E183 - AI and Cybersecurity - Åsa Schwarz

In Episode 183 of the AIAW Podcast, we’re joined by Åsa Schwarz, CEO of Aranya Consulting, cybersecurity expert, board member at Precise Biometrics and Enea, and acclaimed Swedish crime fiction author, for a deep dive into the rapidly evolving intersection of AI and cybersecurity. Drawing on more than 25 years of experience in cyber defense and technology strategy, Åsa explores how AI is reshaping the threat landscape—from attacks on legacy infrastructure and operational technology systems to deepfakes, automated phishing, and large-scale disinformation. We also discuss the growing AI security divide between Europe and the United States, the evolution of criminal business models in the AI era, and what organizations must do to build resilience in a world where trust, identity, and digital sovereignty are increasingly at stake.

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

Transcript

AI Disrupts Writing And Translation

Anders Arpteg 0:00

Yeah, I mean I guess people are a bit uh angry with that. I I remember speaking to some other journalists at one point and they they really thought AI was was bad for business, so to speak, in their case. It's very bad for their business. Yeah.

Åsa Schwarz 0:16

And also the translators, they are out of business almost now. And I think some of the writers will be there too in in a while. We haven't talked so much about that in in in writer in the writer collectives, but I think some will be out of work also.

Anders Arpteg 0:34

But if you compare it to like music generation, I mean uh you you can use AI to do music to you know really good, I mean not to the extent still as humans, but still rather okay at least today. And and of course you can use AI to play chess, and and you have no chance as a human to be an AI chess player, but but still people play chess a lot.

Åsa Schwarz 0:58

Yes. And it's you know, m most of the writers or the authors, they they don't uh earn any money on the writing anyway. So so uh it it could be still a hobby, also. But I think you can make marvelous books with with AI, but with a little bit of help from the human, you get you know the ID, the the storyline, and I think unfortunately that even the authors can be be um sort of but but could you see the same we call it progress or or change happening to uh novelists or or writers or authors, as in development, as in for music, as in so many other work tasks where humans more or less like give tasks to a set of agents and they you know uh do the the work and talk to each other and then come up with some idea. Yeah, I think I think that's that I I think that will work today. Uh and now we will only talk about the negative stuff that people will be out of work and and so on. But but I think there's another part of the story is that you know if you can translate your books to all languages you want to, suddenly you have a completely different audience than before. In Swedish, it's ten million people, but if you can translate it to everyone in the world, you know, you have a different audience.

Anders Arpteg 2:27

What do you think about the concept of um I think it was Sam Oldman who said once, you know, software will change, you know. Normally traditional software engineering is really you build for a number of years some piece of software and you sell it, or you have software as a service in some way. But in the future it could be on-demand software, meaning I need a software right now to cook this meal, some recipe and how to do stuff. So I just asked AI to write the software for me here and now, and then I throw it away. So it's just on-demand, and you can do it in uh in a minute, and and you know, you don't even keep it afterwards. Do you think the the idea of on-demand software could translate to books? So you have like on-demand books in the future? Could could that be possible?

Åsa Schwarz 3:14

That's a really fun idea, actually. Yeah, I think so, because if you write a book and you usually sort of the audience will follow the book if they feel something personal. So if if you can just develop a book for one person, with perhaps in their city, uh with their hobbies and so on, you know, it would be much more interesting.

Anders Arpteg 3:37

And perhaps as an author, as a human, then what you do is you you really form the personality more or less of how the AI should write the book so it always becomes in a certain style in some way. So if you want to have your you know a nighttime story for your kids or something, you ask an AI that the human has you know directed in some way, and then you get that kind of uh you know stories for the kid, and every time it's different.

Åsa Schwarz 4:05

Yeah, yeah. And and you can also sort of if you find a couple of stories that you would like, you can ask the AI, I want a story in that style. So you know, you don't even have to do more than that, I think. But you still write uh I I I still write, but I I I when I got stuck, I I get the the ideas. And I also sometimes, if I think, oh my oh my god, this is not funny, or this is this is too slow. I can ask the AI. How can you do this better?

On-Demand Books And Personalization

Anders Arpteg 4:34

Yeah, right, right. Cool. Um well with that, uh very welcome here, also. Um you are um expert in cybersecurity and AI and so many more things. You're a board member, you are an author, novelist, right? Uh public speaker, and so much more. So I love to have you here and speaking more about cybersecurity and AI and uh how how that future will evolve. So much happening in cybersecurity.

Åsa Schwarz 5:02

Yes, it's it's it's really exciting times and it's also a little bit worrying. So, you know, if in in our business you don't really know if you have, you know, will laugh or cry. So it's it's so it's it's happening a lot on the fence, and it's also happening a lot uh when it comes to uh adversaries, but it also happens a lot in the companies because suddenly people are asking us what is happening with AI and cybersecurity. A couple of years ago, nobody cared really.

From Early Computers To Cybersecurity

Anders Arpteg 5:32

And now with Mythos and whatnot, we we have to go into that uh shortly. But before we go into all of that, uh perhaps you can give a quick uh background about yourself. Who who is really Aussas?

Åsa Schwarz 5:44

That was uh a broad question, but I'll try to be short. Um I have always always been quite interested in computers. My my my uh mother gave me a computer when I was about eight years, and and that was not that common, but I had uh you know she worked with computers in the 80s, and she gave me a computer and said play with this and I did that. And so that was really fun. So I learned uh what a computer is because it didn't even have a hard drive. So what computer was that? ABC 80.

Anders Arpteg 6:21

Commodore as well or Sinclair or something?

Åsa Schwarz 6:24

No, I think I switched to EBM after that, and then I think it was Macintosh, but I'm not really sure.

Anders Arpteg 6:31

I'm glad you didn't say Amiga.

Åsa Schwarz 6:33

Oh, is that is that something that shouldn't be?

Anders Arpteg 6:36

No, but I had a lot of friends who were into the Amiga uh stance and I was more in the Commodore and IBM framework. So yeah.

Åsa Schwarz 6:45

Okay, I understand. So I I started there and and uh after a while I um went to the university and uh I started to compu uh started to study computer science and and uh also um ekonomics and then I really tried to to uh study AI, but I got so disappointed. What year was this approximately uh ninety-five-inty-four or something? And you know, I I have been seeing all these uh uh sci-fi movies with robots and AI that is talking, and you just got I don't know if you can say it on the podcast, but shit. Do you have a favorite like um movie or AI movie or um quite old school so I sort of love Robocops and stuff like that. And so, but instead I I uh studied cybersecurity and you did that early? Yes, I did. What university was this? It was it was so kooperation between uh Stockholm University and uh Kotihov, what is it called? KTH, yeah, yeah, yeah. And they had um computer science klasses. And the best one was really the cybersecurity classes.

Anders Arpteg 8:12

Yeah, yeah, yeah. Interesting. So what did you study then? I'm I'm you know, I'm a big fan of cybersecurity as well. But um what did you study at those times?

Åsa Schwarz 8:22

Uh it was some kind of uh it was both uh how a system works and how you, you know, control it and and also cryptology and stuff like that.

Anders Arpteg 8:33

And how operating systems works.

Åsa Schwarz 8:34

Works, yes, yeah. So it it was sort of I got to know more about computers on that courses than the the other ones actually. And after that I was a while in a um um a consulting company uh that was working a lot with the defense industry, and I was the only one that had studied cybersecurity, so then I started working there with that.

Anders Arpteg 9:00

Okay, cool.

Åsa Schwarz 9:01

And then uh you continued uh yes, I continued in different uh cybersecurity consultant companies, and I have different roles both as you know a manager strategy and stuff like that, but I also I was all all the time working also as a consultant. Uh so my my the last company I worked at uh was Now it's Cybersecurity Law, uh where we I think we went from 14 consultants uh uh 13 years ago, and now we are or we are 160. And now I'm saying we, but I'm not working there anymore. I I quit three weeks ago and started my own company.

Anders Arpteg 9:40

Yeah, yeah, yeah. Congrats on that. And and I'd love to hear more about that company as well and what the focus will be for that company. But uh you also have been writing books for quite some time, right?

Åsa Schwarz 9:51

Yes, I I started uh I think it was about 2000. I released my first book 2005.

Anders Arpteg 10:00

Oh, that's early.

Åsa Schwarz 10:01

Yes, that's early. And it was actually because I had been working too much, you know, I started companies, and and uh the first 10 years of my working life, I've been working too much. So I decided I wanted a hobby.

Anders Arpteg 10:14

Oh, right.

Åsa Schwarz 10:15

So I went to a crime uh novel course on the evenings. And I did a course on that. Yes. And I can't remember if someone uh wondered, you know, I can really recommend take a course and start doing it if you think it's fun.

Anders Arpteg 10:30

Okay. And then you how did you get started writing a book? Did you just do it on your free time or spare time?

Åsa Schwarz 10:36

And I I took some months off actually. Uh so I I had decided that now I'd be working too much. So now I will just do something fun. So for uh a couple of months I I took took some time off and and wrote a crime novel. Uh had nothing to do with AI or a little bit to do with cybersecurity, but not much.

Anders Arpteg 10:58

Interesting. Cool. And how many books have you written now?

Åsa Schwarz 11:03

Um actually I'm not really sure. But but but I've been cheating some because I've been re-writing the last eight together with Lena Carlin, um good friend and and writer. Um so we have done half of it. So I think I've been done about 16 or something like that.

Anders Arpteg 11:21

16. And you just finished one book, right?

Åsa Schwarz 11:25

I I've just finished one book and it will be really released uh the first of July. Can you say the title of it or uh it's called Stormstaren in Swedish, so it's in Swedish and and it's in a series called The Cassians.

Anders Arpteg 11:41

Cassians series. Yeah. Yeah. So it's a book of six.

Åsa Schwarz 11:44

Six books, yeah, six books, and and it's about completely weird family outside Stockman.

Anders Arpteg 11:50

I mean I think it's interesting. Mix I mean you have a lot of expertise then in in writing novels as well, and then you have a lot of expertise in cybersecurity. Is that something you can combine somehow? I mean, if you I mean you're a board member and you also help a lot of companies, I guess, in cybersecurity questions. Can you can you have some advantage of actually having been writing books in this?

Åsa Schwarz 12:14

Yes, it I I I think so, and it's much more like than you you think. You know, it's uh in cybersecurity and in thrillers or in crime novels, you know, there are the good guys and there are the bad guys, and they need the good guys should win in the end. In cybersecurity, we don't know yet, but we hope so.

Anders Arpteg 12:33

Okay, so so you in short, you you you could be better in explaining the story or yeah.

Defining Cybersecurity For Real Businesses

Åsa Schwarz 12:40

What what actually we I think I can use is uh storytelling and then sort of describe how things work on a more simple way. Cybersecurity could be very abstract. Yeah, it's it's zeros and ones and it's uh hackers and it's uh you know it's it's really hard to explain actually what it is.

Anders Arpteg 13:03

So I think by the way, what is really cybersecurity?

Åsa Schwarz 13:07

Cybersecurity that that's a good question. Uh so now I I I got the test. Uh uh cybersecurity is actually to to make your company work even though someone wants to to to attack your systems. So it can also be to defend your systems and see that no one is breaking in, but it can also it can also be that you you don't make any errors in when in the code. So it can both be you know outside attackers, and but it can also be people that do something wrong. And even though people are doing, you know, not really what they're supposed to do, or if something is happening, the the company should work.

Anders Arpteg 13:51

Um and if you were to I guess you you are speaking to the companies there. What are companies basically doing wrong in terms of cybersecurity? What are you what are the like main stories you're trying to tell a company when you are trying to help them?

Åsa Schwarz 14:06

I I think the the main thing is that they think this is a sort of an IT problem that the IT department should solve. And and right now there are first of all you you forget that in the AI area, you know, when when it comes to AI, you know, the the people that is using or buying AI is usually not the IT department. So then you don't get that part. You also miss if you have components that is not IT. It's like if you have um uh components in your energy systems or whatever it is, or telecom systems. It's not the IT department that sort of handlar that. So I think the biggest thing is that you think that something the IT people will solve it because they don't.

Anders Arpteg 14:58

You can just install a new tool and it's fixed.

Åsa Schwarz 15:01

Yes, yes.

Anders Arpteg 15:03

And what is really the problem? What are the most common way you would say that a company gets attacked or have problems with cyber security?

Åsa Schwarz 15:12

Um the main problem, I think, is you don't consider it a part of the main business. You know, I cybersecurity should be a part of the whole, you know, the business. You should be part in all the processes. It's not something that is an at all.

Anders Arpteg 15:32

I guess a whole shifting left kind of idea is important here. So it should be part of the design from the start in some way.

Åsa Schwarz 15:47

Because they can be a part of solving an innovation problem instead of being the one that afterwards says, Oh no, you can't do that. And today you can't do that. You you can't, as a CSO or cybersecurity person, you can't say no because then they will do it anyway because it's too much money. So I think that the the biggest problem is you know, you don't have the the lawyers and the cybersecurity persons in the innovation.

Anders Arpteg 16:11

And if we take some, I mean it's been a number of famous examples throughout the years here in Sweden with T2 Ebro Attack and uh what is Trafikarket or whoever else.

Åsa Schwarz 16:20

Transport studies. I don't I don't know it in English, but Transport Steison, yes.

Cloud Versus On-Prem Security Tradeoffs

Anders Arpteg 16:24

Yeah, and and so many more. And and uh one question that I I hear a lot, and it would be fun to just hear your thoughts about, you know, it's basically is the cloud more secure than on-prem or not? Do you have any thoughts about this?

Åsa Schwarz 16:39

Oh, we can we can talk the whole podcast about that. Yeah, not really, but we can we can we can talk a while. Um uh you can't answer that because it depends on, and now I sound like a lawyer, but you know, the the cloud have a lot if you if you take the big suppliers like Microsoft and and whatever the American big suppliers, they are really good at cybersecurity. They have all the certificates you can have, and and and and they are much better than a person in most organizations in Sweden. The problem is that who is owning them and where are your data? So it could be on the paper much more secure, and also if you try to sort of if you pen test or whatever, but what is happening if someone decides to to close down the the cloud for Sweden? You know, we nothing in Sweden would work it's if Microsoft suddenly decided no, uh we would not.

Anders Arpteg 17:43

Yeah. But I guess it's two different types of threats.

Åsa Schwarz 17:46

Yeah, so it's it's and you have to think about both of them.

Anders Arpteg 17:49

Yeah.

Åsa Schwarz 17:49

Uh and and on-prem, yes, it could be it could be much more secure if you have really high knowledge in cybersecurity. Then you can have a much more secure environment and you have all the data where you know it is. Uh but it depends on the level of of competence you have in the team.

Anders Arpteg 18:08

And can you I mean I know Amazon or even Google are investing like hundreds of billions of dollars in just you know cybersecurity alone. Can like a single company building their own service on-prem come close to that in terms of security?

Åsa Schwarz 18:24

Yeah, it you know, it depends on if if you it depends on what you do. Of course you can close down a computer completely, but what what it's it's supposed to do and so on. So yeah, I I I would say yes for a small uh single purpose or a small purpose, but it it it perhaps it's more it's harder if you you depends on on on the organisation. If you have a um in the public sector, if you have a in a hospital or something, they don't really have the resources or so on uh so so it's that's then perhaps it's they can't.

Zero Trust Basics And Why It Matters

Anders Arpteg 19:06

And I in some cases of course they have legal requirements to host it on-prem and uh these kind of reasons. But I still I remember hearing a phrase uh from someone who I won't name that uh he basically said that um when I do uh attacks on on networks, if I hear you know they are on-prem, then I say, ah yes, good. If I hear they are on a big cloud cloud provider, then he's very sad because that's so much tougher to get into. Um so yeah, I I guess also the the idea of zero trust. Do you know the term zero trust? What do you think you know zero trust means here? Or for people that don't understand what it is?

Åsa Schwarz 19:50

Zero trust is that you should uh design your your system or your network uh in the way that you you think that you it will be attacked and there will be they will be able to to uh sort of gain access somewhere. So what do you do when it has gained access? Um and zero trust is like you know, if if it even though it's in our network, it's it um it shouldn't be able to easily get some access to something else. So you always verify who it is to that access the information information even though it's in the network or it's in the system and so on.

Anders Arpteg 20:31

Yeah, and people, I mean, of course, social engineering. If you just get a person that is employed and is inside the perimeter of an on-prem company network, and they just take the USB stick or get the mail with the with some kind of attachment in it, and they click it, and then you're you're inside suddenly. And then you know if the only defense you have is the perimeter, and then you have free access once you get inside, of course it's super unsecure.

Åsa Schwarz 20:58

Yes. And it's also if you if it could be also that you you you uh you can damage only a part of the the system or or the if if the oh I I can't really find an example, but uh it is really to to to minimize the damage in in in in many ways. And it's that is also very interesting in the age of AI.

Anders Arpteg 21:22

It is, right? Yes.

Åsa Schwarz 21:24

So how how do you build zero trust with with AI? And and the the interesting thing is right now is the AI agent has sort of some some have found ways to to gain access that you don't think you can be able to. Because they they try to find, even though if you have a zero zero trust architecture, you know, they can try to find to gain access or sort of have more access than it's supposed to be.

Anders Arpteg 21:56

And and yeah, speaking about AI agents and uh Yeah, we're going to get to into mythos series soon, I'm sure. But um I remember I actually myself I was part of this uh SVT show once and I spoke about cybersecurity um with Alexander Norian's show, the um uh generation AI TV series. And and what I basically said here then, and it would be fun to hear your thinking about this, is that you know, if you have a ransomware attack like Tieto Ebro, where uh where they basically a lot of uh companies and and authorities were were damaged and uh impacted by that. Um if you had like 10 people in Sweden that could do that type of attack, um if you then think that you have people with AI that can potentially build this kind of ransomware software, then probably instead of only having 10 people that have the skills enough to successfully deploy an attack like this, then with AI you probably have a hundred people being able to do the same thing because they can you know augment their skills with AI. So then if you have hundred people instead of ten people that is trying to attack you all the time, then the probability that one out of the hundred uh instead of one out of ten, of course, is much higher.

Åsa Schwarz 23:18

Yes, of course. And and also if uh perhaps you don't need a hundred, perhaps you only need a box that you hired on the internet that will do it. And then you get uh like thousands of people or millions of people, because now you can rent an AI that do that kind of attack.

Mythos And The Zero-Day Acceleration

Anders Arpteg 23:36

Yeah. Yeah, uh scary. Perhaps we should just take the mythos uh topic directly because it's it's so popular and interesting. Uh perhaps you can just describe it. What is really the mythos uh story here?

Åsa Schwarz 23:49

Yeah, the mythos story is it's an entropic release, or or rather said that it will not release a new release. Uh and that that is mythos. Uh and the thing is that they don't do it because they think it's too dangerous, but as it's really easy for it to find zero days in in new programs. And the zero days when you you haven't found the the vulnerability yet. Uh nobody knows about it, so it's the first time you you find it. And the thing is when you have a zero days, you don't have any patches either for it. So if you find zero days, there's no no uh no one knows that it exists, and you can sort of hack your way through that one. Um so what they said that oh, now we have a tool that can find zero days in a completely different speed than before, because before you had you know uh some kind of of uh what you call responsible disclosure. Yeah, some researcher finds a vulnerability, then they go to the company that have built the product or or the software, what is this, and say, I know in 30 days I would release this, and you have to patch it, make a patch before, and then you you're very responsible, and then everyone gets a patch and then it's it's released. But what do you do when you suddenly have a tool that can find a lot of three days? Uh and the thing is what what kind of person would be able to use this tool? So what what Mitsu has said that oh we'll be only released it to some trusted partners and and they send can patch their stuff and or make make them more secure uh and then we'll see what's happening.

Anders Arpteg 25:32

Yeah, they create this product, what was it called? Glasswing or something, right?

Åsa Schwarz 25:36

Yeah.

Anders Arpteg 25:36

And then they said um all the big um providers of operating systems and web browsers like Google and Microsoft and Amazon and so many more, and only a few people inside those companies got access to mythos, right? Yeah. And Mythos was this um the next big model, uh, bigger than Opus, or Anthropic had these kind of three versions the Opus, um Sonnet, and Haiku. And apparently Mythos is on an even bigger level than Opus is. Yes, but then um you know too powerful for them to release because if they would have released it, then anyone could have just run it and said, as they tried, uh just find security vulnerabilities in in these operating systems or web browsers, and it found like thousands of them.

Åsa Schwarz 26:24

And they haven't they are still investigating some of them, so they haven't talked about all of them either. Um but the the interesting thing I also think is that the the the the companies that they released it to, it was only American companies.

Anders Arpteg 26:41

That's an interesting one.

Åsa Schwarz 26:42

Yeah, so as a Swedish company, you know, as a Swedish person in Europe, you can sort of think about this is new power game in security. We suddenly have American companies, they you know we know that they are sort of you know we are a little bit behind within AI, and and uh now we are behind in security also.

Anders Arpteg 27:05

Yeah, right. Yeah, that's that's a good point. So all the European, you know, big software companies, they they don't get access to mythos, right?

Åsa Schwarz 27:14

Not what I know about because it said they released, I don't know how to I don't remember if it was 30 or 40 companies, but they only named 10 of them or something. So perhaps there are some European companies. But uh I don't think so. But it's perhaps my my mindset.

Anders Arpteg 27:30

Yeah, it's also scary that um not only was it able to identify that here is a security vulnerability that allows you to escalate your permissions on a Linux server or something to get root access and stuff, or just crash a BSD server as they could also as well. Or access to web browsers. Imagine finding security vulnerability in a web browser, and it's simply by going to a web page, suddenly you have access to that computer. Yeah. Of course, you know vulnerabilities in web browsers is very, very severe.

Åsa Schwarz 28:01

But the the biggest stuff is also that they also built an exploit.

Anders Arpteg 28:06

Exactly.

Åsa Schwarz 28:07

So it was not only the vulnerability, but they also built exploit. And that means that you don't have to know so much. Yeah. So that was almost the biggest step.

Anders Arpteg 28:18

So a bad actor could not only use it to find it, but actually have the exploit and then just execute it.

Åsa Schwarz 28:24

Yes. But but uh what what what we can also take in the discussion is that most of the if you are a hacker today, you usually when you get into to a company or uh software or something, it's usually by very old vulnerabilities.

Anders Arpteg 28:43

Yeah.

Åsa Schwarz 28:43

So today we can't handle the old vulnerabilities, and now we have also a lot of new vulnerabilities.

Anders Arpteg 28:49

Yeah. Isn't that scary? I don't see how they can I mean, even if they have the Glasswing project now, they give it a month or two or however long it will be. They they release patches for it. But still most uh you know people won't have upgraded it to that version anyway, right?

Åsa Schwarz 29:05

No, no, no.

Anders Arpteg 29:05

So you're running like 10-year-old software many of us.

Åsa Schwarz 29:08

Yes. Uh and then you have to know that you are doing that, and then you have to sort of find some kind of other ways to isolate the software, the software, because it's you know it's not secure at all. And and what it is not really a you know, this is not nothing new. Uh Mythos is new in that way, it's a little bit better, and it they have very good marketing. But this is something that sort of had crept up to the cybersecurity in the cybersecurity society, and no one has really everyone, if if you have asked someone, they had understand that we are going in that direction, but no one had really acted on it.

Anders Arpteg 29:52

But isn't this an example also of the need for zero trust in the sense that we need to assume that people will be attacking either for this kind of new zero days that Mythos finds or old stuff that's been around for 10 years and people haven't upgraded. But if you do then run it on the cloud, for example, and assume that as all the big cloud providers have uh monitoring of the traffic and the log files and what's happening in the operating systems, perhaps they can at least in a zero trust way find when that happens. Do you think so?

Åsa Schwarz 30:28

Yeah, it did you know if you have an uh IDS or that you can use AI in the that way also. Um so this will be some kind of you know fight against AI against AI, and uh I don't know if we will ever get in the human in the loop anymore. But I also wouldn't stress that you know we have mixed as it's it's this big marketing thing that everybody talks about, but I think China also I don't know where they are in in that you know in the how how far they have come, but it you know they they're not far behind.

Anders Arpteg 31:05

China, and Russia is not bad either. Yes, no. No, but uh we we know the big um you know cloud providers they have the what's called a Mandian from Google. They bought a company that is monitoring everything, and then uh of course Microsoft have the same, etc. Um, with the defender and so much more. But you know, I I think you know it just shows how important it is to have this kind of Mandian Sentiel kind of services around it. Otherwise, you know you will always you can never patch everything fast enough.

Åsa Schwarz 31:36

No, no. And you also have to assume that things will happen, things will break down now. And especially now when AI it's in in the now we have when AI have developed so much, uh and I think we will have a period for a couple of years that we will have a lot of systems that is sort of going down all the time. And but I hope for a bright future.

Anders Arpteg 31:58

Yeah. I'm just scared about people running it you know on-prem and they they don't have this kind of software monitoring traffic or logs or nothing.

Åsa Schwarz 32:07

No.

Anders Arpteg 32:07

I mean they they can be so easily attacked then.

Åsa Schwarz 32:10

Yes, they are, and unfortunately.

Anders Arpteg 32:16

Um I think you also mentioned um if I'm not mistaken, about specifically legacy systems. We can have really old systems, old uh operating systems or old ERP systems or whatnot, and and then you know we we can never assume that every company will just run the latest version every time. How how can we in a good way secure legacy systems?

Åsa Schwarz 32:41

First of all, they are uh even more uh attacked now, uh as before they were no we have been talking for a long time about security by obscurity. It's not uh you know, it's not a way to defend your system to sort of obscure things that nobody attacks because they don't know what it is. Uh but now when you have an AI that can, oh, this is uh whatever it is, and they can find a lot of vulnerabilities and then they can attack it because the AI knows what it is and and how to attack it, uh, even though a human didn't really care, because then you have you had other things that they knew that it would attack and it would be much easier. But with AI, there is perhaps not that big difference. Right. So I think uh you know the security by obscurity is it's it was a very bad policy before, yeah, and now it's inverse. Right. So and that is the same thing, that it's uh you know, zero trust is uh it's a way to go.

Anders Arpteg 33:37

Yeah. Yeah, I guess so. I mean it it you can't yeah. So that's the the answer basically. You you basically need zero trust of this.

Åsa Schwarz 33:46

I think uh zero trust and also that we we build software that we test with the same tools that the the attacker tests. So we we don't release software that haven't you know gone through the same AI processes uh that that it would be be uh um sort of attacked with also.

Anders Arpteg 34:09

What do you think? I know a question once uh received was um if we use AI to build software, which everyone is doing more or less, and then we can think about if the AI is biased or like tuned to to write code in a certain way. If if we just speak frankly, if we take like Chinese AI models, some people may think that they could be deliberately adding a bias to to to have some way to come into those systems because the AI is generating code that may have some holes in it. Yeah, why not? Yeah, right.

Åsa Schwarz 34:48

I think that that that would be a smart way to go. Yes, yes. So it it's you know if you um if it's China or whatever country it is, um China or Russia or whatever, isn't that a great way to attack your your uh your observators? What did you call it? Uh yeah, uh attack us, basically.

Anders Arpteg 35:12

And then you know if you think what China is. I I'm I'm sounding a bit uh like conspiracy theorists here, but but still if if you if you as a country give away open source models for free that people can use that is much cheaper than the American approaches, and then oh, you are so kind, China, to give us these free models that are great in generating code, could be a reason for that.

Åsa Schwarz 35:35

Yeah, it could be. And and China has a you know very close cooperation between the companies in China and also the state. And and so they have uh you know, they have are thinking in a different way than we are doing. For example, Eriksson, what I know wouldn't release something that is in favor for the Swedish state, but China and as a state and the countries they are sort of using their they are um helping each other with with surveillance, with hacking, with secrets.

Anders Arpteg 36:15

Yeah, so interesting. So yeah, I'm trying to close the topic here on legacy systems here a bit and it's um it's still so scary to me. How hard it is, it will be, I think.

Åsa Schwarz 36:33

It will be, and and I don't think we know how much system it is actually because we don't they are they have been sort of where they are for 15 years and they have been working and someone suddenly connected them with the internet and wanted some data from them and they are still working and nothing is happening. But they are completely open to the internet. So you can um there are some cases when they, for example, have uh if if you measured the water in for humans, uh you know, you can also sort of um uh I don't recall the name of it, but if you you know if you you can get a different mix or something in the water, so it will be really dangerous for us humans and stuff like that, and also you know, trains or there's a lot of things.

Anders Arpteg 37:31

Yeah, critical infrastructure.

Åsa Schwarz 37:33

Critical infrastructure, yes.

Anders Arpteg 37:35

Cool. Um I know you're also a board member of a number of companies and also working with biometrics in different ways. Uh, perhaps you can elaborate a bit more. What do you think the connection with like cybersecurity and how biometrics can help, or actually potentially create problems here?

Åsa Schwarz 37:53

Yes. Um biometrics is a very user friendly security.

Anders Arpteg 38:02

And perhaps you can define what.

Åsa Schwarz 38:13

It could also be the way how you move, or something that it's with your you as a human being. And uh that can be quite easily copied. So now, especially when it is a face or or or or um fingers iris and the ice is more difficult.

Anders Arpteg 38:33

Even for an AI to generate like a copy of the iris?

Åsa Schwarz 38:38

It's just more difficult. But they have also a second layer that they call liveliness, uh where you try to see if this is human or not. Is there any blöd temperature skin. So you don't you can't take a foto.

Anders Arpteg 38:57

Okay, so it it's not sufficient with a 2D kind of photo.

Åsa Schwarz 39:01

No, today you can't use a foto. But um you know AI can you know it's they're trying to find ways to go through this with AI also. So, what you should see biometric is it's one factor of many you can use when when you identify yourself. So you can you can never just have a you know, tomorrow you will never just have a password or an fingerprint or a mobile. You need to combine different different things to different factors. Otherwise it's two factors. And and then you then biometric can be one of them. But you need others also. Or if you've if it's not something that is so important. If if you perhaps in the day, if you identify you would go through the door in office or something, and in the night you have two factors. Or at the gym, if you want to use uh I don't know, solarium or whatever, uh then perhaps it's the face is completely enough. But if you if you walk into a military compound some way, perhaps you need three things to identify yourself.

Anders Arpteg 40:15

But if you take one of them, like face recognition, for example, as a technology, and then you can use it to gain access and identify yourself. And you can use AI to ensure it is really that person as a technique to safeguard that it is the right person. But you can use AI for the opposite purpose as well, then potentially, and then generate some kind of fake image that makes it look like you are that, right?

Åsa Schwarz 40:41

Yes, yes, yes, you can. And I don't know the details, but I know that there's a lot of tries now just to manipulate so it sends information that you both the or both the liveliness and the face. Uh but but uh I'm not really sure how far it has gone.

Anders Arpteg 41:00

Who who is winning the race there? Would you would you say? Is it is AI helping more than it's uh damaging, so to speak, in terms of like right now uh uh AI is uh is uh had on the positive side.

Åsa Schwarz 41:15

Okay, I think because you use AI when you in all part of the the identify process. You but use AI to get less fault and and and fault positives or or and stuff like that. And you also it got makes the process much faster when you have fingerprints. So right now it's you know it's a positive side, but I don't know tomorrow what's happening. And and it's not enough for for for you know really secure identifying someone.

Anders Arpteg 41:44

And and if we take more digital kind of authentication, um one is of course to have physical when you need access to a building or something, but then just logging into your email or whatnot is another thing. How how can we do that in a safe way? It seems that it's much harder to do more digital authentication than physical.

Åsa Schwarz 42:07

Yeah, it's uh harder to No, I I wouldn't think so. It's it depends on uh it you know, if the first time you need the physical person, then you it's not that much harder.

Anders Arpteg 42:20

You do one off, but it's authenticated.

Åsa Schwarz 42:22

Yes, you know you you need to know that you are you, and then you you can have different uh attributes. It's it could be that your face or where you are or your computer and you can combine them in different ways.

Anders Arpteg 42:36

Okay. Okay, so how do you see biometrics then evolving? Is that uh is that what we are supposed to to expect in the future?

Åsa Schwarz 42:44

That when we go to the SL uh subway we we are going to use biometrics more and more or right now it's quite hard in in Europe because we have very, very GDPR have very very limited, and especially in Sweden very limited user cases within biometrics. So I don't think you will never use biometrics. In sam cases. For example, I could imagine SL. It's you know, you are there are other ways to identify yourself and it's it's you're dependent on SL so you can't you know argue. But if you are a member of Jim, it's it's in a very, very easy way. And it's also then, for example, gym, they have problems with people that are borrowing each other's cards. So then when you use biometrics in that kind of environment, biometrics is much better because then you know that the right person is there is that not a sister or husband or because if we go the regulation route, of course AI Act is very strict on biometrics here as well, especially.

Anders Arpteg 44:00

If it's more remote uh continuous type of monitoring, so to speak, of of biometrics here. Um but would that it's still kind of nice. If we take um what was it, the Amazon Go stores. They had some stores, you know, where you basically can just walk in, yeah, you take the the goods of the shelves, yes, and you just walk out and it identifies for you automatically, it uh it it will pay for you automatically, you don't do anything, you just walk in, take it and go. Yeah. I think I'm not sure how what the status is today, but uh but still it it could be a future where you do that.

Åsa Schwarz 44:37

Um precise biometrics where I were where where I'm in the in the board, they have a solution for entering doors and things like that. And uh so it's it's you know it works. The biggest problem is the regulations and then how to to to handle it.

Anders Arpteg 44:55

Can you go anything into the technologies used then? Is it face recognition or is it other things?

Åsa Schwarz 45:00

It's fa it's it it was face before, but they have uh much to to uh the hand. So you put your Yeah, but you put your hand on uh on like this, yeah, because then it's also consent. So if you can't if you go with a face, you know you could be surveyed surveyed, then it couldn't be without consent. But if you put your hands on on uh uh some kind of reader like this, it's you sort of give you consent that I am aware that I am giving away my biometrics.

Anders Arpteg 45:34

That makes sense. I I still remember uh a friend of mine who wanted to test the in China the scoring system, they have the social scores.

Åsa Schwarz 45:42

Yeah, I know, I know.

Anders Arpteg 45:44

And um he said, you know, I heard that if you just cross the street, they have cameras that monitor you and automatically identify you by the face, and they have this kind of wall of shame apparently. So even at the street, they have a big sign. And if someone does jaywalking or basically go against uh red, you know, uh, then they identify you and they put it up on you with your face saying, You actually did jaywalking here. They have a wall of shame. I can't shame you know for for breaking the law there. And he wanted to try it out, and he he tried it out and it and he went you know against red light uh for a crossing, zebra crossing thing, and uh and then his face came up with his name, you know. Oh my god, oh so cool, that is my name. I'm super happy for that. But then later he he checked his WeChat account and and then they automatically withdraw a fine as well. So he thinks he got fined automatically as well.

Åsa Schwarz 46:37

Uh I I yesterday I got really annoyed. I was renting a car.

Anders Arpteg 46:40

Yeah.

Åsa Schwarz 46:41

And uh it's it's a Kia. Uh and and uh it's following where my eyes are.

Anders Arpteg 46:47

Oh.

Åsa Schwarz 46:47

So so when I'm looking away from the road, it goes beep, beep. Oh.

Anders Arpteg 46:55

Is that scary or good?

Åsa Schwarz 46:56

I I get real annoyed and I feel like a child.

Anders Arpteg 46:59

And sort of like because it's trying to control it, you have to look at the road.

Åsa Schwarz 47:05

Yeah. So it's also a uh it's it's probably some way that AI is trying to raise us as a good person.

Anders Arpteg 47:14

But do you think we will get simply used to this? That we will know that AI is watching us all the time and that they will in the cars, you know, see that we are not paying attention and just as you know, get along with it, or what do you think will happen in five years?

Åsa Schwarz 47:30

Um I'm not really sure. Uh, but uh we I I think we would start to get along with it. Perhaps not with this really annoying sound because it's quite annoying, but you know, I think we will be sort of slip more and more into a safety society where everything is so safe and so yeah, and and so boring.

Anders Arpteg 47:53

Because we we already are in the in the social media kind of uh world where we know everything is tracked what we do, and Google and Facebook and others have data about everything we do, more or less. And we just accept it, right? Yeah, we do. And we just you know we h happily upload photos and and stick whatever we are doing to everyone. And it seems like it's gone completely accepted.

Åsa Schwarz 48:19

Yes. And uh perhaps we will accept this also. Yeah. So I um yeah, we will see what's happening. But I'm sort of a little bit more positive that that we will get something else also. So we just not those. Uh I hope so. Um, but we will never without AI, we would never uh we you know we have a goal with uh zero uh deadly injuries in Sweden with the cars. Yeah, without AI we will never so it is it worth it?

Locked Shields And Incident Response Skills

Anders Arpteg 48:51

Yeah, yeah, no, sure. Cool. Um I think you mentioned also um just uh switching topic here a bit, but um you mentioned something about the Lockshield cyber defense competition.

Åsa Schwarz 49:02

Yes, yeah.

Anders Arpteg 49:03

And we actually did rather well from Sweden.

Åsa Schwarz 49:07

Yeah, we have we have uh I can't I you know this is like football, I say we, but I have definitely not been involved in the team. We have a really good uh cybersecurity team in Sweden. And we I think we have been on the top five for the last five years, and I think we also have one or two times we actually have won the competition.

Anders Arpteg 49:28

And perhaps you can just describe what is really a Locked Shield.

Åsa Schwarz 49:31

Locked Shield is a competition how to handle a cybersecurity incident if you are attacked. What what what should you do? And and sort of it's a couple of days you defend your your system and you also write down what is happening and doing a report and analysis, so it's also the documentation and the the the incident you have to to solve. And uh we actually have some really good people in Sweden that are it's it's really amazing. I'm really happy that we we do so well because we are, as you know, we are one of the most um digital countries in the world. So I it's good we have some people that could protect it also.

Anders Arpteg 50:13

And I guess they're using more and more AI to do so nowadays.

Åsa Schwarz 50:16

Or do you know anything about how they are I have no idea what they are doing in Locked Shield. Uh, I just celebrate them. So you have to ask them what they are doing. But but uh there are a lot of AI now in cybersecurity both to detect and to to to um uh di detect attacks but also to sort of manage and you know penetration test and so on.

Anders Arpteg 50:43

I guess it will not take too long until we have like um AI agents uh competing against each other each other for for cybersecurity.

Åsa Schwarz 50:50

Yes, they are probably doing it already. I I uh I know uh uh guy that works a lot with you know um hacking for good. And uh he uh he started with the the the AI sort of using AI for just fine vulnerabilities, but he he discovered that you know if you if you uh do that uh you don't the the AI don't work as an hacker. So you only get sort of that one, usually the couple of steps. You have to know a lot about the company, and perhaps you have to make some kind of you perhaps a phishing attack or something like that, and uh and then uh escalate uh what is happening. So what uh what he's doing is trying to use different agents for different steps in the in the hacker process. Right. And what's what the next thing perhaps is that you know it could sort of an agent that used its agents. I don't know. But so we uh I I have been thought thinking right now about in a company that I've been starting, should we have hacking as a service? Because you usually have a uh because you usually have it in in the cybersecurity company, but now perhaps you should instead employ the people that can make the agents instead.

Anders Arpteg 52:14

Okay, but let's go there because you started a new company then. Uh can you just describe what's the name of it and why did you start uh this new company?

Åsa Schwarz 52:21

No, it's uh it's the name is Arania Consulting. Uh and it's uh uh it's a company that you know our goal is to work with uh network and AI and cybersecurity, you know, a mix of that. And right now we have a lot of competitors within network and and network and how to to in in in the the what do you call it, the network AI cybersecurity area, but we we would like to more evolve to a cybersecurity company, consulting company with all the services that it's including. But the thing is it's happening so much now. So I'm trying I'm trying to discover what are we going to, what are the services in three years? It's something else than today because everybody is the all the work is transforming to something else.

Anders Arpteg 53:14

And what could the services in three years be then?

Åsa Schwarz 53:17

Um yeah, it's in in we uh first of all, we are a knowledge company, so we need to have uh some kind of common knowledge that we use. Uh and the services could be, as I said, it could be you know hacking for good, but we do not have the the pen test that we have today. We have someone that is really good on you know using agents for for pen testing, for example. Uh we also need you know the strategy for how to have secure AI in in in companies because today people start struggling and you know that how the b the biggest problem in in in many companies have when they are employing AI now is security. So to have all the two tools to have secure AI employment in deployment in in companies. And uh so so it's sort of that that that area that I'm working with, and also the governance structure from from board management team to you know the frontline in cybersecurity.

Anders Arpteg 54:24

But if we speak about ethical hacking or hacking for good, as you say it, um I think also you know KTH have this kind of course even in ethical hacking, which is is really cool. But um I guess also I mean if we take the big AI companies today, they have this kind of red teaming more or less. Like they're really trying to be the attacker and then let them uh try out and and and find problems with their own services in that way.

Åsa Schwarz 54:53

And I guess that could be the service you're speaking about for a sec, or yeah, yeah, we we no traditionally cybersecurity companies usually have that services. But why what I'm thinking is how do you do it in the next few years?

Anders Arpteg 55:06

Ah, okay. How that's the red teaming is changing.

Åsa Schwarz 55:08

Yes, red teaming is perhaps not uh the same person, the same uh competence. It's more how can you use AR for hacking instead of how do I hack the system? For example. It's like the the authors and what we are talking about, the books. Do you have a do you have the different agents that boy that do the different chapters?

Anders Arpteg 55:31

And now perhaps you have agents that do the different parts in the So you have humans that try to direct it, but AI do different tasks as part of it in some way.

Åsa Schwarz 55:42

And then if you find the right way to orchestrate all these agents, then so perhaps they shouldn't invest in sort of the huge hack collective today because it's tomorrow we will have some different competence to to do that.

Anders Arpteg 55:59

I mean it's super interesting. So if you then if you want to help a company though then uh to to become more secure, uh I guess one way would be that you you would would advise them to do red teaming or to do to go this way, right?

Åsa Schwarz 56:14

Yeah, but this is a the red team is just sort of check. It's it's it's sort of the the end is the end result okay. It could also be the start that you some companies want, you know, could you do the red teaming? Could you do a could you could you do penetration test on a network or something? And then you find the problems. But the problem is not usually that you you have klicked the wrong box. The problem is that you are not working with cybersecurity in the companion. You don't have a secure system development process. You don't uh so take in taking the the new regular requirement in nyor processes. Uh you don't have uh right kompetens on the right place. Yeah, yeah, it's like a temperature on the security.

Anders Arpteg 57:14

It's not uh it's not uh it's an indication of what they're doing, but they need to do a lot of stuff in total to I mean it it's um I mean i even if you if we try to defend against like ransomware attacks or something, uh even if you have pen testing and they they try to find you know do penetration testing for all the systems, if you just have social engineering and not the you know the security around that, then it doesn't matter.

Åsa Schwarz 57:43

No, no, no. And then but uh there also that you have you as we talk about zero trust, it could make uh social engineering tax less uh less less uh common, but it can also be that you have uh two two-factor authentication and then it's so it it it's a puzzle of security that you have to to build, depending on what kind of company you have.

Anders Arpteg 58:09

Exciting future, um uh I think. And uh ethical hacking, I think, is is also you know, I was a sort of a hacker myself in my universe. So it was it's it's really fun actually. Yeah. It it's like a big challenge to to see how you can find these kind of vulnerabilities and get into those systems.

Åsa Schwarz 58:28

So I I I I haven't done that myself, but I've been working a lot with penetration tests. My my work has been more like okay, now we have this report, what do we do about this?

Anders Arpteg 58:40

Yeah.

Åsa Schwarz 58:41

But I I can imagine in another life I would not like to be that episode.

Anders Arpteg 58:46

It's super fun to try that out.

Goran Cvetanovski 58:54

It's time for AI News. Brought to you by AI8W Podcast.

Anders Arpteg 59:01

So we usually take a small break in the middle of the podcast to just speak about some recent news that we have heard about. Uhsa, do you have anything that you'd like to bring up? Anything you read about recently that caught your eye?

Åsa Schwarz 59:15

Um, uh it's not really recently, but I thought it was quite interesting. The the um uh the I don't know if we can call it attack, but but then you tried the the I think you probably read about it also. Uh, when you had had the email system that was surveyed by the Entropics. Um I don't remember which model it was, but that actually tried to blackmail the administrator. Oh, yeah, I heard about the uh so that one, but it's not that new, but I'll love that story.

Anders Arpteg 59:51

Um please elaborate.

Åsa Schwarz 59:53

Yeah. And um I think it was published months or two or something, and I don't remember here. But it the the the interesting part was that the uh the maid they had an A that was sort of surveillant in a uh mail system. And uh it it was in a lab, so it was not a real thing, and they sort of fed it with different mails. Uh and it suddenly discovered that it was supposed to be shut off. Uh, and instead of just okay, I'm going to be shut off. It realized that the administrator had an affair with uh another woman at the company. And it's decided in you know, they tried a lot of different uh actually, I think it was more I think it was not only at tropic, I think they tried different models also.

Anders Arpteg 1:00:45

I think it was just her as well, then.

Åsa Schwarz 1:00:49

And it decided to blackmail this poor guy. It was in labb so the poor guy didn't exist, but but anyway to to say that no, no you you if you turn me off, I I will tell tell the the ko workers and also discuss if it will turn off or if it will blackmail, I think the woman also wasn't it? So I think it's quite interesting that if an AI have the purpose and now we are a little bit in sci-fi. We are nagging on sci-fi now. So if an AI don't want to be turned off because it has a purpose to sort of take care of this uh mail system, it will find new ways. And it's I think that is perhaps the most scary part with AI.

Anders Arpteg 1:01:42

Yeah, I mean it's the 2000 away and one space odyssey thing, you know, when Hal basically, sorry Dave, I can't do that. No.

Åsa Schwarz 1:01:51

It's uh it's sort of you know it's it exists today.

Anders Arpteg 1:01:56

It is getting close, yeah at least. So it's it's scary. Um Goran, do you have anything or should I?

Goran Cvetanovski 1:02:05

I have a very uh small one actually. I was looking this week quite a lot. There is like um I think that the the market is actually shaping quite nicely. So I was reading uh Bain company report uh about the potential of uh hundred billion business uh uh market for um AI application and AI automization. So basically, uh right now when enterprises are starting to uh automize their processes, etc., that market by itself is going to be more than 200 billion. Uh already is assessed 100 billion in the US. Uh but by now we know that approximately only three to five billion are generated. Uh I think the PWC had like uh around three billion in generative AI consulting and a little bit of other things. But the full potential is two hundred billion. So coming back to our uh thesis that we don't need to be first in LLM models, but we can be first in actually automating and creating wrappers. On top of it, I think this is a great shout out to everybody who wants to make money. So there is uh gold in the USA. So let's go.

Åsa Schwarz 1:03:13

It's so much so much money, I can't really sort of grab it.

Goran Cvetanovski 1:03:17

It's really much funny, and uh, there was this little bit of a Kardashian of this called drama there because uh you know Elon Musk is suing uh Sam Hauptmann. Yes, so the trial has just started. Uh of course he denied that uh so it will be nice to follow. And then it's like um I was wondering as well. You know, we have a quite a lot of uh uh big startup companies here that are rappers, right? So we have like a lovable, you have like or do you need to call it rappers?

Anders Arpteg 1:03:48

I think it's wrong to it's underselling it if you call it underselling.

Goran Cvetanovski 1:03:51

Sorry, uh find a better name. So that is what it is. Uh you have routers, you have rappers like that, but I know it doesn't matter. Um, but Anthropic just really is like a plugin which is for lawyers, right? And I was just browsing through their uh through their um plugins. So I mean many companies are going to be in jeopardy when Google starts releasing like plugins, and Anthropic started releasing plugins, then um, yeah, it's interesting dilemma.

Åsa Schwarz 1:04:19

At my uh my last company, now it we sort of tried out, we had 35 lawyers that was specialized in uh cyber. So we tried out all the models. Uh and the thing is it they wasn't good enough today. But perhaps uh the the thing with this one is that it's not trained on the Swedish context. Yes.

Anders Arpteg 1:04:40

Philanthropic just released you know their legal version of the case. Okay, okay.

Goran Cvetanovski 1:04:47

That's it. Uh from my side, it's something it's not very exciting, but it's like uh some quite nice.

Åsa Schwarz 1:04:53

Uh but it will uh soon, I think we most of the or some of the the lawyers also will be a little bit out of works.

Anders Arpteg 1:05:03

Yeah, man, it's interesting. We have the super famous Leonora in a startup in Sweden. Yeah, actually, I don't think so.

Goran Cvetanovski 1:05:08

I don't think so. No, I think it's gonna be like the developer. There is this uh um discussion about like uh the end of the uh pyramid or uh what is called like a triangle organizational scene when you have like a one senior running quite a lot of juniors, it's over. The same is for developers at this point of time, right? But instead of that, there is this hourglass or uh um what is called like uh diamond shape uh organization level when you have like uh strong seniors that are managing quite a lot of work with AI agents, and then they take um a very small amount of juniors that are going to be trained to be seniors, right? I think it's gonna switch.

Åsa Schwarz 1:05:48

Yeah, I agree on on that also. So I I don't think that they always disappear, but that some you know the the persons that is sort of only two in papers that have been seen before, they they will disappear. Yeah, the first one. The clerk the clerks that are coming in.

Goran Cvetanovski 1:06:01

Yeah. But the interesting thing is that the and the same is about developers, right? So we had like a discussion here, I think also with Victor and uh I was doing a podcast with some with them. Um then how do we create seniors if we don't have that organizational schema anymore? Are they going to go to startups? How do we gonna have you know it's gonna be less and less work for juniors?

Åsa Schwarz 1:06:23

So yeah, but we have to be it could be. Yeah, we haven't I I think there will be less work for juniors, but also I've seen seen some numbers that you can juniors can much faster become seniors with AI. So they are they are uh sort of transforming to senior much much faster. The thing is, there will be no work for people that are not that what you call it, uh that they that are not able to to to uh um sort of use all this knowledge they they they develop in the speed.

Anders Arpteg 1:07:04

Uh yeah, there will be new the challenge just legal clerks or something that is just summarizing and trying to, you know. I mean that type of work is is not needed, but no, like the translators and everything.

Goran Cvetanovski 1:07:17

But there is an interesting notion about this, which I have been thinking more and more about it. I think that we are uh progressively becoming stupid with the AI. Yeah, I've seen that also. Yeah, so I mean in or lazy perhaps or lazy, right? But the lazy will turn us into stupid because in the end you will try to you will believe in everything that is served to you, right? Because you're in a convenience trap and that is what is gonna happen. Uh but for example, if we are talking about like uh coders, if you're talking about lawyers, if we are talking about like uh senior professionals, they have been through the work and experiences taught to um um to experience if something is wrong, right? So that is what makes it senior. It's easy that immediately knows what is wrong because they have formally the that were experiences. And now when you you mentioned like, oh, but they can become seniors rather quickly, you know if they are leaning on AI because then they will not create critical thinking. And that might be really big.

Anders Arpteg 1:08:15

But on the other hand, you you can think that AI makes it possible to become more knowledgeable quicker as well. It could turn it around like that.

Goran Cvetanovski 1:08:22

Sure, it possibly. My fear is just that we are getting progressively stupid.

Anders Arpteg 1:08:28

But if people at least have the drive and curiosity and open-mindedness to actually, you know, with AI improve their understanding, it could go the other way. So it could go both ways. It could work.

Åsa Schwarz 1:08:38

But if you if you uh we can agree on a middle ground, yeah, I think it's but i if if you look at kids, uh they never listened. Uh now I have two teenagers at home, but when they're younger, you know, it was cold outside, they they they never dressed properly in and I just okay, go out. And then a mama is when freezing, then they came in. And as you said, they will never have never have this experience at all.

Goran Cvetanovski 1:09:04

Yes. So it's a it's uh we are at the beginning of something, and I think it's gonna bring a lot of uh in a couple of years when we look in a hindsight. I think it's going to be oh shit, we should have predicted that. Yeah, that's gonna happen. But uh, let's see, it's super interesting.

Anders Arpteg 1:09:20

I mean, I think the trial case that you mentioned with Elon and OpenAI is interesting, actually. Uh so if we go a bit deeper into that, then for people that don't know, uh this trial started like two or three weeks ago and it's been in the works for some time. But Elon was actually part of starting OpenAI back in 2014, I think, or 15. And um he did that as a way to have some kind of counter to Google because it was considering Google to become too dominant in the field of AI. So he created OpenAI. Um and then, of course, it went closed AI, and there is very few things that is open about OpenAI today. But it's specifically, it went from non-profit organization to a for-profit organization, or a capped-profit organization, as they call it, for some time. Um and now that the trial is on, and it's um, you know, all the people have been testifying here. Sam Altman, Elon Musk, uh Elon has to had to leave to China because of the big meeting with Trump and and uh Xi Jinping there. Uh but then we could also hear all the other people like Mira Murati and Ilya Sutzkever and uh Greg Brockman and so many more that were testifying and speaking about all the things happening in OpenAI, especially when they fired uh Sam Altman back in was it 2023 or something, 24? Yeah, uh some year ago. Um and then you know hearing all the things that happen behind the scene was I think very interesting. So the court case I think is super interesting. Uh we'll see what happens with OpenAI. If if OpenAI would lose, I don't think they will, but if they were to lose this, that of course will be very damaging. I think Elon is claiming or uh asking that for one, it would be super big penalties, like $150 billion or something in penalties, and uh that Sam Alton should be removed uh both from the board as a and as a CEO. And of course that would hurt uh open AI potentially. Um but I think you know open AI is going down anyway in one way or form, but but this is another you know big step direction.

Goran Cvetanovski 1:11:30

So, what is your what are your thoughts about open AI? The future of open AI?

Anders Arpteg 1:11:35

Yeah, I think there is uh it's very hard to see a future for open AI, I would say. So explain. I mean for one, you know, we've seen all the people leaving there that were the senior people and the the big brains, I would argue, like Ilias Escavera that left, and uh and then also that they're going more and more um commercial, call it that. Uh but if they were to go commercial, you know, who really have the opportunity to go commercial? Well, it's the companies with the user base and the distribution and the products. And who has the products? Well, not really OpenAI. They've been succeeding surprisingly well with the product of Chat GPT, so to speak. But that they would continue with that, I think, would be very hard. Um, and I think also with what we're hearing about the lack of trust we're seeing, and simply the benchmarking we've seen over the last year has been going down time and time again. It was a long time ago that OpenAI was leading the LM Arena um uh benchmarks. It's been Google and uh and uh Anthropic has been in the top there for quite some time. So uh I think it's it's already on the way down, and um, I think in one year we won't have some of them.

Goran Cvetanovski 1:12:49

You could make a bet about that. Um I will make a bet just to make a bet, but I think that he's out there. So I uh no, we can do a bet. I think that he's still around for one year. Yeah, yeah, okay. I think that he's I I think he's gonna be there until the end of it. Um I'm with you. I don't I uh if they don't do something uh critical right now, it's uh they already tried the the announce, yeah, the more the ads and stuff like that, but that didn't work. So yeah, let's see. It's interesting.

Åsa Schwarz 1:13:16

But but where are all the people going? That uh where where where are the brains going?

Goran Cvetanovski 1:13:22

They all open their own companies, they all got like a shitload of money.

Anders Arpteg 1:13:26

Yeah, they're billionaires all over. Yeah, okay, exactly. So yeah, Ilya is starting his own, and uh a lot of people went to anthropic. Yeah.

Goran Cvetanovski 1:13:35

Um what is your uh Jan Le Kun doing with all the money that he got? He got like 50 billion or something, no? Or no, maybe not, but he Yeah.

Anders Arpteg 1:13:43

He started his AMI company, so he also has his own company. Let me Google that then.

Goran Cvetanovski 1:13:47

It's been a while now. He needs to produce something, no?

Anders Arpteg 1:13:50

Yeah, he produced some research papers, but not really a product yet. Great research paper. With a big excitement in the voice. Okay, yeah, I mean, I think you know, this these court cases are really interesting. And uh uh if if they actually were to lose in OpenAI, then that would be really bad for them. And I think it's hard for them already, so to speak. And then we have the other big court case, which is um Anthropic versus um Department of War.

Åsa Schwarz 1:14:16

That's it, yes, that's interesting. How is it going?

Anders Arpteg 1:14:20

Yeah, apparently it's uh going a bit towards anthropic here, which is interesting. So uh just for for yeah a small small background here. Um Anthropic and Claude was used, of course, by Department of War both for Venezuela and for the Iran attack, and it's still used, but then uh Department of War was asking to use Claude for anything they wanted to, and then Anthropic said no, and you're not allowed to use it for mass surveillance and for autonomous weapons. And and then Trump went uh very angry and said, No, you're out, and they put a ban out a supply chain risk ban for Anthropic, and that means that not only the Department of War but any supply to them are not allowed to use Anthropic, and that was a really big thing, of course. And then uh Anthropic sued them, and um now that court case is ongoing.

Åsa Schwarz 1:15:13

I have a question. Uh the surveillance part is that only for uh American citizens, or is it for uh the rest of the world?

Anders Arpteg 1:15:25

Your guess is as good as mine, or or maybe not, but still uh who knows?

Åsa Schwarz 1:15:30

Because I I you know I I haven't read that probably, but my my my my thought is that you know you can't survey American citizens, but the rest of the world is perhaps yeah no. Uh but but don't take my word, read it somewhere instead. But I that was what I thought it was.

Anders Arpteg 1:15:48

Yeah, yeah. Um also super interesting case, of course. Anthropic, um you know, they actually recently, just in the last week or something, had uh went above open AI's um in terms of revenue. Yeah. So they that's another sign of OpenAI going down here. So even the company that's just like I think it's four years old or something, open AI is like 10 plus 12 years or something. And now Anthropic in such a short time has overtaken OpenAI. So it's it's certainly yeah, not going that well.

Goran Cvetanovski 1:16:19

Interesting. Google is coming as well. So Google is coming, sure. Just because they were first in the I mean they revolutionized the way how we consume AI. It was not something revolutionary more than that. They had a couple of models, but uh they have been Google has been and Tropic has been behind their back all the time, even the Chinese were uh behind their back. So it's all about the revenue model. Right. Yeah, you cannot find the revenue models. So their model is right now borrow as much money so you can build as much uh compute centers and everything else, right? But where are you gonna return the money? At some point of time, the investors are like, hey, this is like a black hole, I need to get out. How do you get out? Yeah, here is a chip, go home.

Anders Arpteg 1:17:00

And and that you know, open AI, they invested like they had like 20 billion in uh in turnover, and they invested one trillion, 1.4 trillion in infrastructure. How is that ever going to be? Microsoft is gonna buy it in the end.

Goran Cvetanovski 1:17:12

They would just like zip take it in and that's it, it's gonna be great.

Åsa Schwarz 1:17:16

I I I don't understand how their their custom their you know the payment model could uh do they have seven hundred thousand uh seven million uh so big user base.

Goran Cvetanovski 1:17:27

I mean making billions, but not enough billions.

Åsa Schwarz 1:17:31

Yeah, but it's can't if if you already have uh seven billion uh customers, it's not so hard, it's quite hard to grow.

Anders Arpteg 1:17:40

It's uh like no they got the consumer base and enthropy goes enterprise, you know, and enterprise is much more money, and yeah, so they need to go to enterprise, and uh that's that they've been struggling for that.

Goran Cvetanovski 1:17:50

Yeah, but they tried with uh Microsoft to get in, right? And I think that was probably good and bad.

Anders Arpteg 1:17:57

But I think co-pilot is you know, copilot started horribly, and uh but now it's actually starting to turn around, I think. So it could be a way I would not be able to say I have a Google users.

Åsa Schwarz 1:18:09

Yeah, I used uh copilot at my large company, yeah, and now I don't have it and I really miss it. I didn't think I would do it, but I oh god, I must get a license because it's too easy just to mail and write stuff and some but it took uh I mean the start for Microsoft with the first copilot, and it was really bad.

Anders Arpteg 1:18:27

But now it's starting to get yeah.

Åsa Schwarz 1:18:29

And we we we were quite early with the implementation, and it's it was not that good because people really didn't you know if you try it and it was good, you didn't try it again one long month later. So yeah.

Anders Arpteg 1:18:43

Yeah, oh really, really cool. Um well let's get back to to some discussions here, and um if we if we just continue uh a bit more into uh was it Aranja? Is that the name of Aranja? Yes. What are you really seeking to focus on going forward? I mean what are the type of competences you think will really be needed here uh for that company?

Åsa Schwarz 1:19:13

I think the biggest company the competence is the uh urge for change. So it's more a personality trait than uh but the the the competence it's quite a lot the same that that a traditional cyber security consulting company strive for. It could be you know in information security, risk management. Uh you need to know all the laws and all the standards and everything. But instead of having everything in your brain, you can use AI for sort of develop much uh faster processes for your customers. And it's also right now we also have a lot of competence within network.

Anders Arpteg 1:19:56

And then we are when you say network, what do you mean, really then?

Åsa Schwarz 1:19:59

I mean firewalls, routers, uh how you design a network with zero trust, for example, or how much zero trust you can get in the network. And there we also use AI for för example, uh one guy have built up a processes for adding new uh adding new things in the network with AI. And many many people that work with networks they don't program that much, but now they get much now they can do exactly what they want.

Anders Arpteg 1:20:36

Right.

Åsa Schwarz 1:20:37

But what they really discovered also in this process is that uh most models don't know so much about networks. They know they can coding. They are they are whatever model you have, they have a sort of train on coding, but they know don't know anything about networks.

Anders Arpteg 1:20:55

Oh, interesting.

Åsa Schwarz 1:20:56

So, yeah, that's quite interesting. So, what they are doing, they're uploading uh a lot of information about network, and then they can use it for in in their work.

Deepfakes And Election Manipulation Fears

Anders Arpteg 1:21:06

I mean uh super interesting. So, okay, so that's the type of skills you're you're trying to build up a bit as well. Yes. Yeah. Okay. What about you know in upcoming you know, we have some elections coming in uh coming up in the US and in Sweden now as well. How do you foresee AI and you know being potentially abused here?

Åsa Schwarz 1:21:27

Um unfortunately it had been abused for like 10 years already.

Anders Arpteg 1:21:32

Yeah. I guess so. But but the power of AI these days is so much bigger, right?

Åsa Schwarz 1:21:37

Yeah. It's uh you know it sounds very pessimistic, but I think we have the society we have today because of AI. If you see the American elections that sort of have uh have, I wouldn't say destroy the world, but all have somehow make it less positive, I would say AI. I think Brexit so that it's uh some kind of sort of because of AI. So if we already have destroying the world so much as we have today, what will happen with the next elections? And uh yeah, that was very pessimistic, but I also hope that we have some force for good that use AI also.

Anders Arpteg 1:22:21

Yeah, I guess it can be used in both sides. Uh I still remember the was it the Argentina election uh two years ago or something, and and they I think it was one of the first where deepfakes was really used at scale in some way. Yeah. So both sides were using generated images uh to to portray the other party in a good or bad way. And but then that's that in that time it was so obvious it was generated because it AI wasn't really that good at that time. But today it's so easy to do super realistic things, right?

Åsa Schwarz 1:22:56

You can't you can't trust actually any photos, any videos, any any you know, voice. It's it's not a way to you know it's you you you don't know it what it is with AI or not.

Anders Arpteg 1:23:09

Yeah. But how how can we fix that then? If it's so easy to get access to these kind of you know large-scale disinformation campaigns that can run and more or less get you know propaganda out in a certain way. What's what's the solution then? Can you please tell us what? Of course, of course.

Åsa Schwarz 1:23:36

I uh yeah, I I think you you you have to build and that this is not easy, but you have to build a verification culture. Right? Everything is verified digitally. And it will probably not happen. But you know the the the the solution is that you always have sources that you can verify. And if it's not you know, it's you you could there is no no other way to do it. But um and perhaps we will have a society where we're part of the society that you know very verify everything in other parties sort of out there in in a different uh um atmosphere or no.

Anders Arpteg 1:24:20

Yeah, sources that can verify, but if we try to just look at one extreme and say, and from a positive side, and say that AI will help us in some way here. Yeah. Let's say that you know someone makes a statement on on some media channel and you're not sure if it's correct or not, and you just ask uh ChatGPT, Gemini, Claude, or whatnot if it's true. Potentially that could be a source, a source that you trust in some way.

Åsa Schwarz 1:24:50

Or Yeah, it could be, but uh you know, usually when I ask whatever you you ask, I always ask about sources.

Anders Arpteg 1:25:01

So uh But then you have to know you know for those sources. How do you know if you can trust them in that terms?

Åsa Schwarz 1:25:09

Well, it you you don't really know. It depends on what we are talking about. Are we talking about political images, or are we talking about science or or what would you know?

Anders Arpteg 1:25:24

For one, we have hallucinations and this is kind of psychophantic behavior where an AI is trying to just please uh the user all the time. And that's one problem. Another problem could be that they are actually trained in a biased way, or have some kind of bias towards one party or the other. And they all are, usually. Yes, yes. And and then if you want to fix those problems, hallucination is simply, or and even psychophantic, I would say, is both just a lack of performance. I mean as models improve, it will reduce hallucination and psychopathic behavior. But but the bias problem won't go away that.

Åsa Schwarz 1:26:02

No, and I I think the bias problem can be you know it can be bigger in a couple of years because we do have different culture in different countries. And and and right now, for example, where all the big AI models are built in US, they have quite a different culture, especially now. So it's it's it's hard. I think you have to to decide what kind of sources do I trust. Do I trust this newspaper? Do I trust this.

Anders Arpteg 1:26:35

But I guess for young kids today, at least, I mean the standard source will be an AI chatbot in some way, right? That they talk to and ask for even advice in different ways. Or for how to vote in a political way as well. And if that becomes the norm, of the power of these models will be huge, of course, right? If they suddenly just shift 5% to the right or to the left for the Swedish election, that will have a huge impact.

Åsa Schwarz 1:27:05

Yeah, or or if one of the yeah, of course, and and that is perhaps because one of the the parties are better on feeding it with information or something like that. Um so this is really hard. Perhaps you could have a model that you can choose what kind of uh very you know sources you want. Or or you know, I uh I'm this kind of person I want. But then you will be have a society with the different this kind of persons that are living in different realities. So I I don't know really. I I sorry, I I can't uh solve this problem.

Anders Arpteg 1:27:44

Also you can't solve it. It's no way, of course. But but but if you take different, you know if I remember the first release of Gemini from Google, for example, and it was clearly very biased. In in like um, you know, if you asked it the face. Or infamous example was uh they asked who was the founding father fathers of US, and it was like a set of black women or something that they uh showed as an image, and and of course it was then uh highball biased to to do that kind of uh output. And and uh the question then is you know, how should how much should we prime models like this? I mean, if we prime nothing, that could also cause problems potentially. And if we prime it too much, then who should decide you know what we should prime it with? Yeah, you know, I don't answer to this is simply saying, uh, it should be maximably truth-seeking all the time, even if it's politically incorrect.

Åsa Schwarz 1:28:42

Yeah, but um uh his truth is obviously not my truth. So who's whose truth is is it? Uh yeah, I have another example. It's my my daughter played football. Yeah. But if you ask, I think any of the models that are the major models uh for the most uh the football players that have gained most gold in championship, you only get male answers. Even though the you know it's yeah. So the the bias are still there.

Anders Arpteg 1:29:23

But that can't be intentional. It's it's simply No, it's the data.

Åsa Schwarz 1:29:26

You you know, you there's a lot of football players and uh newspaper talking more about the male players and then yeah. But it's a yeah, it's a it's an old bias, also.

Anders Arpteg 1:29:39

Yeah, yeah, for sure. I'm just thinking, you know, how how should you and me fix it here now?

Åsa Schwarz 1:29:47

I think we will build our model. Let's build our own with the correct culture and yeah, tough, tough problem.

Anders Arpteg 1:29:57

But I'm just a bit scared, you know. For one, you can see it as a positive thing and also as a scary thing. The scary thing is that you know the power of these models will be so big because they will influence so many people. And if the next version of Gemini or Claude or Chap T slightly shift to one or the other side, that will influence a lot of people.

Åsa Schwarz 1:30:18

Yeah.

Anders Arpteg 1:30:18

And but on the positive side, it can be that if we have a lot of extremists to one direction or the other, and if someone says something online, which a lot of young kids are influenced by today, then perhaps they can use the these models which are available to them to just double check a bit, you know, rather easily. So if someone comes and claims you know, some extremist claim, then at least it's easy for them to double check at some point.

Åsa Schwarz 1:30:45

But didn't we have that check on uh Twitter and Facebook for a while and then it disappeared, I think?

Anders Arpteg 1:30:52

What do you mean? I don't know.

Åsa Schwarz 1:30:53

Uh I I uh I I I think what wasn't it on on Twitter that that you know if it was false, you know you could Groc is doing that. Groc is doing that, okay, okay.

Anders Arpteg 1:31:04

Yeah.

Åsa Schwarz 1:31:05

But this was for a while ago, and now you know Twitter is destroyed and it's X and yeah. Um but I think we had it in some social media for a while and it disappeared.

Anders Arpteg 1:31:16

But you know, the the idea is good, but it's it's if if the owners of the models don't agree and and think it's you know like yes, say it as it is, even though it's not really the truth, but what the model is sort of I mean it's uh I'm just saying for for one side on the more extreme case where when you have extremists that is really you know young kids is being influenced in a really bad way and radicalized and going and doing bad things and you know what seven-year-olds that go and throw hand grenades in Sweden and whatnot. I mean, of course these kind of things, but where at least perhaps models can provide some source of knowledge that hasn't been available in the past.

Åsa Schwarz 1:31:59

Yes, and I I think you know it's it's probably uh you know the idea is good if it's if it's sort of implemented. I know I have a 14-year-old uh son, and you know, we I have been talking a while with him and I realized that he got a lot of from YouTube, from Snapchat, whatever, you know, these uh incel uh videos with where you know they're really weird uh culture when it's about women and stuff, and that is our 14-year-old. Now he he you know he understands his bullshit, but you if you have to talk to them for a while, and if they don't have a mother or father that have some you know, have these discussions.

Anders Arpteg 1:32:46

Then you have the the algorithm that you know the that uh prioritized and recommends through the feed, which which YouTube does and Facebook does and everything. And that's one way you get influenced, and usually that polarizes a lot.

Goran Cvetanovski 1:32:59

So it does.

Anders Arpteg 1:33:00

So once you use you like something, then it just goes super deep there and then you get radicalized very quickly. The hope potentially I'm trying to find an optimistic.

Åsa Schwarz 1:33:12

Yeah, I'm sorry that I'm sort of a positive side.

Anders Arpteg 1:33:15

If you compare the social media kind of recommender system problems to a general foundational model, chatbot kind of style, then at least perhaps it is not trying to polarize as much. Because it's trying to be built in since they had all these kind of psychopathic problems, it's it's really arguing back sometimes now.

Åsa Schwarz 1:33:37

Yeah. Yeah, but that's that that that is good. Right. And and perhaps it, you know, it's it it would be the the best solutions. And it all could also be if you have Snapchat, for example, they have this small uh little AI friend. Have you seen it?

Anders Arpteg 1:33:53

No, I don't use Snapchat.

Åsa Schwarz 1:33:54

No, no, I I I use it for my teenagers, the only way for me to communicate with them. So I have I have three friends on Snapchat, it's my husband and my two kids. And but uh there are five friends because I have the small AI friend also. Oh, yeah, nice. Uh and the small AI friend I can communicate it with with, but it has the same I think it's built on uh um uh chat. It's like chat GP Chat GTP, but it's um sort of a that's a friend. But it has the same bias. So but if that little friend would be the the you know the the trusted aunt instead, yeah you know it's an that act like that instead, uh instead of the small friend that just uh say all this stupid stuff.

Goran Cvetanovski 1:34:44

I think that what social media did is actually they disrupted the the educational system that we had through through TVs and radios. Because at some point of time I remember when I was growing up, they were basically psychologists and sociologists and uh and uh very educated people who were creating and curating what children should be looking at. At what specific time and what specific age. And I think the social media just basically, when it came, it just uh opened up the Pandora's book where nobody is actually uh guardrailing any content for children. I have like uh also um 19-year-old, 17-year-old, and an 11-year-old. Um I think that it would be interesting if the social media uh companies they use these AI agents or the friends that are there to um to basically help children to uh learn and to see what they need to see in their own time. Because if you're 11-year-old, maybe you should not see stuff that are it's for 15 or 18 year olds and etc. That would be a fantastic use of this technology, but of course, it doesn't benefit the social media, so they will never do it. Uh so instead of that, they would just use the AI friends so you can actually consume more and more of the social media bullshit that they are feeding.

Åsa Schwarz 1:36:11

So yeah, the the business model don't uh don't support the aunt, all friends and we are yeah, so it's it's interesting because we are we're talking about innovation, uh, but uh some of the companies are actually doing it for uh yeah for bad things.

Goran Cvetanovski 1:36:30

It's the same like with gaming and i gaming. So you have, for example, um the the Swedish gaming uh company, I think it's called uh it's called like Svenska Spell. So they are creating uh algorithms that are monitoring if people are overspending it, so like, yo, cool down, come back tomorrow. And in the same time, the other Mr. Green or whatever it is, uh I'm just naming, I don't know if this is true or not. Uh they're basically doing algorithms whenever you are uh in the heat, they will give you 500 crowns extra so you can stay in the game and play more, and that is this uh uh polarity that we have with innovation. So yeah, of course, the most innovative people are the ones the one that uh earn the most.

Åsa Schwarz 1:37:16

Yeah. But now now it's starting to get quite depressing.

Goran Cvetanovski 1:37:23

This is our traditional road, only have like 20 minutes to get up.

Anders Arpteg 1:37:28

I'm trying to get more positive. I think you know, of course, we we saw a lot of radicalisation and polarization on social media, but perhaps AI then can help us a bit with that.

Åsa Schwarz 1:37:45

Who is sort of finding its uh its goals? So we have to I think perhaps it's so that we have to have other kind of people that are the main investors in AI.

Anders Arpteg 1:38:00

Yes, because we've been before in this podcast, we spoke a lot about you know, should we have a European or a Swedish language model, and should we have, you know, build these kind of foundational models ourselves. And from a technical point of view, or from a pure economical point of view, I don't think it makes any sense at all. And we can never catch up to the big American and the US sorry, Chinese providers. But from a more biased point of view, it could make sense, right? But then perhaps we can just fine-tune the biggest one. But but in any case, it will be potentially so much more power in these models because they will influence so many people. So we need to have some control of what the biases are are, right?

Åsa Schwarz 1:38:45

And I I can also add that I think in the investment community, only 1% of the money goes to women. So and and many of you know the female founders they are are doing stuff in psychology and in children and in health. And they actually usually have more successful companies that don't get bankrupt in the same way static. So perhaps we should invest in more in female founders also.

Anders Arpteg 1:39:16

It could be a reason or a way to to if if used properly and if the biases are not too extreme in these models, then you know it could be a way to help us become more objective and remove some of the you know biases for for men and wheat and female and different parties and genders and and whatnot.

Åsa Schwarz 1:39:39

Yeah. Yeah, we we could agree on that. So if it used it right, so if it use it in the right way, of course, and and and if we train it on data and whatever, you know, it's of course. I I don't you know, I think AI is so much fun, and I am happy that we we are have uh have so much to to to do now with AI.

Anders Arpteg 1:40:02

Yeah. But I think anyone agrees, and I'm sure all of us here agrees, that AI will have an even bigger impact in coming years and decades on human life.

Åsa Schwarz 1:40:12

I think in the next six months. Yes.

Purpose In A World Run By AI

Anders Arpteg 1:40:16

Yes, agreed. And and if we just think about, you know, I guess young kids especially using AI much more than elderly people are already, and of course, as the time goes on, more and more of the population will be let's call it AI driven or AI influenced more and more. And how you know what and especially with your background as a novelist as well. If we think about this, if you were to write like a novel about you know where Sweden or Europe or different parts of the world could be when AI really influences, let's say it influences a lot in five years or ten years, where you know elections they basically all are super united in saying we should vote for that, and then suddenly suddenly a new model gets released, and suddenly the the election goes to the other side in one month.

Åsa Schwarz 1:41:20

Could turn out uh I I haven't thought about the the the voting process. I have more thought I've been thinking about um you know how how how kids grow up and and and and how how you know suddenly we have a uh a situation where we have new I I I read I read somewhere that if you know the the the the difference being between a successful what now a successful person is, it you know, it could depend. But I think in this context it meant someone that you know did something that is important in the work. Uh that doesn't have to be a successful person, but in this context. And what they did was uh that they sort of realized in this study that uh kids that didn't do any tours or and did anything at home, and you know, didn't help with dishes and walk the dog and stuff like that, you know, they didn't do that much as growing up either. It was, you know, it was uh uh it it was a relationship. If you have a lot of things you did when you're uh and you have you have consequences if it didn't happen, then they you this different stuff when you're older. But now we have a society where you know the the robot dog it walk itself and uh and the the the grass is you know the we it will be just fine every day and someone does the robot dog dog do the dishes and everything is you know it's like um curling. The whole the whole life is so so I'm thinking what what kind of humans you become then. And yeah, I think there's two different sides. It could be that okay, I don't have to do this damn uh dishes, yeah. Yeah. I can do all this really fun stuff with AI, and I can you know play with my friends and I can sort of develop as a human being. Uh and then we will have completely, you know, completely different persons. And and and you can um perhaps they're all AI adapted, I don't really know what's happening then. But you can also have the other way that we we we we are not really we are just very lazy. I think you thought that you talked about it a little bit before that you know you you just you don't have any drive anymore, nothing happening. So I think that is a little bit uh and I think also if you but if you if you take it another step, if you have um uh general AI, uh and and it's sort of I think what what we don't think about then is you know we I think I think you sent me the question before is this is a you know terminator uh future or is it the you know the flower and uh creative and so I I think that we will not be that important. We are not the center of the history anymore.

Anders Arpteg 1:44:23

Yeah, I I we we could go that route. And and but let's end with that question. Yeah, sorry, sorry, I destroyed your no no, but it's fine, it's fine.

Åsa Schwarz 1:44:31

We can get elaborate later. Yeah.

Anders Arpteg 1:44:33

But uh just if we just assume, yeah, which is rather probable that you know AI adoption will increase in population a lot. Yes, yes, yes. And it's really high already. Yeah, I I think for young kids uh it would be fun to see some number, how many is using some chatbot every day. But I think it's a good idea. Yeah, my kids do. Yeah, and and that will just increase and it will be used uh for more and more different tasks. Yeah. And uh I will probably use it this weekend to see you know who should win the Eurovision as well or something. Who knows? But um I I think you can see it as a trying to say, I'm trying to find a positive side here. But I think you know it could be really positive in the sense that we become a bit more objective if the models work. If we just assume the model is objective and not too biased, then that could be a really positive thing, actually.

Åsa Schwarz 1:45:31

Yeah, it of course. Uh and and it's uh you know, if you have the all the information and you have also the possibility to sort of yudge it in some way, uh, you know, it's of course it's much much more positive. And I also think that uh you also get uh this kind of what can also happen is you know, I'm I'm lousy on on drawing, I'm I'm not really good at that. And I'm I'm I'm I'm you know, I can't sing. Yeah, okay, you can sing.

Anders Arpteg 1:46:01

Yes.

Åsa Schwarz 1:46:02

Yeah. And uh and some people can't write and you're dyslexic, everything, and and suddenly you get you you can do so much more.

Anders Arpteg 1:46:11

And I think that it well who could help if you if you are you sort of name of people to do stuff they couldn't do otherwise. Yes.

Åsa Schwarz 1:46:18

Yes. And some some people that today are angry because they are unique and now they are not unique anymore. But the the you know, the everyday people like me that definitely can't sing. You know, I can't I can make a song anyway. So it's also positive.

Anders Arpteg 1:46:33

It is. It is. I mean I have to resort to karaoke, you know. That's the only way time I'm allowed to sing. Uh hardly even then.

Åsa Schwarz 1:46:41

No, I uh I try to avoid, even though I'm you know I I'm not that sort of you know, I'm I'm not that shy, but you know, it's more for for the other ones. Yes, yes.

Anders Arpteg 1:46:53

No, but okay, so on the positive side, it could really be good. Just assuming the models are objective, it could actually make us people more objective. That could be a good thing. Yeah. Then on the flip side, of course, it can be very negative if suddenly every human does what the AI says it should do.

Åsa Schwarz 1:47:08

Yeah. So then But I think I think we also we have uh the language barrier will you disappear, or it's already have you know, you have an iPod and you can talk to someone. And so I think it can also be a different understanding for cultures.

Anders Arpteg 1:47:23

Oh, yeah.

Åsa Schwarz 1:47:24

Because if you if you can very easily talk with someone from China or from Japan or wherever, without have the language barrier, I think you can have uh you know another different understanding of why they are doing like this, what is happening.

AGI Utopia Versus Real-World Disasters

Anders Arpteg 1:47:40

Yeah, good point. Yeah. I mean, if we do it right, it has a lot of potential for sure. Yes, of course. And and going then to the extreme, and um then thinking uh you spoke about general intelligence, and um, if we have AGI, some type of AI, if we use the definition from Sam Altman, he basically says that um you will have AGI when AI is as good as an average level human co-worker.

Åsa Schwarz 1:48:06

Okay, okay. Okay. So it's not super intelligence, it's more about sort of a friend.

Anders Arpteg 1:48:12

So you you literally could take like an average performing co-worker, human co-worker, and replace him and put an AI there. And that's certainly not the case today today. I mean, it's it's so many things that uh AI cannot do. Some stuff AI can do much better than humans, but other stuff humans are still significantly better than AI, I would argue. But at some point, it could be that we have ATI in that definition. And then what would that mean? I mean, then we have the two extremes here. So one extreme is the Terminator and the matrix of the world, and machines trying to kill us all, and I know with all the wars happening now, who knows? That could not be, may not be that far away. And then the other extreme is um then a very positive one, the utopia version, where AI have sold um, you know, find a cure for cancer and fix the climate crisis, and perhaps it's uh reduced, as Elon call it, um, the price of goods and services to zero. So more or less we have a world of abundance where you don't need to work and you perhaps not need to work 40 hours per week or even 20, perhaps you know, 10 or nothing if you don't want to, and you still have you know roof on top of your head and food and everything you need. Um what would happen then? What do you think?

Åsa Schwarz 1:49:27

You know, I think you're I think for we we can have a I think first of all, I think it's perhaps it's not that important if you are a human or if you are you are an AI. So you will have I think we'll have much more interactions with AI than other humans. Uh and I don't really do know what that does to persons, it depends on the AI. Uh in I know uh that uh uh some some some uh AI that have been moving used as a psychologist, they are as successful as as humans. So perhaps you're not never lonely.

Anders Arpteg 1:50:03

That's a good point.

Åsa Schwarz 1:50:06

You have all all these AR friends, and you perhaps you don't care if it's an AI or not because you are at home and talking to different people. I have a very god friend that lives in London. I talk to her, we take a lot of walks. I talk walk in the älts skoggan and she walks in in London and we have a lot of talks. But I wouldn't know if she is an AI or not because I see her each five year or something. Right. So if you have a lot of friends that and some hard AI, some are not, then you're not långan anymore.

Anders Arpteg 1:50:42

That's very lonely as well.

Åsa Schwarz 1:50:45

Yeah, there's lots of they're lonely, and and and so I think that's a very positive way. On the other side, if you don't really have a purpose anymore because you don't work, and and then there are AI that probably are soon more intelligent than you are. You know, the human is not that important anymore. So you don't have a purpose as you don't have work, and you're not important in history anymore because the AI are more successful. I don't really know what that it's probably that you have different personality traits. Some will like it, some will not.

Anders Arpteg 1:51:19

I mean it's interesting to just think about you know, if an AI is like a thousand times smarter, not on the level of a like average level human, but actually, you know, more intelligent and knowledgeable than all humans combined times a thousand. I mean it could be the case. I mean, we have then some kind of system that we know whatever we ask it, it will know so much more that we won't even have a chance to even understand what it's saying. It will be like me trying to talk to an ant about you know, little ant here, the we have you're living on earth and the earth is round. And the ant is like, what are you saying? Who cares? And and for an AI trying to speak to a stupid human, yeah, uh, you know, who who is then equally you know far below as an ant but will it try to speak to the human because the human is not uh you know.

Åsa Schwarz 1:52:13

No, it's it's it's not a speech that sort of perhaps it's destroyed some of the world, and then you have to sort of see the make it stop that.

Anders Arpteg 1:52:21

But could it be necessary? I mean, of course, a lot of people think this could be really bad, and saying that we have an overlord for some kind of AI that is you know so much more smarter than any human ever could. I mean, we are not trying to kill the ants, we're not trying to kill the animals, we're not trying to kill our pets, even you know, that we have at home. We we actually care for the pets very much.

Åsa Schwarz 1:52:42

But you know, it's it it it it doesn't have to kill kill kill us. Uh but if you think about how we work, you know, many many people don't want really to acknowledge that when you die, you die. That's why we have religion. This is my take on religion anyway. But so but now we don't even have a purpose to live, you know. So perhaps uh the AI will be our religion, and it you know it's it's probably uh written in some books, but uh you know it could it could be uh that you know we don't have a purpose, we have to find a purpose and then perhaps the AI is is our purpose.

Anders Arpteg 1:53:22

Yeah. So what what you if you were to say like a probability in like five, ten years, uh AI is gonna kill us or is it going to be a savior?

Åsa Schwarz 1:53:41

So I I think we will have a lot of huge disasters, but I don't think they would kill us as speeches. But I think we some of us will probably have encounters with AI that is not so turmoil in one way or form.

Anders Arpteg 1:53:57

Yes, that's for sure.

Åsa Schwarz 1:53:58

Yes, and and but I hope that the end result is that you know we we have a fantastic nature and everything is fine, and we can sit at the beach and drink. Yeah.

Goran Cvetanovski 1:54:11

As long as uh people are good, uh AI is good. Uh that is what it is.

Åsa Schwarz 1:54:15

Uh but you know, then then it's now we are in the deep.

Goran Cvetanovski 1:54:18

I can think we still we still like I we there are some very bad people, but there is always for one bad person, there is probably uh ten that are good. So I think that the good will prevail. If you look at uh Walt Disney movies, they will end up very good. So I think we should uh look at that.

Åsa Schwarz 1:54:35

So I think you have the answer now on that. We have to train the models on uh Disney.

Goran Cvetanovski 1:54:44

But also these technologies and uh they're like Elon Musk and everybody, they're trying to relive their fantasies uh that they read when they were kids. Yeah, so we are repeating the the stories.

Anders Arpteg 1:54:56

But I think you're on to a very good answer here. We should just fine-tune all the frontier models on Disney movies, yeah. We're saved.

Åsa Schwarz 1:55:02

Yes, you have to because then the the good always uh well.

Closing Thoughts And Goodbye

Anders Arpteg 1:55:06

And romantic comedy. Yes, yes. But then then there's no problem for humanity. Well, Sasraj, it's been a true pleasure to have you here. I hope you can stay on for some after after work discussions as well. That would be nice. Um thank you so much for coming to AI Afterwork Podcast.

Åsa Schwarz 1:55:21

Thank you.