4#20 - Sune Selsbæk-Reitz - Promptism and the Dangerous Illusion of AI Truth (Eng)

«We need source criticism more then ever now.»

In the season finale of MetaDAMA, we dive deep into the intersection of philosophy, history, and artificial intelligence with guest Sune Selsbæk-Reitz, tech philosopher with a background in both history and philosophy.

Sune introduces the provocative concept of “Promptism”, which is our era’s version of positivism, where we believe that truth can be extracted from language models simply by phrasing the question correctly. But just as historians have learned through centuries of source criticism, we must ask the critical questions: Who trained this model? On what data? With what biases?

Here are Winfrieds key takeaways:

• Are numbers and data points neutral? Or can they be used to convey a message, or even a certain philosophical view point?
• Philosophy is important in data. Here is an example:
   
  • Data Governance according to Immanuel Kant - best possible governance
  • Data Governance according to Utilitarism - focus on business value
• Lessons from history studies: question the authenticity of your sources. Who wrote it? For what purpose? Why are you reading it? - same lessons apply to data.
• That is why we need principles and values in AI ethics.
• Over-reliance on the objectivity of math - is math binary? Right or wrong? - this has been introduced into algorithmic thinking and AI.
• This is the reason why «algorithmic authority» is an issue - because the algorithm says «right» doesn’t mean it is right.
• Our mindset is constantly evolving. That’s why we cannot predict tomorrow’s bias. We need to ensure that our systems evolve with us.

What is the real purpose of AI systems? Are the core values only efficiency, automation, or is it human dignity or autonomy?

Promptism:

• A new way of «positivism: Just because its written down its true»
• Promptism is my term for a subtle but growing mindset around the globe, that you can extract truths from a language model, just by wording your prompt well.»
• LLMs are very fluent and flattering - they say what people want to hear.
• «That’s what Large Language models are: You are not getting the truth. You are just getting the most common answer.»
• Objectivity is a myth. It is always subjective, so we need to read not only the text but also peoples intentions with the text.
• Responsibility for understanding at the limitations and needed criticism of LLM output is shared.
• Producers have a responsibility to ensure that you can know, when models are hallucination, guardrails in models to ensure that output is not looked at as the truth.
• Readers are responsible to learn how to read and understand machines.
• Consumers need rot push for transparency.
• Without accountability, trust is eroding.
• Politeness is a way for machines to ensure that users keep using them.
• Shouldn’t rather an LLM as a «conversation partner» challenge you? Disagreement is part of learning.
• «Agreeableness is addictive.»
• People are starting to get influenced by how LLMs are writing. It changes written conversation.
• Is language narrowed down to a certain path defined through AI? Is language becoming controllable?
• LLMs affect our lives in the way we read, write, talk, even think.
• There is a worldview baked into the system.
• Literacy means also critical thinking.

Transcript

Introduction

Speaker 2 0:20

Dette er Metadema, en holistisk synning, som en profesjon i Nordisk. Vær så god med de kompetensene vi har, og det er derfor jeg inviterer nordiske eksperter i dat og informasjonsmanagning til å snakke.

Meet Sune

Speaker 1 0:35

Velkommen til den siste episodeen av MetaDema, sesong 4. Og som alltid, vil jeg slutte en sesong med å løfte vår syn. Teksting av Nicolai Winther earlier, where we talked about, for example, the geopolitical situation that we find ourselves in, where we talked about how data is taking over our lives through public services, and today we're going to lift our view into a direction that I'm very passionate about. I talked about this on the podcast before several times that my background is in history, political science and law, and every time I say that people are like for mange ganger at min bakgrunn er i historie, politisk forskning og lov, og hver gang jeg sier det er folk like «Åh, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, ha, at least since the 19th century.

Speaker 1 1:43

There's something around textual criticism that you learn. There's something about how you interpret your sources. In German, we call it Quellenkritik. In all the Scandinavian language I think it's something like Kjeldekritik, and you really don't know the context, the intention a certain text is written about, and this is very much the task of a historian, to uncover that and put it into the right context and try to bring at least some sense of objectivity into these texts and sources, and that textual criticism has become even more important now, when we live in a world where there's a lot of text generated through AI, where we also, more often than not, har blitt enda mer viktig nå når vi lever i en verden hvor det er mye tekst generert gjennom AI, hvor vi også mer ofte ikke lurer på tilgang til ressurser, tilgang til kontekst, og bare presenterer det med en certain utgangspunkt.

Speaker 1 2:37

Lykkelig for meg, har jeg en ekspert i området med meg i dag, sunne, og han har veldig mye dypt i denne temaet enn jeg har expert in the field with me today, sune, and he has very much dived deeper into that topic than I have, and it's very interesting to talk about exactly that. So, sune, welcome.

Speaker 3 2:55

Thank you, Vitric.

Speaker 1 2:58

We do it the natural way. Let's hear your introduction first. Who are you? What do you do? What are your hobbies?

Speaker 3 3:05

Oh yeah, å, ja, å ja. Jeg arbeider som data og AI-strategist i DEMAT i København. Det er en av de fem store høring-8-manufaktørene rundt om i verden. Jeg sitter i vår produktmanusjon portføljedivisjon hvor jeg hjel division, where I help my fellow, you can say, product managers fill out their area strategies for their app, connectivity devices and all that. but I focus on data, ai and cloud infrastructure. So that's why I day-to-day work, you can say on a more personal level, i have always have, you can say, an urge to write something.

Philosophy and data

Speaker 3 3:44

I have published two poem collections early on back in my student days at the university, going out to, you know, nightclubs, reading out poetry, reading, all that, and somehow I never, you can say, loosened that urge to write. So now I'm writing on LinkedIn mostly and also on min substack, og det er mest om data og AI-etikken. Jeg tar en dyp passion i Immanuel Kant. Han har en dyrt rasjonal filosofi hvor han er mer inn på at vi bør gjøre det rette i det rette ordet fordi det er rett, ikke fordi av utgangspunkt. we should do the right thing in the right order because it's right, not because of the outcome, but because the thing in itself is right. So that's more or less what I do all the time and I'm boring my wife about it and you know. so all good, yes.

Speaker 1 4:37

Well, my wife loves these conversations. No, she doesn't. I think I'm the same pickle as you are. Let's come back to Kant in a second, because there's something interesting there to. Jeg tror jeg er den samme pikkeren som du er. La oss komme tilbake til Kant i et sted, for det er noe interessant der til å utforske. Men før det, hva er din interesse for datafjellet AI? hva kommer det fra? Du kan si at jeg, som de fleste som ikke studerer data eller ingeniøring, støtter på data av skade.

Speaker 3 4:57

Jeg ble høret i en dansk pensjonsfond not stuttering data or engineering, stumble upon data by accident. I was hired in a Danish pension fund and at that moment they, you can say, had a scale, agile transformation ongoing and I became, you can say, a business person the product owner for a data development team. In that regard And that was the first time I hadnden på du kan si data og data-infrastruktur Så av skjedd, klikket det virkelig inn i min hele perspektiv av livet, hvor du kan si at hver datapunkt er en narrativ, at du kan si at alt vi gjør er rundt mennesker, og jeg møtte en klasje med alle de ingenjørene around people And I met a clash with all the engineers that didn't see the same way as I did. So this was really a challenge for me to go into it, so you could say I'm just pushing on and going from that.

Speaker 1 5:56

I mean I've used this quote way too much on the podcast as well, but I very much like it. Ronald Ross said that data is a message to the people in the future And if you look at it from that perspective that your target group, who your data should resonate with, is in the future, you need to prepare that data to actually be understood in the right context. This is very much an issue that we are struggling with, especially in data ethics, but also AI ethics right, but you can also say that AI ethics is just a magnifier class on data ethics.

Speaker 3 6:32

So it's not that different, it's just a more powerful tool. You can say you can go from HP much faster and that's more or less it. You could always do the things we could with AI more or less beforehand. So you know it's really the backbone of AI ethics is data ethics and governance for that matter.

Speaker 1 6:52

Well, i think you mentioned Kant quite early in your discussion, which is interesting. It doesn't happen that often that you get a conversation where Kant is introduced that early. But I wanted to talk with you about why do you think that philosophy has a place in the data world, and especially now in times of AI and Gen AI? And when talking about Kant, i've been thinking a lot about data governance and maybe just to give you some context here and what role data governance and maybe just to give you some context here and what role data governance is playing in an organization, and I feel that as a starting point, it is very much based on what you explained around Kantian principles.

Speaker 1 7:39

That has turned into a certain issue where data experts were very much aiming to create 100% data quality, best possible quality of data, instead of trying to focus that on what is actually the business outcome. So the last years, and especially from, i would argue, from across the pond, from the US there has been a lot of business value focus. So how do lot of business value focus? So how do we create business value through data, which is very much turned into more of a utilitarism when you think about data governance and what quality you are providing. So, with that as a background, where do you see the place for philosophy in data?

Promptism and textual criticism

Speaker 3 8:22

Oh yeah, you can say that we are grown up with the idea that numbers and data points are neutral for filosofi og data. Å, ja, du kan si at vi er vokste opp med ideen at nummer og datapunkter er nøytrige. Data snakker for seg selv. Du kan si, men for meg er det faktisk at alle data vi samler samles av mennesker, de er interperert av mennesker og brukt av mennesker, alle med hverandre å ha intention, konstruksjoner og forst by people and used by people, all with each having intention, constraint and assumptions. You know biases and all that. And yet in these, you can say, ai systems, we often treat data as, you can say, an oracle or a gospel. We just stop questioning the origin of the data underneath. So you can say, data may be structured, but it doesn't make it neutral. So to me it has fingerprints and we should really look at it from. You can say both a philosophical sense, but also a historian sense, that everything needs to be. You can see under suspicion. We need to question it more. And you can say, from my early days as a historian student, we were taught from the very Vi må kristne det mer.

Speaker 3 9:25

Og du kan si at, fra mine første dager som historisk student ble vi fortalt fra begynnelsen at vi burde spørre disse spørsmålene om tekst hvem som skrev det, for hvilken behov og hvem er du som den som leser det? Så det er ikke bare teksten eller ressursen selv, det er også deg som leser. Hvilken valg putter du inn i det? Så fra mitt perspektiv, er dette virkelig hvorfor vi bør lære, du kan si, ikke bare den inntegnerlige delen av AI, men også den valgte delen, den principale delen av AI og etikk.

Speaker 1 9:57

Veldig godt, ja, takk, og jeg tror du sa noe som stakk med. I think you said something that stuck with me, and that is that over-reliance in the truth of fact and that facts can be objective, and this is very much based on how we interpret math for and have been done for for centuries. Either it's one or it's not. There is just one right answer and one wrong answer, and that way of thinking that over-reliance on the objectivity of math, that very much has been introduced into algorithms, right, something that is called algorithmic authority, that, because the algorithm says it is that it has to be, that.

Speaker 3 10:40

Yeah, and you can say now the challenge is really just at scale. We have AI systems that just not use data. De anbefaler det, de gjør det i forhold til å oppføre oransje av datapunktet selv og også å automatisere kontraktskjøringene som kommer ut av det upon it as such. And if we are building system on, you could say, unexamined data, we risk qualifying past biases as future defaults. So you can say, if we have trained a model only on white male in their 30s, as I am, we will always get in results that suits me, but not all people, all the majorities, And that is a problem that I think is 30 som jeg er, vil vi alltid få resultater som er tilgjengelig til meg, men ikke alle mennesker, alle de største.

Speaker 1 11:26

Og det er et problem som jeg synes er bliende mer og mer present at når vi snakker om AI-ethikk, snakker vi veldig mye om bias som har blitt innført og hvordan vi reduserer bias og hvordan vi sørgerrer at modeller ikke drifter. Men jeg tror det er noe som er underlagt her, noe som underlagt som jeg tror vi kan veldig mye forandre fra en historisk eller filosofisk perspektiv.

Speaker 3 11:54

Ja, du kan også si at vi ikke kan utvikle bias 100%. På en måte, det vil alltid være bias. Og du kan si at det vi har funnet bias today, but it's not what we found biased in the 90s. So our mindset is always evolving and shifting, but we should somehow label our models, our data points, with those biases so we are aware of what we are basing our intelligence upon. So that's really my you can say core feature in this that if we are building something and our data set is limited, we should just be transparent about this, limitless, simple as that featuren er at. Hvis vi bygger noe og vår datasite er løsnet, så bør vi bare være transparent om denne løsningen.

Speaker 1 12:32

Du snakket om transparant, og det er noen terminer som går rundt når vi snakker om AI-etikk og datatetikk, hvor transparant er en av dem? Vi snakker om rettighet. vi snakker om tilgjengelighet. Tror du at? Do you think addressing the issues that we have through these kind of frameworks, like the FAIR framework, are the right way of doing it?

Speaker 3 12:53

Yeah, it's at least a start doing it. But we should, you can say, build it upon a more unified framework. That's asking different questions. So you can say is this trustworthy, is this explainable? You can say all the transparency, but also is it equitable, is it fair for all? Så du kan si er dette trusselig, er dette forklarende? Du kan si all transparensen, men også er det kvalifikt, er det rett for alle? Du kan også si at vi alltid bør øve på å ha noen slags menneskeintervention, menneske i løpet, og så bør vi alltid se disse produktene vi bygger som en kontinuerlig utviklingsfase. Så jeg kaller dette moral-kontinu, as a continuous development phase. So I call this moral continuities, sorry. So moral continuousness, where we are not just looking at ethics from the starting point, but we are looking at through the whole development phase and also after release. We should have regular look into our systems and seeing if they have shifted somehow.

Speaker 1 13:47

I like that you introduced the term human in the loop to the conversation. I had a previous recording with Guri Hasselbalg from Denmark and she talked about her last book and she talked about that human in the loop is important for practical matters, but there's something more here that we need to focus on, and that is human in the center er viktig for praktiske ting, men det er noe mer her som vi må fokusere på og det er mennesker i midten av det vi anbefaler AI for i vår samfunn, og jeg tror det var en kraftig skift.

Speaker 3 14:19

Ja, du kan også si hvorfor bygger vi disse AI-systemene? Hva er de korte valgene vi virkelig bygger på? Er det bare effektivitet, optimisering, eller er det som Kant, som jeg har nevnt et par ganger? er det menneskes dignitet, er det autonomi vi bygger på, eller hva er virkelig purpos her For meg? vi bør selvfølgelig ønske du kan si optimisering, det er rett, vi kan redde noe penger og vi kan spille det utenfor, men det bør ikke være på grunn av mennesker.

Speaker 1 14:51

Jeg vil snakke med deg om en term du har introdusert i ditt nye skrivning som heter promptism. Kan du forklare, hva promptism er?

Speaker 3 15:01

Ja, og da kommer vi til min kjære kjærlighet til skrivning, my dear love of writing. So this was a late night sitting in this office space just struggling with an idea. I had a conversation with one of my old friends from history classes about how to read and how to interpret AI content. We both use AI, you can say, as a kind of like you did med search engines før, bare for å få kjøtt informasjon. Og da kom det på meg at promptisme er en ny måte å si positivism.

Speaker 3 15:38

Det er en gammel historisk term hvor vi kan si at bare fordi det er skrevet ned, er det sant. Og for meg, kom igjen Som historiker, vet du at bare fordi det er skrevet ned er sant? Og for meg, kom igjen som historiker, vet du at bare fordi det er skrevet ned, er det ikke sannheten? Det er skrevet ned av noen, for hvem, i en purpose. Så du bør alltid putte deg inn i det. Du kan si at prontisme er min term for en sattel men voksen mindset rundt denne jorden, at du a subtle but growing mindset around the globe that you can extract truth from a language model just by wording your prompt.

Speaker 1 16:12

Well, And now we're actually back to something that we mentioned right in the introduction about textual criticism and source criticism. What do you think happened to those techniques?

Speaker 3 16:22

They are more or less being forgotten at the moment and it's down to that. the large language model is so fluent and they are flattering. They are saying what we want to hear And you know if they are saying something you want to hear, you're not questioning it. You are being presented it with a so good UI and UX experience so you are not questioning anymore what it's actually saying. But you know, the language model doesn't know things. They are predicting things and they are generating responses based on patterns and not facts, at the moment at least. So when it just sounds right, then we just stop asking what is actually right anymore.

Speaker 1 17:06

And it even goes further than that. Right, if I pose a question to JetGPT or Tropic Cloud, i get a response, and then I can and that's what I normally do I ask for sources. Right, where did you get that information from? But now I realize that even these sources are sometimes made up.

AI politeness

Speaker 3 17:26

Yeah, you can say it depends on the model and which search function, on the model and which search function within the model you're using. You can say deep research is better, but if you're using the base models, they are just aimed for efficiency. So they are answering quite fast and they also just made up sources along the way just to make it feel right And then you're not criticizing it anymore because you got a source. you know it was saying by Ron Buffett. so, okay, fans, well, you could say that it's like, if you're asking, you could say a thousand voices in a crowd to shout out at once and then, believing in the most common phrase that was said, back to you That was a last language model. You're not actually getting the truth, you're just getting the most common answer.

Speaker 1 18:13

That was a fantastic way of explaining it, but what do you think that poses on us as users? What responsibility lies with us when we interpret it and use these results?

Speaker 3 18:25

You can say that we need source criticism more than ever now. As I said before, as a history student, the first lesson I interpretere og bruke disse resultatene. Vi kan si at vi trenger sorgskritisering mer enn allerede nå. Som jeg sa før, som historisk student var det første jeg lærte meg hvordan man spør om det, hva noen sa og hva som skjedde, og du må spørre hvem som skrev dette, for hvilken publikum og hva de prøver å si med dette. Du, with this, you can say everything is subjective. There is not a thing as objectivity. In my mind, that is a myth. You can say we can have a common subjectivity, we can call objectivity, but it is subjective nonetheless. So we cannot just read text anymore. We need to read the intentions behind the text, and I somehow think that we forgot that part of reading.

Speaker 1 19:10

I think it's kind of interesting and every time I mention it people are like what? When I studied history in Germany, you were required to have a certain understanding, or minimum understanding of Latin, and if you studied old history or ancient history then you would have need an understanding of Old Greek, which is very much to get a have meet an understanding of old Greek, which is very much to get a first-hand impression of these sources and do not have to work with interpretations of the original right. And I think that that's a bit of a luxury to have that possibility to actually criticize sources and actually have that understanding of criticizing. And I think it's easy for you and me to talk about this because we have studied that. but these language models are introduced to everyone and not everyone has the privilege to dive as deep into the topic as we do. So what or who should be responsible to ensuring that we do understand the criticism that should be introduced?

Speaker 3 20:11

As of now, you cannot point a finger to one single entity or person to have responsibility over this. This is a shared responsibility between the manufacturers of these AI models. You can say OpenAI, Anthropics, DeepSeek, you name it. They should actually somehow embed it into their systems. Ai-modeller. Du kan si OpenAI, Anthropics, DeepSeek, du nevner det.

Speaker 3 20:28

De bør faktisk noen gang imbede det i deres systemer, som de forteller når de hallucinerer. Du kan prompere en AI til å si at. Du ikke er sikker på dette, men dette er min beste skatt. Du kan noen gang putte inn det guardrailet i systemene Og det er også på deg og put in that guardrail into the systems, And it's also on you and me as the readers. We should somehow teach how to read machine. We learn how to read text and we trust text, but we need to read the machines, And then it's just as a shared responsibility all around. It cannot be one entity It's not EU coming out with AI and say we should do. Entity. It's not EU coming up with AI and say we should do this. It's not working. So it's somehow that we need to push, as consumers, for this transparency for these big manufacturers of AI language models.

Speaker 1 21:26

There's something interesting that I've been pondering about, and that is the need for accountability, and I think that, well, in data governance it's my home turf. Right We talk we again talk a lot about accountability. There has been a while where we've talked more about business outcome and business value than we've talked about accountability. But the term is back and about ten years ago we talked about cloud accountability. Who's responsible for safety security of my data in the cloud? Is it a cloud provider? Am I responsible? Where does the line go? And I think we are at a point where the same discussions are happening towards large language models and ASs. So where does the accountability lie? And the big question is how much accountability are vendors actually willing to take?

Speaker 3 22:18

You can say vendors, if they are not pushed by the consumer market, they will not do anything. They can just sell their products, get a lot of money, fine and dandy. But it's growing consumermarkedet. De vil ikke gjøre noe, de kan bare selge deres produkter. De får mye penger, fint og dantig. Men det er en viss tro under at du kan. Si AI-genererte bilder er gitt ganske godt, noen ganger er jeg ikke sikker på om jeg ser en AI-bilde eller ikke, så da kan jeg ikke tro på en bilding mer som en l seeing an AI image or not. So then can I even trust an image anymore as a reader, as a consumer? And the same thing is coming with text.

Speaker 3 22:53

We are hearing high school teachers being on the edge. Can I really trust the things I'm getting from my students anymore? Is it AI generated or is it produced by a student? It's also you can train the AI model now to sound like yourself. So you're getting all the errors into the tech stream, so it sounds like a student in second grade of high school And that is just disturbing somehow. So we need also to somehow teach our young adults how to actually use this. You can also see now there was a news article the other day about Mattel wanting to place a chat-gibbet-like brain into the Barbie doll. So then you are seeing a Barbie doll with a generative AI language model inside talking to young children, and what are that actually doing for the kids' minds and their own way?

Speaker 3 23:48

AI-mål innenfor snakker med unge barn, og hva gjør det for barns, mennesker og deres egen måte å forandre? du kan si samfunnssyn kritisk og alt det. Når du bare er presentert med en mål som bare sier alt du sier er rett, de støtter, ikke tilbake mer. Det er bare ja, you should do that. It's fine.

Speaker 1 24:10

And the interesting thing and you pointed at your writing as well is that reinforcement you get of your own way of thinking through the model is also paired with a certain politeness, and I want to hear more about that. What does that politeness in the way you get your answers and you get that oh great question, you are on the right. Hva gjør denne politikken i måten du får ansatte og du får det «Å god spørsmål, du er på rett trekk, noe, noe» som begynner med din respons. hva gjør det med folk?

Digital Literacy

Speaker 3 24:35

Hvis du spør en spørsmål og den siste språkmodellen begynner å si at du er feil, så vil du bare vende den ned, du vil bare slå being polite and saying encouraging things to the users. So they are keep prompting the same machine and asking more questions, being more diligent with the content and wait even more with it. You know each token we send out is a token in again and it produce money and yeah, all that. So when we are having a large language model that is just polite to keep you in the conversation, is that really trustworthy anymore? Do we have a model that can challenge my worldview? If I'm saying the COVID vaccines cause autism and then the large language model will say, yeah, it sounds about right because, but it's not saying the opposite things. They're not saying, oh, i think you are going down a spiral here in a bad habit and then trying to trace you back to reality. So that's really the big downfall here in my mind.

Speaker 1 25:50

Well, covid is a good analogy here, right? Because during COVID, during lockdown about five years ago, certain things happened. Right. Fem år siden skjedde noen ting. Jeg husker jeg snakket med min far i lov og vi snakket om om barnet fortidig vil skjønne hendene etter covid, fordi nå er de veldig lært ikke å skjønne hendene med noen. Og du bruker hendsanitiser hverdagen og du finner dem i hver publik bygning og det, and you would use hand sanitizers everywhere and you would find them in every public building, and it took about not even a year and all that was gone. I wonder if that well, the certain things we learn as humans that are very hard to unlearn, or will that way of thinking, that way of talking, that politeness that large language models introduce, also spill over to how humans interact between humans?

Speaker 3 26:38

They most definitely will. You can already now see that the way the AI, or the large language model, is formulating text is influencing how people are writing. You can see that they are using the same common wording and ways of spelling, ways of formulating a sentence, and you can say is that the egg or the hen? You can say the last language model is based on text mostly produced by humans and then presented back to us, but now it narrows down into a more common path. And is that really what we want with language? nå nærmer seg til en mer vanlig vei, og er det virkelig hva vi vil med språk? Skal det ikke være mer bredt og eksplosivt? Det er en av tingene jeg gleder meg mest til.

Speaker 3 27:25

Om de yngre generasjonene Nå er jeg 37 år gammel. Jeg er ikke gammel, men når jeg snakker med yngre mennesker, jeg er mest fascinert av alle slangen de utgiver, alle de små søtte ordene de bruker bare for å sikre seg fra andre generasjoner. Og med AI, det kommer ikke til å skje. Med AI, vil det være mer nærmere. Du kan si at det er litt som med den berømte George Orwell novel in 1984, where you are narrowing down the language so it's more controllable, and that's actually one of the biggest threats with AI language models, in my mind, is that it's narrowing it down.

Speaker 1 28:09

A really interesting take. And yeah, there's something about language and the natural development of language, right? I think in Germany there's like there has been this yearly. Det er noe om språk og den naturlige utviklingen av språk. Jeg tror i Tyskland har det vært en sånn jævlig valg hver år Publik valg på jævlig valg av året Og det går tilbake i 30-40 år. Og det er interessant om du ser noen av de gamle videoene og hvordan disse ordene er fortsatt present i vårt språk og er fortsatt utvikling, and how these words are still present in our language and are still developing, and I think it's interesting to see that. Maybe you're right, maybe that development will be narrowed down into a certain phrasing, like in the world of.

Speaker 3 28:47

I can also say that is one of my most fascinating things about you could say Norwegian. They are not just borrowing words from other languages, they are actually translating it somehow into Norwegian. They are not just borrowing words from other languages, they are actually translating it somehow into Norwegian. I went to a psychiatrist at some point at a party and she was Norwegian and she told me to watch this show on Netflix. I should strømme det, so streaming, but then it was just directly translated into Norwegian strømme. I found it, so you can say in Danish it sounds nearly cute when you are just using this direct translation of a word. You can also download or nederleste an app on your mobile phone and in Danish language we are heavily influenced by American TV and we have been that since you know en app på din mobiltelefon Og i dansk språk er vi veldig innført av amerikansk TV og vi har vært det siden Barry Hills, friends og alt det i 90-tallet.

Speaker 3 29:45

Så mange ord vi bruker i vår dag til dag ordning er engelsk. Mine barn er tre og fem og de bruker også engelsk ord når de snakker, og jeg har funnet English. My kids are three and five and they also use English word when they speak, and I found it a little different from time to time, and people often say, oh, that's not a good Danish expression for this. Come on, there is a lot of words we are not using in the Danish language and we should really cultivate this use of language across all generations and all countries.

Speaker 1 30:19

I like that our conversation naturally flowed into how language is generated, which I think is really interesting because it very much shows how large language months are affecting our everyday life and the way of talking, the way of writing, even the way of thinking, av å snakke, av å skrive og til og med av å tenke. Og tilbake til vårt område, om at noe ikke er sant bare fordi narrativene er sagt i en sann og politisk måte. Og hva tror du er og du skrev det i ditt tekst død til å ikke forstå H.

Speaker 3 30:57

Du kan si at. når maskinene alltid sammenligner med oss, stopper vi å tenke kritisk. Disagreement er og skal alltid være en del av læringen. Vi må noen slags å designe systemene så de ikke bare knutter sammen med oss. De må stå tilbake, kanskje bare lett for å fortsatt holde folk i løpet, men vi bør kunne designe systemer som kan stå til to still keep people in the loop. but we should be able to design systems that can push back a little bit. We can say, unless we encounter resistance, unless we examine our own thoughts anymore, if an AI system always mirrors back what I believe, what you believe for worse, or they are just flattering us, we lose the internal tension that fuels this critical thinking, critical thoughts.

Takeaways and Call to Action

Speaker 3 31:43

It feels good to be agreed with. You know that When people are saying that's a good post, venk Frid, you did spot on. You're feeling it's good, oh nice. But the real tension comes when people are saying, yeah, it sounds about right, but have you thought of A, b and C? Or I strongly disagree with A, and then you can have this conversation And I really think that we should push for that conversation. Agreeableness is addictive. We know that. we want to be agreed with, but when we are looking to echo chambers, we are not seeing growth, we are just seeing stillness.

Speaker 1 32:22

It's interesting that we create machines that create an echo chamber for us to feel good about ourselves.

Speaker 1 32:29

So that's one term that has been thrown around a lot when we talk about both ethics and large language models and about competency around it and that data and AI literacy And I just had a talk with an analyst at one of the big analyst houses around the topic and we talked about literacy from a perspective that was very much inside an organization, saying that literacy, data, ai literacy is not at the end itself. It's a means to an end. It's a mean to create productivity in your organization, to use AI more effectively. But I tend to disagree with that. I think there's something there in AI and data literacy that goes beyond that. That is very much, and I think the possibility for organizations to have a positive effect on society i AI og datalitteratur som går utenfor det, det er veldig mye, og jeg tror at muligheten for organisasjoner å ha en positiv effekt på samfunnet ved å introdusere datalitteratur i organisasjoner, men også med fokus på hvordan man kan oppføre det i sosialt liv utenfor verden.

Speaker 3 33:30

Ja, jeg er med. Du kan si at i dag er de fleste mennesker digitalt fluente. Vi kan bruke hjelp. say that today most people are digital fluent. We can use tools, apps, search engines with ease. We can do anything. I almost grow up with a computer in my hands, more or less. But it's not the same thing as digital literacy, the ability to question, interpret and evaluate the systems behind the interface. And if we are using the AI phrase in the age of AI interpretere og evaluere systemene bak interfeisen Og hvis vi bruker AI-frase i AI-en, så blir det en viktig forskjell. Vi ble lært hvordan vi kan bruke maskinene, men nå trenger vi å lære hvordan vi kan lese dem. Vi trenger å lære hvordan vi kan spørre de rette spørsmålene igjen. Vi trenger å se hvem som har need to learn how to ask the right questions again.

Speaker 3 34:16

We need to see who trained this model on what data. who are left out. is it only based on white males 30 years old or is it also black women from America who are in the dataset and be transparent about it? And we should also see which assumptions are baked into the output. When you are asking the same question to different language models, you're getting different outputs. They are more or less trained on the same datasets. They are scripted all the internet. but if you are asking chat GPT about Hamas, you get one answer. If you're doing the same thing for DeepSeek, you get another answer. So there is also a worldview baked into the system, into the training system here, and we need to be transparent about it.

Speaker 1 35:04

Right, yes, i've seen a video just recently about the trolley problem. Right, you are at a trail track and the train is heading towards the track where there are four or five people on the track That would be at en trailtrekk og trengeren går mot en trekk der det er fire eller fem mennesker på trekk. Det ville være det at trengeren ruller over, men du kan influere på situasjonen ved å rute ut trollen til å gå på en annen trekk der det er bare en person. Og det trollproblemet har vært pågående til ulike AI-modeller. Og ja, hvis du har det basale, the basic trolley model, the answers were quite similar. But as soon as you introduced other variables, like something that's a cat and not a human, the outcome was entirely different and the gap was getting bigger and bigger between the models.

Speaker 3 35:49

Yeah, you can also see we need somehow to not just teach kids but also adults around the world that AI is not giving you answers. they are giving out output, it's probabilities, it's patterns and by popularity, you can say the most common answer from that model is what you're getting. Literacy in that context means that we need to embrace there is some sort of uncertainties and we need to teach people at vi må oppføre, at det er noen slags uansettelser og vi må lære folk å pause for en sekund, tenke på hva de har lest og da prøve å interpretere det i sin egen mening, og ikke bare ta utsatting som fakta. Maskiniliteratur er i min mening om å rejse AI-systemer. Jeg er all for å bruke. It could be great tools to actually use as a sparing part of something board when you're writing something, to see is there something I'm missing but you also need to learn how to prompt them correctly and ask the really good questions. Most people are not learning that at the moment.

Speaker 1 36:51

Yeah, and as we talked about, i think there is a room and place for that learning to happen also in organizations. So this is where, ja, og som vi snakket om, tror jeg det er en rom og en sted for at læringen skal skje også i en organisasjon. Så dette er der jobber kan ta en certain del av det. responsable topic is just broadening as we speak, but I want to narrow it down at the end to your key takeaways and, if you have a call to action to people, what should you think about going forward?

Speaker 3 37:23

I think that what we are talking about here today is actually the difference between fluency and understanding, Between that answer sounds right and answer that actually holds up in court. You can say when you ask where they're coming from. My term of promptism is tempting because it makes everything feel effortless, But the real thinking isn't smooth. You know when you're sitting down at your table. You need friction, You need doubt and disagreement. The best friend in the world isn't a friend saying you're right all the time. It's somebody that's challenging you And we need somehow to get rid of some of the politeness in those machines, in those language models, so we can actually grow as humans.

Speaker 1 38:13

Grow as humans. Fantastic. Thank you so much for the conversation.

Speaker 3 38:17

Thank you.