Root Causes 530: Introducing the AI Iceberg

Hosted by

Tim Callan

Chief Compliance Officer

Jason Soroko

Fellow

Original broadcast date

September 29, 2025

We compare AI in 2025 to Internet in 1995 and describe the AI iceberg, including the majority of applications which are below the waterline.

Podcast Transcript

Lightly edited for flow and brevity.

Tim CallanSo recently we talked about did an episode where we said, AI is the room.

That made me think about, let's go back in time to 1995. So it's 1995. The World Wide Web, all of a sudden moves into the public consciousness and takes the world by storm. And yes, internet and even World Wide Web had been growing for a while, but suddenly Netscape goes public, and common browsers are commonly available, and there's an IP stack in Windows, and just all of a sudden, everybody's using this thing. And I remember at the time, there was this idea of “internet technology” or “internet products”. What I think a lot of people didn't grasp was it's not internet technology or internet products. It's just technology. It's just products. That internet is just part of the fabric of things. The same way that we don't say, hey, this is a nice earth couch I'm sitting on. It is just a couch because, of course, it's an earth couch to the point where earth doesn't need to be said. Hey, Jason, this is a good earth podcast. I like your earth microphone there. Now just take the word earth out. It's all earth. The same was true of internet. And it took a lot of people a long time to get that. Now we do.

Jason SorokoWe don't call this an internet podcast.

Tim CallanWe don’t say this is an internet podcast. I think we've got a similar thing with AI. AI is just part of the basic fabric of procedure. Like if there's a chip, there is going to be AI. And we need to stop saying, well, I have AI applications, or I have AI businesses, or these AI use cases. They're just use cases. They're just businesses. Like internet.

Jason SorokoPrecisely. And that is why Tim, we said it. I remember saying it right here on these couches - earth couches - -

Tim CallanEarth couches. Like that earth chair you're sitting in.

Jason SorokoThe elephant in the room – AI. Well, it's not the elephant in the room. It is the room. Therefore we can't distinguish it from anything else anymore. This isn't an internet podcast, and that's why you're going to hear us talk about AI unironically, and we're not going to explain ourselves.

We're going to be talking about AI topics on this podcast because it's just, how can you not?

Tim CallanIt's just living. It’s just being in the technology space in 2025. Pure and simple.

Jason SorokoSo on a PKI digital identity, certificate lifecycle management type podcast, where we also delve into cybersecurity and automation of all different types, AI is going to pervade, and currently pervades a lot of it, and will pervade it even more.

Tim CallanI think also we're going to see, just as you and I have had AI episodes and AI topics, expect that also to be less the case moving forward, as just that the episodes and the topics are going to be informed by understanding of AI the exact same way that everything else is. Just like we don't have internet topics. Or earth topics.

One of the things about how a lot of people are thinking about AI right now and again, this is the same as it was in 1995 with internet, is there's what I call the AI iceberg. So there's this little bit that sticks up above the water that we all know, which is, I go to ChatGPT, and I ask it to give me a picture of a fluffy kitty, and I get a fluffy kitty.

But the stuff that's below the surface is the vast majority of it. I talk to people who aren't really in the tech space, friends of mine, and I say, oh no - - they say, oh, I've never used AI. I say, well, sure you have. Have you searched on a major search engine? Sure. Well, those results are informed by AI. Do you use any kind of digital process? Do you use social media? Well, that's informed by AI? And so most of how AI is going to serve us is not going to be in I'm sitting at a prompt and asking for a thing and getting a result. Most of the way that AI is going to serve us is that a service that we use will be augmented. It'll be faster or more accurate, or superior or better anticipating my needs, or have new capabilities that wouldn't have been possible otherwise and how it's augmented is implementations of AI technology, and that's the part of the iceberg that's below the surface. And that's the majority of it. And that's the most powerful part of it. So that's the AI iceberg, and I think most people don't think about it that way.

Jason SorokoNo. Think about the internet. The TCP stack. If you're a network engineer, the amount that's above the water of the internet has gotten really quite small.

Interestingly. Which is not a bad thing, because it means that we all get to use it, and don't have to think about it much at all. It's below the water. Like we don't think about it so much now, the only thing we panic is when we lose connectivity, Wi-Fi, things that connect us to that TCP stack without us thinking about it.

Tim CallanI think that's an interesting way to put it too, is people always talk about the only time you notice the power or the only time you notice the water is when it doesn't work. I think the true is same is true of connectivity. Bandwidth. I think same is very shortly going to be true of AI. When this neat thing that just happens automatically for me stops happening, that's when I'm going to notice.

Jason SorokoAbsolutely, Tim. So you had talked about how functions, a lot of functions will be automated, improved based on AI. You won't notice it.

It will happen in ways that unless you're even a developer, you might not realize what's going on underneath the scenes. A lot of what had to be coded procedurally in the past, things where a lot of difficult edge cases in code have to procedurally be coded can be quite costly to code. Those are the ideal use cases where you collapse that code into one AI call and AI makes the edge case decision that feeds back as a function into your software.

A lot of people won't realize that that's happening at all, but it is a place where AI is right now rampant in all new code bases. But Tim, you said something interesting, which is, the top of the pyramid. At top of the iceberg is small. You and I were talking about this last night, I'd like to make the argument that the top of the iceberg is, in fact, going to become really microscopically small for the majority of people, but there's a gigantic however. For those people who want to dive deep into what AI can actually do, the top of the iceberg is actually way larger than most people realize. Here's my argument for that.

Most people are completely unaware of the concept of an embedding model. Like in other words, most of us understand the GPT that we use, the generative transform system that responds back to us in a chat bot. However, there are entire models that I play with that have to do with chunking information into generative systems to be able to train and fine tune LLMs of my own. Both small language models, offline models that I used for very specific purposes for me, and these things have become incredibly powerful. And they're incredibly powerful because there's an entire set of tools that people who are aware of how to use them can make AI sing and dance and do things that none of us could have even imagined not that long ago. So there's an entire class of technology above the water line that I think, Tim, is worth talking about. And in fact, we probably should have some more episodes about this down the road, because it will literally define the haves and the have nots in the future. That's what I believe. It just means that there's a set of tools out there that I think people who are listeners to this podcast should be aware of. It's too easy to become a technical geriatric right now.

I don't mind that my mom only thinks about ChatGPT as that GPT box.

Tim CallanI think your mom is going to benefit tremendously from AI without knowing it. Your mom is going to benefit from AI by using technology, products and services that themselves benefit from AI.

I think that most humans, that's how they're going to - most earth humans, I should say - that's how they're going to benefit from earth AI. What you're talking about, I think, is a relatively small subset of people with the skill set and the predilection to put their hands directly on the levers.

Jason SorokoWhat is remarkable, though, is that, let's say you became a networking expert back in ‘95 and therefore what you were doing above the water line was - - most of us have collapsed what we do above the water line with respect to the internet.

I think with AI, there's the difference. I think there's always going to be a significant amount of stuff above the waterline that I think that people who are listening to this podcast should consider themselves. This is applicable to you. In fact, Tim, we talked about some of the use cases last night. Something as absolutely basic as, how do you connect via an API to an AI system in order to be able to have it generate something differently in a different way. And it could be something as simple as, hey, I want to get the absolute best thumbnails for my YouTube podcast from a model that I can't just get from the chat box. That's like kindergarten level one.

Tim CallanYes. The YouTube thumbnails - - by the way, we've never said this on the episode before, are AI generated.

Jason SorokoI do that on purpose. Imagine if we had hired out somebody, or you or I did the artwork in Photoshop, whatever. Choose your art tool of choice. Well, we would have been doing things the old way, and there's still absolutely a need for all of that. But to me, it's like, well, if we're gonna have a podcast that's AI is the floor and the walls of what we talk about, why not right from the thumbnails themselves generate something from a system that's not human at all. Why not leverage that kind of a tool? And so yes. Ever since episode 500 all of our thumbnails are AI-generated. And not even partly generated. Something imprinted on top of a raster. They are 100% every pixel generated.

Tim CallanIncluding the words?

Jason SorokoIncluding the words.

Tim CallanI guess that's a nice, concrete example of what you're talking about.

Jason SorokoYes. Let me push it one step further for you, Tim. So let's say you and I were to teach somebody how to do that. We give them that little bit of code template, we fire up an integrated development environment, and we teach a person how to actually fire up an API call to make AI do something a little different than what they could do right from their chatbot box, that's the basis from which you can do a ton more things.

In other words, all of a sudden, and I think I said this to you last night, you can actually start to ask AI, hey, now that I have this environment available to me, and I have a template script, a little bit of Python, whatever it happens to be, all of a sudden, now you have the environment and some of the basic know-how to start asking, AI, hey, I want to do some other things. Write the code for me.

So that's when you realize, wow, this no longer has limits. And I think part of what we were just saying is that the sum total of potential for the future have never been bigger than right now.

Tim CallanMy whole life has been a story of acceleration in the pace of technological improvement. You can see that continuing and possibly continuing exponentially where the pace of acceleration itself is accelerating increasingly beyond what we're used to, just as a product of this true, absolute integration of AI into every process. Like I said, if it has a chip, it has AI.

Jason SorokoTim, let's bring it right back to home. Certificate lifecycle management. I don't want to call it AI-CLM, predictive CLM.

Tim CallanIt’s earth CLM.

Jason SorokoEarth CLM. It's just CLM. Yes, it's going to have AI on it. It’s going to. Therefore, why talk? We don't talk about - -

Tim CallanPredictive CLM, AI-CLM. It’s just best CLM.

Jason SorokoExactly. We don't call it an internet product. That would seem silly. We're declaring right now calling it an AI, AI anything is silly because it's just what it is.

Tim CallanIt'll be interesting. I didn't manage to make it to RSA this year. First time in a long time. It was a little disappointing. But the year before that, I reported that there was like every booth had to have the letter A followed by the letter I. I think we got to get beyond that.

Jason SorokoWe do. What you're referring to, Tim, what you're referring to, very specifically is that below the waterline of the iceberg is gigantic already, but it's going to become most of what's under the water. That's the truth. But here's the funky thing that the analogy no longer holds. Where the internet, the top, what was above the water line of the iceberg got so small so that, who even plays that much with the address bar in their browser? That's above the water stuff like. That's about the closest people get to actually doing a TCP call at its core. Now people just playing with their apps, and people are just browsing away, and they don't even think about and it's so abstracted away that even though there is an ice break above the water line, people don't even think about it anymore. My argument is that analogy does not hold with AI. My argument is this, the above the water AI stuff is actually going to be the most interesting. Now it's going to be far larger under the water. That is something I agree.

Tim CallanIt’s gonna be everywhere.

Jason SorokoIt's going to be everything. I completely agree with you. It's just, if you ignore the stuff above the water, all I'm saying is that you're missing out.

Tim CallanI think let's challenge ourselves to return to this topic and start providing a greater level of specificity on detail and what we mean by that.

Jason SorokoIn fact, Tim, I think what we're going to do is we're going to start to teach on this podcast how to do certain things. We might even have a little bit of you and me interactive sessions where we record ourselves.

Tim CallanI love it. Let's do that.

Jason SorokoAnd so if Tim and I can do it, you can do it too.

Tim CallanAbsolutely. All right. Thank you, earth, Jason. This has been earth root causes you.

Stay informed with expert insights

Subscribe to Root Causes for engaging discussions on PKI, digital security, and best practices for protecting your organization's critical assets. Don’t miss an episode!