a person you know, a personal friend you know, that worked with Gem and I and he's a very good guy. That I know and I'm sure. I'm sure he wouldn't say this, but I speak in his words when I say that I think he's very annoyed with the marketing department of Google. I think they would never, as a scientist, in deep mind, produce a demo like this, but then they have some kind of need to be superior to OpenAI and then they have the marketing working with the scientists and developers in a way that I think is questionable and I don't think internally in Google they are really happy with.

Henrik Göthberg 1:09:44

The benchmarks are impressive and we should, but we let us. Now is very back to OK. Let's see when we get it and see what it's going to look like in real life.

Anders Arpteg 1:09:54

It's still super impressive with Gem and I. They didn't have to fake it. I think I'm going to be disappointed. I feel a bit treated, at least. What are you about you?

Patrick Eckemo 1:10:01

Yeah, I think so as well. I mean, they don't have to no they don't have to, you don't have to.

Henrik Göthberg 1:10:07

Do you have any news?

Patrick Eckemo 1:10:08

Yeah, I just look into why today, and as so one of the noticed on Neko, the full Buddhist car, full body body scan which Spotify founder came up with as one of the probably maybe one of the new unicorns from Sweden, seems to be very, very promising and popular service when you can go into for healthcare?

Anders Arpteg 1:10:34

Yeah.

Patrick Eckemo 1:10:34

So it's more like you can go in, have a full body scan, scan with use, make use of AI to prevent being sick basically could be skin diseases or whatever and I think it's an interesting use case in general when it comes to prevention. And I see it in other parts of the world as well, for one year ago actually, when they use been using it for the same purpose, but more for diabetes, and with very successful results.

Henrik Göthberg 1:11:07

And we talked about this before the pod. I mean like the whole angle of prevention. I mean, like you, we work in the public sector. The compounding effects of value and saving money is probably very hard to even calculate. It's massive. Of course.

Patrick Eckemo 1:11:24

Prevention is the way to go where we can, of course, I mean just when it comes to wrong payments within the government. It's, it's tens of billions each year, and hopefully we can reduce that by the new agency. When it comes to payments, it's about like a core, but also using new technologies like AI. So, yeah, yeah, interesting what this is going with.

Henrik Göthberg 1:11:46

Daniel Lake has been quite successful so far. He's behind it.

Anders Arpteg 1:11:52

I'm sure they have some good tech and motivation, and he certainly has a network to make success, so I think it will be successful. I actually put myself on my waiting list and I got the time, yeah, but I didn't have the courage to go through with it. Why didn't you do that?

Henrik Göthberg 1:12:06

Because you were unhealthy, right Again, some weight, yes.

Anders Arpteg 1:12:09

All right.

Henrik Göthberg 1:12:16

One more, one news from me, and I'm not going to talk to you in a TV AI. Instead, I'm going to talk about Tesla Optimus Gen 2. And could we have some clips so we can have a look at it? So let's start with the video. It's very smooth, so it's the generation.

Patrick Eckemo 1:12:36

Optimus. This is the first one. This is from the first generation one.

Henrik Göthberg 1:12:41

So this is not the new one.

Anders Arpteg 1:12:42

This is March. Yeah, but we'll see the fingertips.

Patrick Eckemo 1:12:49

So Tesla, you know, is not only building cars, but also building a car.

Anders Arpteg 1:12:52

Tesla, you know, is not only building cars, they're building a humanoid. This is what the humanoid look like. It's not the Terminator, but it is a humanoid robot.

Henrik Göthberg 1:13:02

They have improved the neck movements, one of the key things, and they've improved the next generation of the hands as well. So to the actuated neck. And you know, the first time they had it on stage, it couldn't walk right so it's a bit of a shaky and then you know the really remarkable thing is the speed that Tesla has done this. If you take sort of Boston Dynamics, I mean like they use slightly different, I think, technology in terms of AI.

Anders Arpteg 1:13:39

Boston Dynamics does not use machine learning.

Henrik Göthberg 1:13:41

This is what I mean. They use a different technology, but they've been at the game for 10, 15 plus years. There's no one who kind of knows they don't take. And you know when did they start? 2021?. I mean like, because what we're looking at here picking up an egg, it is really quite cool, right, and they are just getting started and look wait for us to wait for the last bit? The dancers, of course.

David Magård 1:14:12

How much editor this is.

Henrik Göthberg 1:14:14

Oh, no, no, no, this was the funny thing In the press release explicit no editing, no editing. They made explicit in the press release to make sure this is real. He loves to bash other people. Yeah, he loves that was a direct bash for sure how much it cost to buy it. What do you mean so far? But you can buy it.

Patrick Eckemo 1:14:37

You can buy it?

Henrik Göthberg 1:14:38

Yeah, but it's going to come. Yeah. What is the first? What's the realest price? It's like a car. It's like a car. I'm going to buy one too. Man, for what purpose? I'm just going to have it like can't eat it all. Another beer.

David Magård 1:14:53

Okay, yeah, why not? This is cool for conferences.

Henrik Göthberg 1:14:57

When you come, you pick up all the best from this guy.

David Magård 1:14:59

Perfect. But I mean, what's the idea of this? Do they have a special purpose?

Anders Arpteg 1:15:05

I can't speculate, but you know, I think for one it's surprisingly often many. Okay, let me backtrack a bit AI today. For one, what Chatbot is doing is basically just a perception. What they want to add is planning and reasoning. This is not really what ChatBT has today, but they want to do it. Potentially, the Q-Star thing that OpenAI is doing is that thing.

Anders Arpteg 1:15:27

But they need another third thing, which is control. So if you take a self-driving car, that is not general, it's narrow, but still it has perception through the cameras, it has planning, and now with version 12, that's done through deep learning as well. And they have control, and also now in version 12, it isn't released yet, but also be done in a machine learning way. So they have all the three components to do something that is truly autonomous as well and acting in the real world, and that is something very few AI systems actually do. So what we hope and I think Ellen is thinking with this is that optimus, when it has the humanoid form which the rest of society is all optimized for, can do so many more things in an automated way. And I think initially it will be just manufacturing, because so many things there is optimized for having a humanoid doing things, but that can then be extended to whatever kind of task you want, and the humanoid form is surprisingly general and useful.

Henrik Göthberg 1:16:27

So the bottom line is to go in and do tasks where you can then install a robot and automate without changing the environment. So much so if you think about, amazon has done an amazing job to robotize, put AI in their warehouses, but they need to build from scratch like a warehouse that is optimized for how to have robots in them. So how can you put humanoid now at places where you don't need to change the environment? You can simply put them where the human was. That is the general idea. And then you come into manufacturing. You come into if I take my old employer Wattenfall, we have many tasks where cleaning inside the nuclear power plant. So you're doing high risks, high unsafe stuff. So the bottom line with the humanoid is to be able to put a robot without changing the environment. I think that's the bottom line use case.

Anders Arpteg 1:17:26

Should we move to the next topic?

Henrik Göthberg 1:17:31

I think this is the end of news. We end with a bang. Optimus is cool.

Anders Arpteg 1:17:38

According to Elon, that will have a bigger revenue impact than the cars potentially will.

Henrik Göthberg 1:17:45

We don't know, anyway, but I think that's the news section, and then let's go for the next topic.

Anders Arpteg 1:17:50

Yes, and it was a big news happening to AI Act. But perhaps before we go to the AI Act, David, you've been part of this and you were, as I understand it, a part of the Swedish team developing the AI Act. Can you just describe with how you get started with that and what you worked with?

David Magård 1:18:07

at that time. Sure, so the Act, as any EU regulations come from the European Commission.

Anders Arpteg 1:18:14

they propose that was in 2021 or something.

European AI Regulation and Negotiations

David Magård 1:18:20

Yeah, that seems about right. I'm horrible with years and numbers. Small watch brothers.

Henrik Göthberg 1:18:27

Exactly.

David Magård 1:18:29

First thing that goes out right numbers. But then of course, every member state have their own government offices and it's the people from the government, the officers, who negotiate the regulations on behalf of the government. I was leading the Swedish group there in the first round, and then, when I left over to the Swedish company's registration office, I also helped out as some kind of experts to the team Are you still in that?

David Magård 1:19:03

No, so that's a disclaimer as well. So I haven't been part of the trilogues, which is the last part here, so I don't know anything more than anyone else, basically, so I can't say anything. That's secret in that respect. And just to explain the trilogues and EU how it? Works. You have the.

Henrik Göthberg 1:19:20

Commission right the.

Anders Arpteg 1:19:21

Council and the Parliament right, and they all have to come with their versions of the.

David Magård 1:19:27

Yeah, and basically the Commission has already come with their version and then the Parliament and the Council negotiate from their versions and then they try to strike a deal, and that's what we see now and that's also a disclaimer for this. So no one, pretty much, has seen the actual provisional agreement. So provisional is the agreement that will be then up to vote in the Parliament and the Council. That can actually be changed, it doesn't happen often.

Anders Arpteg 1:19:55

It hasn't been written, to my knowledge either. It's just an agreement on a superficial scale, right?

David Magård 1:20:00

No, it has.

Anders Arpteg 1:20:01

It's a little nutcase.

David Magård 1:20:03

It's been from experience. It has been written in many details but they have to kind of ensure that it's correctly from what they also agreed to and their details you know in language and everything. So they have to map out. Then they will release normally the provisional agreement so it's open to everyone, and then they will take that and vote on it in the Council and the Parliament.

David Magård 1:20:24

And so if you look at when can the votes the earliest happen, there are two votes we're talking about still here right, yeah, council and Parliament, and it should be like Q1 next year, both of them Most likely, but it could take longer. So anyone who kind of says like this, I know everything about the Act and you know they don't. And it's always like this with, of course, legislation there was in the detail. We don't have detail to date, but of course you know, we do know a lot and we know what it kind of came in with and also some kind of official reports on what they agreed upon. So we know some things that's there that wasn't there from the start.

Henrik Göthberg 1:21:12

How does the negotiations and the mangling for hours and then days right? I mean, like I heard like the first mangle, or like ridiculous 30 hours and then they slept and then need some more mangling and this is what they've been doing that for three days or four days, Three days as far as I see.

Henrik Göthberg 1:21:27

And then they came out and said so you know what? In my opinion, at the end they had a couple of beers setting in sitting in the bar. All right, if I do this this is the final negotiation is a napkin. I promise you and we will never know about that, by the way, because now you know you need to write the napkin- out. That's what you were saying that was my joke.

Anders Arpteg 1:21:46

I mean to just move into the news. Then I mean for one.

David Magård 1:21:50

It's been work in progress for two years at least Even more, because actually the that's one part, I think, people, if you want to understand the regulation, there was actually an high level expert group. I think it started in 2018 on the mandate of the commission, where they asked a lot of expert for clients from Sweden, one of them on like what is AI, how do you define that?

Anders Arpteg 1:22:18

And they changed it in the AI Act.

David Magård 1:22:19

They changed it? Yes, definitely, they changed many times and they also asked like what are the risks with the AI, how could we see the future with AI, and so on, and those things actually went into the legislation or the proposalization. So if you read the report from the AI high level expert group, that's a very good way to kind of understand the regulation, because the regulation comes from a perspective of securing safety for humans, but also securing human rights in the EU. Of course, there are a lot of we can have a lot of discussions on, like innovation and all of these things, but you have to remember that that's like the core of this.

Anders Arpteg 1:23:04

The first one was just to focus on the ethical guidelines, basically Exactly.

David Magård 1:23:08

But ethical guidelines. If you look at them they're like how can you show that there's trustworthy AI? So it's like transparency, there needs to be some kind of governance structure, there needs to be sanctions. You know, all of these things are kind of there, and then you kind of interpret that into a legal text with some additions and so on.

Anders Arpteg 1:23:30

And then if we just move to the 36 hour over three days kind of discussions Before that, you know, just like three weeks ago something France, germany and Italy came out. They basically said we don't like AI act. And you know France and Germany, just have. You know the big mistral language model and Germany had their alpha.

Henrik Göthberg 1:23:50

The little yeah, the little yeah, the little investment.

Anders Arpteg 1:23:55

And of course they're going to post it because it would basically kill at least I understand it the open source movement of these.

Henrik Göthberg 1:24:01

A risk at least.

David Magård 1:24:03

I think they were, as far as I understood it and when I read what they said. So you have to remember this, and this is the tricky part, of course, with regulating technology, which I would argue that you cannot do here.

David Magård 1:24:18

Of course the commission said yeah, they said, like we're not regulating technology, we're regulating risks, right, but you know, that's my personal opinion can you regulate technology to a large extent? Okay, in the first, in the proposal from the commission, there's nothing about foundation models Nothing. And actually, when the mandate from the council was done, three days after that, open AI really and everyone's like, oh. So in the council there is this group that's called General Purpose AI, which is could be defined yeah, the European Parliament, they write about foundation models. So it's a bit of a different angle to it, but it's basically the same thing and it is this large language model.

Anders Arpteg 1:25:10

But isn't the General Purpose, ai focused on the system and foundation and models focused on just the model? Am I mistaken there?

David Magård 1:25:18

I think that's one of the kind of they will in the details differences between the European Parliament and the council, and they can emerge that, as far as I understand it, in the Indian, so in the final hopefully final version Now, as has been leaked, of course and what they write in the official memo from the parliament is that they also regulate now the General Purpose AI, because, as far as I understand it, they merge them and then they have General Purpose AI that should have some transparency demands on that, and then there's General Purpose AI with a systemic risk that has to have more demands on them when it comes to releasing them on the market.

Anders Arpteg 1:26:03

Because otherwise, I think you know, if you just take Chatti BT as an example, I mean it has a model underneath it, for example, but then they have a system with this Chatti BT on top of it. If you do release a system which is ready for end user usage, that's a very different thing, I think, than just producing the model. And I think if you do produce a model that is ready for use and can be abused as well directly in a very easy way which Chatti BT potentially can, but Chatti BT4 perhaps not I think it should be potentially regulated differently.

David Magård 1:26:37

And so I agree. But then you have to kind of step back to how this regulation is actually built up and you have to think about it as like a period. I mean they speak about the risk-based assessment, so it's focused on risk for violating human rights and, of course, security risks, so that could be more like something blows up or something like that, more of that kind of risks, and what to say is like okay, think of a pyramid. So these are some risks. That's unacceptable for using AI systems, of course. So it's AI systems could be some other system potentially. Ai systems can't be used for these purposes.

David Magård 1:27:21

The most discussed area here is the social credit score. That's prohibited now with AI, which will be most likely prohibited. And then there are some other targets or areas, which is called high risk. So that's high risk for, and if you use an AI system in those areas, then you need to have a certificate or some kind of controlled mechanism to ensure that you abide by some rules. Rules are on taking with specifications, proper data use, tools to ensure human insight, oversight so that's human in the loop, and some other things. Transparency that's basically like transparently, because you have to document.

Anders Arpteg 1:28:15

What happens if you ask OpenAI to document GPT4?

David Magård 1:28:19

That's really interesting what happens if you ask Google to do it.

Anders Arpteg 1:28:22

What happens if you ask Microsoft to do it? Will they do it? I don't know. We'll see. It's their IP, right.

David Magård 1:28:28

But they don't have to share all of this, of course. High risk AI systems I would say that Church PT is not a high risk AI system, because it's just a. You can do whatever you want with it, right when it goes over to something that's in the high risk areas and use it for that purpose then you have a high risk.

Anders Arpteg 1:28:48

Please tell me how to manipulate the election in US.

David Magård 1:28:52

Yeah.

Anders Arpteg 1:28:53

Isn't that potentially nothing? Or in France or in EU.

David Magård 1:28:57

Coming there, then you use it for high risk, but of course then it's a regulation, so then they will target the user of that.

Henrik Göthberg 1:29:07

Not normally.

David Magård 1:29:08

Really, but of course there are demands on the transparency of the model and also they have to report to some degree to the commission and or the AI board. How that would look like, we're not sure. And also if and we not know this as well but if it's an AI, if it's a foundation model or a proper AI with systemic risk. So we will see how that looks like, because the definition isn't out.

Anders Arpteg 1:29:34

It's just strange though. I mean to linger on this question is Chattibit T, a medium risk, which I believe they said it would be potentially and not a high risk, but it would be so easy to abuse it for really, really malicious purposes. Wouldn't it and, if nothing else, to manipulate people's opinions, which is what they said, would be not even high risk. It would be unacceptable, exactly.

David Magård 1:29:59

Depends on.

Anders Arpteg 1:30:01

And how can then Chattibit T, if you use it for that, be considered medium?

David Magård 1:30:06

Because it's the usage of something that kind of I mean, this is where we want, but it's regulating the technology, the Chattibots.

Henrik Göthberg 1:30:16

This is the argument that we have discussed on this pod many many times. Like, you want to regulate the use of the hammer, not the hammer itself, right, and how the fuck do you do that?

David Magård 1:30:28

It's very difficult in this case Exactly, and I think it's going to be very interesting to see when it goes over from one area to another no risk or high risk and when it kind of I might be afraid that they will take the approach of GDPR, which is basically let's write a terms of service agreement and once you click accept, then you can do whatever you want with the data.

Anders Arpteg 1:30:54

Now, if you have a terms of service agreement with Chatti BT and says you are not allowed to use this to build bombs or to use this military purposes or use it to manipulate elections, and then you press accept and then it's fine, that would not be very satisfying in my view.

Patrick Eckemo 1:31:13

There's a difference. I think the one thing is integrity in your data that's one part of it.

Patrick Eckemo 1:31:18

The other thing is the usage of it and what you're going to do with it, which is more like the AI Act, and I think that is more difficult, since I think we, in the end, we will probably see a lot of discussions and movement when it comes to how to regulate the usage of it, because I mean, probably, I would say they will end up with some type of regulations that they can't give the service as it is to open.

Patrick Eckemo 1:31:53

Otherwise it would be too easy to manipulate. Whatever it is you can use it for.

Anders Arpteg 1:32:00

But then it should be the usage that this regulator and not a technique right.

David Magård 1:32:06

Yeah, and the idea is that it's a usage that should be.

Anders Arpteg 1:32:11

How can you do that if you have a system as its general purpose? And then you say okay, the system is now deployed. Should you say that this person does something wrong, but not the? I mean you can't regulate the users, you have to regulate the providers of the system, right?

David Magård 1:32:29

Both yeah. And you also have to, of course, remember that there are other regulations in place here. So if you try to manipulate, you have normal laws the elections.

David Magård 1:32:42

There are certainly regulations on that as well. Then you can get fines from the Act. But what I understand that they try to do with this is okay. So the usage something here that we don't want to see. But we also want to have some regulations on the providers of the AS system. So we're sure, do we know what's coming out there? Do they understand that something's happening out here? Then we can also say to them guys, you need to limit something here, and so on. So it is the usage. Of course, the system in itself to some degree gets in the frame of the regulation.

Henrik Göthberg 1:33:23

Because I think this is a good segue. I want to steal the angle a little bit now Because I think there's a pink element in the room now, so we can bicker about the regulation itself and it will have pros and cons and you will have your ass backwards either way you turn. The angle I want to take now, the only way we can sort of now judge if this is going to be a good or bad regulation, in my opinion, is going to be about what is the type of support package that we now put together to actually execute on the regulation. And this is where I think the scary part comes in and where I think we can learn a lot from GDPR.

Henrik Göthberg 1:34:05

So GDPR I was at Vattenfall at the time. We had no problems dealing with that. We had our general counsel, we had 20 lawyers and we hired five more lawyers and we sort of trained the whole company Boom, and we actually knew how to maneuver very, very fast. I was in B2B and we had B2C. I was very, very close to action on this and we knew how we could work with legitimate reason. So we didn't have to go down in the R you need to have consent for everything because we could argue that in order to operate your electricity bill, you would need your information Boom. So the point we're trying to make now is that there was very, very little support in Sweden at the time in terms of how to act, and the lawyers and the management consultants made a killing on it.

Henrik Göthberg 1:34:58

And so I argue you know what is the packages we want to see now, what needs to happen in Sweden, in EU, to basically make regulation. How can we make that?

Anders Arpteg 1:35:14

useful and adoptable, or you can give me for the sandboxes kind of question.

Henrik Göthberg 1:35:20

No, we can lead into that. We can lead into that because that's a big, big part of that, Because I think there's more and I used to top it off. This is used to really nail it down. If I contrast Biden's bill, what he came up with, the executive order. I mean Thank you for correcting executive order, I was looking for the word, so I actually forgot. His executive order to me is a full package. I mean like so he's highlighting regulations and stuff like that, but it's also highlighting, you know, putting money, investment. We're going to open up this commission, this department, we're going to do this, we're going to do this and this and this and that. So it's a 360 package in that executive order. And right now I'm a little bit looking at the pie and there's only one pie piece. What is the other pies? What is it that we need? And we can now talk about regulatory sandbox.

David Magård 1:36:15

But there's more. I think it's more, and I think I mean in the act itself. Of course there are regulatory sandboxes, those things. We can say something about that, but if you speak about the EU level, there's actually quite a substantial number of financing programs. Digital is one of them that has a lot of investments into different kind of supportive actions when it comes to AI.

Henrik Göthberg 1:36:45

No one talks much about this. No one talks about this.

David Magård 1:36:48

And they already. There are already things existing out there called the European Digital Innovation Hubs, the TEFs, the testing experiment facilities, which is actually kind of big investments, but I don't know, it doesn't seem that these things come out so much and also, I think, from, but it's also because we live in Sweden and Sweden is, by any report you can read about this, we are one of the worst in actually attracting money from the.

Anders Arpteg 1:37:21

EU. We give a lot of money but we don't take it.

David Magård 1:37:23

We give a lot, we take nothing.

Patrick Eckemo 1:37:25

But even be impractical at the European level.

Henrik Göthberg 1:37:28

Well, if we have these, things, we need to know about it and we need to make it accessible to the startups. I mean, like it's the startups that or it's the normal companies that need to figure this out.

David Magård 1:37:37

And there is actually a fund out there for a kind of venture fund set up at the commission that aims to provide help for funding of SMEs in AI.

Anders Arpteg 1:37:52

Applying for EU funding. Do you know I?

David Magård 1:37:56

know, but it is an easy in the EU as well, I would say. But you know, it's not. I wouldn't say it's like the best thing to do, but it's still there, it exists, so it's not only regulation. I mean, if you look at the Swedish level, then I would argue that there's nothing else pretty much Coming from, like a central wasp of course is there, but otherwise there's pretty much no.

Anders Arpteg 1:38:21

We're always doing some things.

David Magård 1:38:22

Yeah, a few things.

Henrik Göthberg 1:38:24

But I would argue, if we go back to the other topic, you know what, if the commission could come out with that, we need to invest heavily to make it super accessible and easy to follow the EU Act. So we need to have a very, very strong department on this and we need to have a very, very strong digital service on this and we need to digitize like Bullock's very so you just moved to the regulatory sandbox.

Anders Arpteg 1:38:48

What is that really and what are they trying to?

David Magård 1:38:51

do. Yeah, you can start with what Henry kind of went into there, because it's actually in the act that the the CIPL-8 was here again. But most likely there will be new agencies that should support the ecosystem. So it's and this is one of the first we see this in the EU Act so like the new government agencies that support this ecosystem in this aspect.

David Magård 1:39:16

Of course there's. You know they have two years to kind of and they're already, you know, of course, some movements on that and they have written in that you know these agencies should have the right competence and right financing and so on. But we'll see about that. The regulatory sandbox system, that's part of kind of the work of those agencies. So an agency that has the mandate from the member states to provide help when it comes to the Act, but also when it comes to actually putting an A-assist to the market. So it actually we will see but from the mandate from the council at least, is test, develop and operate and ensuring compliance, ensuring that being compliant with the Act. So it's more than just the kind of regulatory things. It's also aiming at providing some kind of sandbox for helping you to develop your system?

Anders Arpteg 1:40:21

What is the regulatory sandbox?

David Magård 1:40:22

And then it's regulatory sandbox. It's an environment, for a certain amount of time, where you should do these things. You have come in with your A-assist and, and each country basically has to set up their own right. We will see, but it will be. It looks like there's a mandate of at least one per member state. Could be cooperative between the more.

Anders Arpteg 1:40:48

Could do it in Sweden, do you think?

David Magård 1:40:52

A new agency, I think, but it could be IME, it is. Maybe they will definitely play a role in this because, of course, there's data into this and there is also one of the. The idea with the regulatory sandboxes is that you should be able to use data in a way that it was collected for.

Patrick Eckemo 1:41:13

So it's a different purpose than the original usage.

David Magård 1:41:16

Yeah. So it's kind of going around GDPR to some extent, because it's like I have this data, I have this idea of this A-assist, but I can't use this data because it was collected for this purpose in this A-assist. Okay, but here you have a controlled environment with government officials where you are able to test this if it works and if it's in line with the AI Act and so on. But this is the idea, of course.

Anders Arpteg 1:41:47

Then it needs to have a lot of. I think so much simpler to say than actually to do If I were a startup or even a mid-sized company and I have a new system to do I don't know some recommender system for Klasolson or whatnot, and I want to put it through and make sure that we're compliant with AI Act. How would you use a regulatory sandbox to try it out? Is it checklists? Can you run the model in some kind of sandbox, as it sounds like?

David Magård 1:42:22

We don't know and that's something that I actually left out to a large extent of the regulation and say this is something that comes in the implementing act. What I think, patrick, when we were discussing, is that we must do something in Sweden from the public sector, where we kind of pilot this. How would this look like? What is like the best way to do this? What should we focus on?

Henrik Göthberg 1:42:48

Is it?

Anders Arpteg 1:42:48

guidance or is it actually?

David Magård 1:42:50

infrastructure, or you know both, or so on.

Anders Arpteg 1:42:52

Why didn't anyone like think of this before we define what the act is?

Henrik Göthberg 1:42:57

I mean, this is what I mean with the package Some concrete example.

Anders Arpteg 1:43:00

You know, this is really what we should do Give some kind of stereotypical kind of examples of okay for startups. I want to do this. This is how we could help out.

Henrik Göthberg 1:43:07

This is actually my biggest complaint on the AI Act is that this is what I mean with the package that you almost need to have a cohesive strategy here, I think it's gonna.

Henrik Göthberg 1:43:20

I mean, now, okay, now we do it sequentially, that will work, but it will be muddled through a bit more, and that is the but. Of course, you can muddle through as long as someone throws a lot of money on it and brain people on it. We will solve it, of course, but it's gonna require some brains and some money to really do that. I mean, like I have this idea in my head. I mean, like, how is Tonya went for? You know how you can. We're going to be the best to provide. If you want to open up an EU company, it's going to be the smoothest, easiest, most sharp approach through that. So how can you sort of reverse engineer that into the smoothest user experience to build a start AI service? I mean, like, people are working on this, of course, but I think this is super important now, and I mean, of course, regulatory sandboxes, doesn't?

David Magård 1:44:12

they just didn't fantasize about them? There are existing samples of that. Spain has a pilot sandbox actually for this purpose, and that's that's one I'm part of through my expert role as well. So, we have a kind of an insight on what they're doing Not so much at this moment, I have to say, but still they are thinking of what they're going to do. So there are.

David Magård 1:44:37

I mean, people are thinking about it, yeah people are thinking, of course and there are moments here, not all of them being being, of course, official and so on. However, of course, everyone can do more, and I think my, my, I'm not. This is not going to be a GDPR or thing. I think it's not that intrusive.

Henrik Göthberg 1:44:56

And it's not that you don't think it's going to have as big impact.

David Magård 1:44:59

No, I don't think it's going to have a big impact. Why?

David Magård 1:45:01

not Because everybody has data, everybody has data already and you kind of GDPR is just a kind of update of several, several older regulations. Ai Act is more like something that's happening now, but also quite new. You don't have to, you don't have to use AI, right, and then there's, then there's, and then there's. Also, you know there's this kind of test before you put it out to the market. So it's, it's something that you have to do before, but then it's out to market then.

Patrick Eckemo 1:45:35

Then for me, use cases as well, but when you don't have I mean it doesn't affect everything under the threshold. So I mean it's, you don't have to do anything about it. You can use an easy, easy PC AI without doing anything about it.

Patrick Eckemo 1:45:52

And when we went to US, for example, and and had a discussion with the government over there and why they come a little bit farther than we'd further than we'd done here in Sweden the simpler said we don't you just use simply AI, no personal data whatever, and so on. So we just use simply AI and then we can move forward.

Anders Arpteg 1:46:15

So I think that would be a thing. Imagine the country. It actually do help the companies in that country the best.

Henrik Göthberg 1:46:22

Yeah, this will be a compedge. Of course this will be it.

Patrick Eckemo 1:46:25

But I think it's an opportunity with sandboxes like that, but regulatory sandbox, because we just as an example we are, the data in Sweden is very locked in in registers for specific purposes and we have a I mean, it's a long tradition and based upon integrity and all that kind of stuff, so it's an important tradition that we need to maintain. But at the same time, the lock in makes it very difficult to fight crime, for example, and things like that and sandboxes like this. Then we can. It will be really interesting to test scenarios and, for example, if you combine, combine this type of data and this type of data and this type of data and use AI to analyze this type of data, we can answer these type of questions and we can stop crime there and there and there, and then we can talk with politicians about this and maybe that can start like like a require, like a force for change laws and other things and the balance between integrity and fighting crime.

Henrik Göthberg 1:47:30

But we can. We can almost wrap up the AI act topic because it's almost like a segue in now to, given what we now know with the AI act and everything else, how should the Swedish state invest their money in AI? And I can go. First, I put the post on LinkedIn and I even PMed one of our former guests here when I saw that, like Jim Dowling I'm not sure if you know Jim Dowling he's been on, he's been on this podcast. He's he's the founder of hops hops works.

Henrik Göthberg 1:48:06

He wrote a very, very nice article in the Irish independent, or whatever he sees Irish, where he basically highlighted for, from Irish point of view, where and I thought was a brilliant article you know what we have an opportunity in Ireland now to be world class in managing startup or the ecosystem on AI, because we are going to be the world class in managing the AI act and making it super smooth. You know, clara, smooth, smooth payments, smooth AI act. You know you can have it, jim. I like that idea right. So I think this is one, as you were hinting right, what a competitive edge in Europe if your country is super user friendly and adoptable, friendly in terms of company friendly company friendly and has nothing to do with not using the AI act is about being super efficient in supporting us.

Investing in AI for Sweden's Future

Henrik Göthberg 1:49:04

I mean, like that's my take, I would spend a lot of money of you know infrastructure. Yes, yes, sharing data tricky. This one is a no brainer and it's a political no risk zone to build a very, very good agency for this. What do you think is other investments?

Anders Arpteg 1:49:21

We're framing the question in some way. I mean, if you were the prime minister, let's say you are dictator. You're Elon Musk of Sweden, elon Musk of.

Henrik Göthberg 1:49:33

Sweden, the knowledgeable dictator.

Anders Arpteg 1:49:35

Yeah, the knowledgeable disperse. How would you spend the money to just make sure the economy and industry would be blossoming through the AI act? Super simple question. I can start.

Patrick Eckemo 1:49:54

Yes, I mean not just the AI act, but in AI in general, I would say, how to make the best use of it. I think I would start with a common model for what we are going to achieve in Sweden, like where do we want Sweden to be in 2030, for example, and from that, start looking at how can we, what do we have to do, to make Sweden like a successful country in 2030 by using, I mean, digitalization and AI in this sense. Then we can start writing down like a plan for that. What type of capabilities do we need to do in investments and things like that? And probably huge, huge investments. And we need to work together.

Patrick Eckemo 1:50:45

We need to put up very clear governance and steering of a program like that and we're talking about tens of billions of Swedish crowns in investments to make that happen, because we need, like, common infrastructure, common data, common standards and formats, so we can even look into how to interchange data among agencies and regions and municipalities, for example, but it should be connected to common goals and like missions, connected to the target picture of Sweden 2030. Then we can start working against that picture by enabling what we have to do like education, infrastructure, laws, regulations and things like that Everything combines down to what we have to do to fill that gap, to make this happen, because I'm pretty sure that if we are going to maintain or develop Sweden as a healthy country, we don't have an option, I would say, because every day now we are lagging behind and we don't do anything about it.

Anders Arpteg 1:51:57

And would you agree that you know? Of course, ai is going to influence our society and the world in so many ways and the only thing I think we know is that the people and countries and companies do not use AI will be left behind.

Patrick Eckemo 1:52:11

That will be left behind and it's probably an exponential growth here, so it will be impossible to catch up later. So if we want to be one of the key performing countries in the world with a health system like we have, welfare system and so on, we need to start now and before it's too late. I would say but how do you invest now in this?

Henrik Göthberg 1:52:33

because you're making it. You know, a utopian plan. This is how you do it. But then you put that in contrast to the realities of how we are organized in Sweden, I mean like quite difficult to make these decisions as one. So are there some fundamental? Are we talking about infrastructure changes? Are we talking about some other fundamental legal changes or how we run Sweden?

Patrick Eckemo 1:52:56

in this. That needs to happen. Yeah, I'm talking about exactly that. But to make it simple, we need to have at least like a common governance and leadership and management of digitalization and AI within Sweden, and challenge we have today is that we don't have an agency that can rule, decide all the others except for Rian's Consulate.

Henrik Göthberg 1:53:21

But, like in any company, if no one has this as their main objective and main task, but the third objective or third task on a very strained agenda, how does it get done? You know, it's that simple right. Someone needs to take a lead on this, and this is what you're proposing.

Anders Arpteg 1:53:37

David, how would you make Sweden the best country for? Or with AI for our economy and society?

David Magård 1:53:43

I guess with AI act, or did we leave that out now? Okay, yeah, now but, can I do more?

David Magård 1:53:57

more like what, what, what if we can see? Now it's so I think from the what's happening is the government has already said, like we have the AI Commission, that that is that's how we're gonna understand what we're gonna do. The thing is that the Commission would end, have his end report in 2025. Most likely, ai act in in all of all our many parts of it will be in real force 2026. I would argue that probably it's a bit late then if we want to be the best one in in like a show in for the AI system that we have the, the smoothest restriction into the act and all of these things. The Spanish obviously kind of work with that already, with the pilots and so on. So I think we need to do things now and it has to be done together with public private sector. If I could decide a bit more and have my Elon Musk powers here.

David Magård 1:54:57

I would probably just kind of take, take some money from the different agencies, pilot together and and use that to to a common effort. Yeah, yeah, I'll probably do that and make sure that they work from day one to support the ecosystem, because have the horizontal approach and have the horizontal approach.

David Magård 1:55:22

that could be like dig with, with some more power or whatever, but but that is what I would do and I think that's not going to be done, of course, but you know it's, it's please dream, yeah, yeah because I think one of the one of the big hurdles with the act will be the public sector, because the AI act put public sector in the middle of this right, the public sector with the natural competence authorities. They are the ones who will say that this your, your product, your system is okay. They are the ones who will supervise you, they are the ones who find you, they are the ones who should support you to ensure the real sort of science. The private companies they work, they run fast and ready, you know. They just need kind of support to keep doing that in the right direction when it comes to regulatory and so on. So you need you need a really strong, you know entity that supports that.

Henrik Göthberg 1:56:16

And if you give the size of public sector as part of the Swedish BMP, I mean like this is the real game in many ways right.

Patrick Eckemo 1:56:25

It's roughly half of the BMP.

David Magård 1:56:27

Yeah, and I mean for for public sector to actually be able to procure these AI systems. We need to understand them and to understand how do you do it. We need to have the guideline, support and all of these things.

Henrik Göthberg 1:56:38

So this is a massive undertaking, and to put to do something that works as an agency that supports the public sector would have a spin-off effect anyway on the private sector yeah, if you support the private sector as well. Yeah, but if you do it, if you know what, if if, using LOU, someone can procure something and do it in AI any private company he can get this done, you know. So if you can solve it for the public sector, the same service, I think, will suffice for a lot of private companies.

Henrik Göthberg 1:57:08

Let's use the Osthivel I like but you're saying she's greater to get money from all the different parts to put in an AI bucket.

Anders Arpteg 1:57:18

I mean, to me it sounds like the obvious best solution and I just simply wish that someone you know takes that action.

Henrik Göthberg 1:57:25

But here's an uncomfortable question then Is our AI commission, with the deadline in 2025, coming out with an utredning a little bit too late? Yeah, if we want to be the best one on AI, I think If you're scared of Cursewise law of accelerating returns and exponential growth and speed and innovation. Right now, 2025-26, is kind of yeah, it's happening now.

Patrick Eckemo 1:57:52

I think we have a sense of origins as well, since there were demographic effects in Sweden and the rest of the world as well. I mean, just look at the welfare system. We need to hire at least 200,000 people until 2025-26.

Henrik Göthberg 1:58:07

It's now.

Patrick Eckemo 1:58:08

It's happening right now. So we need badly to increase the productivity within the public administration right now, otherwise we will have a huge debt, I would say, and we can't even the people, we can't even hire the people and we don't have the money either, so I mean it's there are no people.

Henrik Göthberg 1:58:25

There are no people.

Patrick Eckemo 1:58:26

So it's impossible to solve without using.

Anders Arpteg 1:58:30

AI and it's just give a small positive spin on this I think we do actually have a rather good educational system compared to other countries. We can at least hire some more junior people and hopefully upskill some people as well. So perhaps there are some ways just to at least not be too pessimistic.

David Magård 1:58:48

I mean Sweden is we have a good situation. The thing is that we used to have a good situation or even the kind of the best situation in many aspects, and I think that we have seen now with the digitalization that we are how we built the site. It has a lot of close with some negative, some downsides, and we need to kind of address that. And I think you know, ai commission, that that's actually it's good in many aspects, but it depends, of course, what your target is Now you also target like for AI act.

David Magård 1:59:19

Okay, that that's not gonna suffice. Maybe they will do something else, but we are in generally, in a quite all right position globally.

Henrik Göthberg 1:59:31

Difficult questions.

Anders Arpteg 1:59:32

Should we move to some more philosophical, even more simple questions?

Henrik Göthberg 1:59:35

You know, I had one very operational question but we might need to skip it on time. I was actually gonna go into. I used to know from talking to Patrick that they are doing very cool work inside Bulldogs Market on how you're facilitating the strategy process and strategy and innovation process and I think that is this is something I want to learn about. So I'm a little bit curious of having hearing that story. So this is a little bit like going operational and then philosophical.

Patrick Eckemo 2:00:09

Yeah, I mean we see digitalization and AI as a means for an end. I mean nothing else. So, and when it comes to how we work with, I think we have a very clear vision at Bulldogs budget when it comes to where we want to be and we want to have company information flows real within the society means that we need to change our role within the ecosystem, and the way we're trying to to work with strategic business development at Bulldogs Market is that we're using enterprise architecture methodologies behind the scenes, but we're not talking about it in front of the leadership teams. I have an architect background myself, so that's one of the reasons why we're working with it, but one of the we have certified a whole unit at Bulldogs Market in business architecture, so we are deeply, or say, educated within the area.

Patrick Eckemo 2:01:11

We have a highly expertise in the area, so we're facilitating the management groups, visioning how about Bulldogs Market should operate the business in like five, six, seven years, and talking about how to design products and what type of value propositions to the society, how to integrate with the ecosystems with the system, vendors, banks and all the rest in ecosystems and when we design this type of future picture. Like I talked about Sweden and 2030, but for Bulldogs Market we have a gap. As we are, where are we today and where are we going to go? And the way we're working with that is we need to establish, like, the right capabilities to operate our business in the future, and that it that will be the plan roadmap for us to fulfill during the years. So we have like a strategic plan based upon that, we have strategic goals based upon that, and so forth.

Patrick Eckemo 2:02:11

Ultimately capability roadmap how you're gonna that's one thing of it, of course, and it will end up like a portfolio management for like the 12, 24 month rolling, so we can work with that and work with prioritizing and how do you break it down into from 24 to 12, to how deep do you go? Sort of all the way, because we need to understand how, for example, we are changing the underlying systems in the architecture. We need to get rid of the whole stuff and build new stuff, open stuff, that are integrated with the rest of the market, so everything is API based, basically.

Patrick Eckemo 2:02:52

So we're looking into like being where the customers are having their lives. For example, if you want to start a new company, you can do it at a bank, for example or maybe in your ERP system.

Patrick Eckemo 2:03:06

Henrik wants to start a new company. You do that and you can suggest a name, and it will send information to Bullocks by then. We will test it with AI models and things like that and come back with suggestion for you if you're on the wrong path, so to speak. And so it's opening up our services and integrate more both APIs and perhaps shutouts.

Patrick Eckemo 2:03:29

So yeah, yeah yeah, we're looking to everything. I mean we want to automate as much as we can, of course, get rid of a lot of unnecessary demand on the agency, at the same time as we want to increase the service level with new technologies, of course, like AI. So digital assistance, for example, could be very trained from like a foundation model, and then we have, like it could be trained on our data, find you, but also with like a rag, like content models connected to it, so you can have a rich models that will be updated without updating the foundation models and so forth.

Patrick Eckemo 2:04:16

So so a lot of things are going on, but in so we work with the, with the executive leadership where I'm part of, but also at each level in organization, with vision, where we are going to be and involve all the developers I mean business developers in that process and it will end up in what we do today.

Henrik Göthberg 2:04:39

That brings us to this future picture and I'm just curious now, like having some insights from from companies both like bull leader but, but more importantly, maybe Scania. I mean like it's super clear that when we're becoming more data and AI intensive, the organization changes, that we it's not anymore the business over here and the data IT CIO organization over here. It becomes much more platform and domain for platform and product teams. So you can sort of learn a lot from what the spot device and the colonists and done so. Are you in that process as well, of sort of having engineers embedded closest in the product teams versus infrastructure platform or you know, it's that kind of conversation happening because this is, in my opinion, one of the hardest questions to have with the normal business. People sort of speak what do you mean? Do I need to have the engineers and data scientists in my team? Don't they work centrally? Is that part of this whole re-architecture?

Innovation Strategies and Technology Testing

Patrick Eckemo 2:05:37

It is and we have cross-domain experts working with that and, as part of what my team, I have the AI hub in my team as well. David is in my team and so I love that. Different type of experts working in long-term, like exploratory development, so to speak.

Patrick Eckemo 2:05:57

So the case David mentioned before when it comes to how to, just as an example, if we, if you're two guys are going to make do business, you need some proof from the public administration and instead of connecting over talking with all the different type of administrations, you can download, like, a trusted document from us which will be stored in digital identity vaulting it could be based on blockchain, whatever doesn't matter and then you can share it with everyone you want to, and by doing that, you will have a much more efficient process for yourself, because you don't have to wait for doing this type of process every time or request every time.

Patrick Eckemo 2:06:45

You have documental ready, trust in your wallet and you don't have to access. You take away unnecessary demand from the Danish as well, so it's a win-win scenario, I would say. And that's the type of work we've been doing, which is like two, three, four years in front of us. We are not using this today, which also led us into leading this one of the consortium at the EU level. So it's pretty unique, I would say, how we operate at developing things today, yes, but we are looking into and testing new technologies for tomorrow.

Henrik Göthberg 2:07:21

You have to, we have to. Yeah, you have to be working on the problems of today, but you need to start preparing, shaping for the problems of tomorrow, or the setup of tomorrow, and we don't actually know how we will solve all problems to achieve this target picture?

Patrick Eckemo 2:07:36

We don't know that. We just know that we need to change the how, and we're testing new technologies to achieve that.

Henrik Göthberg 2:07:42

Fantastic.

Anders Arpteg 2:07:44

So we get to more into innovation topics. But let's not do that because the time is flying away.

Henrik Göthberg 2:07:48

One innovation topic, you can have it.

Anders Arpteg 2:07:52

OK, OK, just one, but it's going to be a rabbit hole.

Henrik Göthberg 2:07:55

Yeah, I know OK.

Anders Arpteg 2:07:58

So innovation can happen in many ways.

Anders Arpteg 2:08:01

You can have an innovation team that works separately outside of the development teams or you can have, like Amazon has another kind of innovation strategy, saying every team should be working with innovation, and they even have a way to say that they want to support innovation by saying we have this kind of press release approach. So if anyone a developer or whoever comes up with an idea, they should start by writing a press release, or this is what the press release should be, internally or externally. If they publish or deploy something that the media says about this new product and then from that you go to some kind of management board or whatnot and ask do you think this is a good idea? If they do, you get some extra time and perhaps even build a team to start working with it. What's your idea in the BulaSarget? Do you think we should have both? Should you have both, like the team that just focus on innovation, or should you have innovation happening throughout the existing teams that you have?

Patrick Eckemo 2:09:03

I think it's a combination actually as many things.

Patrick Eckemo 2:09:06

You need some experts within the methodologist and how to do it. We have really high expertise in innovation management, I would say, but innovation part doesn't happen there. I mean, we have a very cross-domain perspective, working with all the people that work with us, but also, I would say, outside of what we expected, since most of the products we are running this one, as David talked about, is a bit totally open. So we're involving other agencies, companies and other parties that will add knowledge and insights and just thoughts about what we are doing. So we're doing that, as you said, we talk about it internally and we sell it internally, like that, and should we invest in this? Should we talk with the government on financing this, for example which we've got financing for this case as well from the government, which is rare and really good, I would say and work with other peers outside internationally as well. So I think the combination of openness and involvement is the key part, and transparency, yeah and also kind of doing, because that's something that I think, at least in my mind.

David Magård 2:10:24

In public sector, sarah, I know there's always like a focus on innovation, competitions and so on. That's good. It's good, but you also need to have do something, so you need to invest a little bit in it. You need some people that actually dig into the mud, because then you can see new things happening, and I think, as Patrick is mentioning here, they should also be open and externally as well, because if you're only in your own setting, then you don't reach new conclusions. Really, you have to get out there.

Henrik Göthberg 2:11:00

But I had two parts to this, because one part is that to work with innovation is really hard to do in your cave. You need to get diverse inputs and look at what other people are doing to see how it works. But here another topic also here is like sometimes people I get the feeling like, oh, we condense down that innovation is equal of hackathon, so something we do like we were super creative and we had this hackathon. We came up with 500 AI use cases. What did you do? We wrote 500 lines. You know you haven't even started right. So what I take out of that comment from you is like then you need to do the work. Someone needs to get the money and the resources to actually dig in and go deep.

Anders Arpteg 2:11:45

Yeah, it's not that you, Patrick. I think I stole a number of slides from you, but I think one slide in particular that I think I stole from you was the prototype graveyard.

Patrick Eckemo 2:11:53

Yeah, yeah.

Henrik Göthberg 2:11:55

I think we have used that word, by the way, extensively.

Patrick Eckemo 2:11:57

Well, that's right. I mean it's very rare to mention. It's one of the reports as well, and I think it's SCB. They mentioned it in the report from 2019, that public administration spent like 150 million roughly by then, and most of the money when we talked with them as well, it went to testing checkbooks, basically doing the same scenario, and nothing went in production. Often it was isolated innovation teams or whatever, without the necessary competence for production, putting in production, dieting all that.

Henrik Göthberg 2:12:34

Innovation to adoption. Yes.

Patrick Eckemo 2:12:36

So it ended up on a graveyard. So that's I mean that's, I think, still much pretty much the same Today. I would say, in many cases I'm really afraid of that because it's some says, start doing, just go out using AI so we can get started, but you need to have the right I mean, you need to have the right prerequisites for like prerequisites to doing this right. What would say is like the competence and technology, knowledge and a lot of things to be a super-comfort management or competent buyer, so to speak, and we lack that today. So it's a high risk just going out to buy from consultants.

Anders Arpteg 2:13:20

Is it such a you know rabbit hole?

Henrik Göthberg 2:13:21

really, it is a rabbit hole, but I mean like we used to throw you know some fuel on the fire and, being provocative, I love to. I mean like I, you know this is a little bit like you know, working with data and AI. I shied away from working with public sector because I see too much holding hands exercises, I see too much sort of you know, not mentioning any names. The same, you know AI consultants talking about AI in a super, super high level and you know there is no operational competence from how do you get the budget? How do you get the resources? How do you get the engineering team? How do you get an operating model around data? You know, if I compare to you know people, that's what we have done, been doing in Scania. But then I look NBS at Spotify, right, so it's a huge gap.

Patrick Eckemo 2:14:09

Yeah. So I mean, if we go back to the dictator? If I had a mandate by 2019, spending like 150 million doing mostly the same thing ending up on the radio. I would say maybe we should use like 50 or 100 of them to put something in common that everyone could use for the public administration and put in and enforce it to production, whatever it is. Exactly, and then we have something useful for the market. And that would be an amazing key thing when it comes to steering the resources.

Anders Arpteg 2:14:41

We're not doing it. Very well said.

Henrik Göthberg 2:14:45

I don't see that as a negative. I see that as a positive, that we know what we need to do.

Anders Arpteg 2:14:49

Which came one step further, at least know what we should do, even though we haven't started it. Okay, we really need to start to close off here, but so let's go even more philosophical. And we, how should we do it?

Patrick Eckemo 2:15:04

Okay, go around thinking let's start with David now.

Anders Arpteg 2:15:07

David, when or if do you think AGI will happen? I have no idea.

David Magård 2:15:15

Okay, patrick, yeah, but that's.

Anders Arpteg 2:15:25

Do you think it will happen?

David Magård 2:15:27

Yeah, but I mean, who am I to say really, you know, I'm happy if it happens, it will be an interesting time.

Anders Arpteg 2:15:33

You are happy If you're my. I will come to that question later. If you're gonna, when I leave, I hope to, I hope to, I hope you Do you think three years, five, 15, or 100 years?

David Magård 2:15:47

I think the summation from most of the research is 10 years from now, right?

Henrik Göthberg 2:15:51

2025,. I think 2029 is. The code file number seems to be the most, but it seems people are even closing in on that.

David Magård 2:15:58

Yeah, it seems people Okay, so far.

Henrik Göthberg 2:16:00

I've been reading Max Tegh Max's book and they had this Puerto Rican conference and no one. And basically he says no one agrees on this, of course.

David Magård 2:16:08

Yeah, of course I stand on the shoulders of giants. You know, that's not my.

Anders Arpteg 2:16:13

Patrick, what do you think? Do you think it will happen or when, and if so, when? I think it will happen.

Patrick Eckemo 2:16:20

I think it's inevitable actually, since what we have seen last five years, but even before then I actually thought it would happen in the next maybe 30 years or something. So if we reach it, I mean we have the levels from Google now defining the AGR levels. That's a good way of putting it.

Henrik Göthberg 2:16:42

We basically got level one right now. Yeah, I want to Maybe two.

Patrick Eckemo 2:16:45

But I think it, we will reach it in time and when we do it it's just matter of time before we reach superintelligence, so I don't think it's inevitable.

David Magård 2:16:56

So what do you think, Arles?

Anders Arpteg 2:16:59

I mean, I think still, ray Kurzweil is correct 2029. Elon has said 29. Now I think he's gone a bit down, so 27, 28. But I still believe in 29.

Henrik Göthberg 2:17:10

It still is, but it depends on the definition. That was my line, man. This is when we sidekick. I was going to say that it depends on the definition.

Anders Arpteg 2:17:18

My definition is basically when it's hard to even find a single task.

Henrik Göthberg 2:17:22

I saw the guy on the Nvidia CEO. What's his name? Jensen? Yeah, I mean like, if you define AGI in relation to benchmarks, blah blah, blah, right, then it's not too far off, but if you're a little bit different on it.

Anders Arpteg 2:17:36

My view is simply to say that when it will be hard to even find a single task, that, whatever system AI system, you have AGI system or not that it's hard to find a task that it can't do better than most humans, then you have AGI. We were far from that. It's very easy to find tasks.

Henrik Göthberg 2:17:52

Yeah, but the chance of it cannot do. But do you put that into AGI, in robotics physically doing things in society, or do you put it as administrative?

Anders Arpteg 2:18:02

I think we could split it too, you know, if you remove the control part and just say perception and reasoning, then you have a simple version of AGI. It can't really do things, but it can say what to do, and that's easier to reach. If you say it needs to actually do things, and let's say that you have an AI system that can beat human champions in soccer. That will not happen in 2029, for sure, or I would say, or, unless you know, optimus really, really explodes.

Anders Arpteg 2:18:31

But I think you know that kind of the control aspect of having the physical control, motoric part. That will take longer time but from a planning and perception point of view, 2029, I think is reasonable.

Henrik Göthberg 2:18:46

I mean like yeah, and so the point is this like the arguments of the dooms sayers or whatever we want to call it, the AI doomers, not boomers, doomers, that was my joke. It's about getting onto the real questions now in order to be ready for them when this accelerates. So it's not about yeah, I think that's fine.

Anders Arpteg 2:19:11

Final question, then If let's say, assume that AGI will happen at whatever time, we can imagine at least two extreme points. One extreme is simply the doomers view that we will have the terminator. We will have the Skynet that is, building machines to destroy all of humanity and the matrix and what not happening. And that could be one way.

Anders Arpteg 2:19:35

And the other extreme is basically the very positive, utopian kind of view where AI will help humans live in a paradise in some way, living a way where we're not forced to work, potentially 40 hours a week. We can choose to work if we want to and perhaps if we need to have some other needs fulfilled, but we can live in a way where we are free to pursue whatever passion and creativity that we potentially want to.

Henrik Göthberg 2:20:02

That could be another very utopian kind of view.

Anders Arpteg 2:20:06

If we start with you, David, do you think we will move potentially towards more of a dystopian or a more of a utopian view?

David Magård 2:20:13

I think, more utopian. I think at least for quite some time. We will see a big divergence between people though, and countries, but people in countries as well.

Anders Arpteg 2:20:25

The divide between the divide, yeah.

David Magård 2:20:29

So we'll see a class of people having these tools available to them and the ones who doesn't, and I think that's something that we won't overcome in a short time.

Anders Arpteg 2:20:40

Concentration of power that we're seeing now in a few companies and even in a few people.

Henrik Göthberg 2:20:47

We haven't said so much about the AI divide for a company.

Patrick Eckemo 2:20:49

I think it's also for the country level.

Henrik Göthberg 2:20:51

Yes, Because this I mean, like we've seen it in the past, right, and you know this quote they taste the new oil, which was completely misunderstood because they thought about it literally. But when the economists wrote that article, it was all about the money, it's all about the power and the geopolitics of oil, and now we're going to have geopolitics based on data and AI for sure. I think it's all we already have right. I mean, like, look at who went to UK. It was both the biggest countries and the biggest players.

Henrik Göthberg 2:21:28

Patrick what do you think We'll have a?

Anders Arpteg 2:21:31

dystopian future or a utopian future?

Patrick Eckemo 2:21:34

I think both scenarios is valid, actually Interesting, yeah. So I think, depending on what we do, so we can. I think it's up to us. That's well said. So I mean we could easily end up in both scenarios and I really, really, really hope we end up in the utopian one, because we don't have an option then to use it today. So we need to be optimistic and do whatever we everything, all the things we need to get in control and do the right things.

Henrik Göthberg 2:22:08

We had a guest Was it Kaiso? Who else was it? Or who had a quite interesting view on this. Where will we end up? Right, which was moralized. It will be the same as we have today, and both me and Andrius did like what do you mean with that? Well, we will have general AI and it will have crept upon us before we knew it. We had it and, but we will have our next level, daily struggles of the Kekuikan or whatever it is so like.

Henrik Göthberg 2:22:34

So this utopia, in the sense that you're drawing a picture in so, I thought that was a quite interesting remark, that it's going to be the same but same, same but different.

Anders Arpteg 2:22:45

I don't have a real humans. We're still humans, right. Potentially we've fixed cancer, perhaps fixed energy crisis, yeah, but then it's another crisis man.

Henrik Göthberg 2:22:53

So when you fix and you know those, there will be other. You know Elon's problem? Yes, definitely. Maybe, oh, we're going to Right, you're going to fix it. That's the problem. We're going to fix.

Patrick Eckemo 2:23:05

Oh see Anyway anyway, we just download our brain and then we have a pure that's the point Right Pure digital twin.

Henrik Göthberg 2:23:14

You know which movies that when you download, your brain in the end in this way, you live forever. Can't remember which movies.

Anders Arpteg 2:23:20

Good one. Awesome, it's been a. It's been a true pleasure to have you here, patrick Ekmo and David Maghurt. I hope you will stay on for the really interesting discussion that haven't even started yet. Yeah, we'll do off camera. Thank you so much for coming.

Henrik Göthberg 2:23:37

Awesome discussion. Thank you very much. Thank you.