Why Data Governance and Modeling Need a First-Principles Rebuild with Joe Reis

ood morning. Good afternoon. Good evening. Good whatever time it is wherever you are in this amazing planet of ours.

I’m Malcolm Hawker, host of the CDO matters podcast. Welcome back. We had a three month sojourn over summer break. We took a summer break.

I mean, I haven’t done this for three months, so I’m I’m a little rusty and…

So forgive me if I massively screw anything up. Also, I’m getting over some sort of lingering plague, and it sounds like I have been pounding Marlboro Reds all day. I have not. It’s just a normal new raspiness of of my voice, thanks to the the plague I’m still recovering from.

Today, we’re gonna be talking about governance. We’re gonna be talking about data modeling. We’re gonna be talking about AI. We’re gonna talk about all sorts of fun stuff because we’ve got two really fun guests. We… You you got… You’re joining CDM Matters podcast, you got a BOGO.

Buy one, you get one. We we we… We’ve got two of the brightest minds in the data and analytics industry. We’ve got Winfried Etzel. Winfried, welcome. And we’ve got, of course, mister Joe Reese. Welcome to you both.

Hey. What’s up? Your voice reminds me a lot of Drewski’s new character.

The chain smoking character. You should go check it out. Anyway, good to see him.

Likewise, Winfredge, thanks for joining us late.

Good seeing you. Yeah, it’s late. Well, it’s 08:00.

Eight 8PM. You’re in Norway, right?

I’m in Norway, yeah. West Coast, Norway.

Okay. Whereabouts?

Davenra, it’s called. It’s Norway’s oil and gas capital.

Yeah. Okay. Yeah.

So…

I’ve I’ve I’ve been to Norway a couple of different times to to the north, this place called Lofoten.

Have you ever been up there?

No, haven’t. No. Oh, it’s stunning. Oh, you got to go. Oh, it’s it’s it’s it’s beautiful.

Just South of Tromso, it is it is breathtaking. You get these mountains that meet the ocean, and they’re these jagged sharp mountains, and and all these little islands and roads can… Anyway, and the beaches, like, believe it or not, like, you’re inside the Arctic Circle and they look like the Caribbean.

Yeah. I I had a friend visit me from The US and he said it it looks a lot like Hawaii.

I was like Yeah. That’s so strange.

Yeah. Without the palm trees. No no palm no palm trees. In and in, you know, and instead of people running around in bikinis, there’s reindeer.

Same thing. Yeah. Exactly. Yeah. There’s a there’s a there’s a lot of reindeer. So so let’s get to it. You both have books that are coming out very shortly.

Winfrey, yours is Data Governance in the Wild. Yes. Unless the title has changed. And Joe, yours is Mixed Model Arts.

Yes. I I love the the title. Joe Joe, tell us what’s the main thesis of your book? What what why as as as a as a data leader should I be going deep on on mixed model arts?

I mean, it’s mixed model arts as the title sort of implies, it’s it’s about data modeling.

The thing we never seem to have time to do, but we seem to go back and fix over and over again after the fact. And so this book, I think, really bridges the gap between, you know, today and the last, probably, quarter century or more of what I would say is the lost art of data modeling, you know, and bring it back into the present and the future. Because now we’re modeling not just for humans, we’re modeling…

We…

I’m including our our machine brethren here, our agents. They’re also modeling, and so we’re in a different phase of data and technology at this point. And so, yeah, this book really will help the next generation of practitioners and and leaders, I think, come to terms with how to do data modeling. But it also takes a step back.

Think there’s…

It’s a two book series at least, probably maybe three, but it…

Oh. This first book really describes the building blocks of data modeling. So we describe data from first principles and what is…

What are the building blocks of data itself, and then how do we model these? I think that’s actually a treatment that has never been covered in sufficient So, the goal of this book is obviously to bring back data modeling and present it. And the mixed model arts, the the…

Guess, the kind of the pun on that is, you know, I I think the data industry is going through its mixed martial arts moment right now, sort of the UFC moment where you had all these different styles, you know, be kind of before the UFC time that there was always a question, like, would win a…

In a boxing match versus a wrestler, and, you know, Bruce Lee beat Hulk Hogan or something stupid like that. And then, of course, you find out, like, you know, with the UFC, like, yeah, you should probably learn all the above if you’re gonna be a a competent, well rounded, you know, combat athlete. And the and the same thing is happening in data right now. You’re having a convergence of, you know, like, applications, you know, software engineering, analytics, AI, ML engineering, and so forth.

And the knowledge world too. Right? The ontologies and graphs and all these are…

I think we’re all fusing together across different modalities of data. Right? So you have unstructured, semi structured, and so forth. So it’s really a recognition that the world is is converged and we need to keep up with it.

So See, you touched on something there around the idea of agents being the consumers.

Right? And now, Winfried, I think that’s core core to your upcoming book, which is the idea of we need to change governance. We need to change the rules of the game in order to better reflect the fact that the agents are the ones that will be abiding by the rules and maybe even recommending the rules. So so tell us a little bit about your book.

I think what what Joe just said that really fits also into the framing, the wider framing of my book.

And I don’t know if you have those. In Norway, we have in the supermarket like mix and match a candy, so that’s like an open candy and you get a little spoon of everything.

And that’s kind of the idea, right, that we’re seeing right now in the data world. And, someone has to make a decision, right? How much candy can you put in your bag? What candies can you put in the bag? What does actually work together? And, the way we’ve done governance has been experienced to have something slow, something bureaucratic, a bit disconnected from the real world.

And, at the same time, not governing is not really an option anymore, right? We see regulatory pressures increasing. We see data being reused across domains, reused for different purposes.

And we also see decisions that scale beyond individual teams, that scale beyond humans. So, somewhere we need to place accountability for it. And I’m trying to look at data governance also from a first principle perspective. So, if we scrape away all the frameworks that we have built over the years and look at the core why are we doing governance in the first place, what is it really about, I think we get a better starting point to rebuild into governance.

That’s kind of the That seems to be…

So what you just said is approaching governance from a first principles perspective.

I I kind of heard that in what you were talking about as well, Joe. And and I really I really, really like this idea because I think that so much of what we’ve built over the last thirty years has become so monolithic and so big and so inflexible and so difficult to understand or to reverse engineer. In many ways, I think we may be better off just starting from scratch. And I… And that’s a tall order for a lot of companies because most companies… I mean, we’re we’re cruise ships. Right? We’re not we’re not we’re not a lot, you know, innovation risk. That… That’s scary.

But what what do you think about the response to that, Joe? I mean, it’s… It seems to me like we… We’ve we’ve created a lot of gigantic monoliths and kind of… Now we’re trying to… We’re figuring out that when we deploy AI into these monoliths, maybe they’re not as effective as they would otherwise have been. Is that kind of a… How how do you see this problem of this… These gigantic data infrastructures that may not be serving us very well?

Yeah.

I I think it’s an accurate observation that we’re all, I think, now figuring out, and I I think we’ll continue to figure out even more.

I mean, the simple matter is that the architectures and the systems that we, you know, came to rely upon and and love and cherish and grow and and feed and care for, these aren’t adequate for the worlds of agents. Right? You’re dealing with agents that… Whose query patterns are immensely, you know, greater in in volume, you know, velocity variety.

It it sort of harkens back to the three Versus of big data. I think we’re actually back to that moment once again about with And so, you know, now it’s a…

Both a producing and a consumptive force of…

That we have to contend with. These things will be producing data both, you know, in the three…

In the matter of the three Versus as well as consuming it in the matter of three Versus sort…

It’s simply that our architectures aren’t able to keep up with this. It’s actually I was having a discussion with some fellows over at Google last week, you know, who are, you know, built a lot of their systems. And and the the recognition there is, and again, they…

I think they know big data because they kind of bent it. But the systems there are not equipped to deal with the concurrency problem, right, of of this…

The sheer number of agents that will be hitting these systems. And so it’s forcing a rethink of what what does the database need to look like. Right? And then, also though, it’s a it’s a big question, like, companies aren’t just going to gut things out, so how do you retrofit Right.

The existing architectures to accommodate, you know, these new these new friends of ours. And this is…

I I think something…

It’s a question that every company is dealing with right now, every vendor and and so forth, but it does come back to, you know, the the primitive as well, the practices we surround with, the architectures are the architectures.

It’s probably the sunk cost. You bought it, get to deal with it, bolt some stuff on, if that works. Otherwise, gut it out, like you say. Think it’s it’s… So right now, it’s interesting because what you’re seeing with AI especially is some companies are just bolting it on to the company and expecting that that’ll just make them AI native, and others are sort of retrofitting their org, really, and trying to think, okay, so what does an AI native org look like? I was just having this conversation with this head of this CTO, pretty big company, just before we joined, and, you know, he’s quite… He’s asking, well, how do we… Is it is it better to sort of rethink our organization for first principles and and and architectures? I was like, yeah, I think that that would be the conversation that should be had.

Obviously, it’s a bit of a riskier one, but I think the risky approach is actually the safer approach these days. And so it’s it’s forcing a big rethink, but…

And and the bolting on part, it reminds me of what’s… You know, I wrote about this where it’s… What we’re seeing is very similar to what you saw in the… With the introduction of electricity where steam engines where companies would try and bolt on steam engines and turbines into their their manufacturing facilities, and, you know, you had to really rethink what what… What’s what’s the factory supposed to be designed as and for, and so it really wasn’t a few decades, you know, people finally figured it out, but it took that long. Right? And I think we have a better chance of doing this right, but it’s going be hard. A lot of companies, you know, got got a lot of institutional inertia that you got to deal with, you got architectures, you got patterns and habits, and all this data that you need to collect, and it’s a gnarly problem. What do think, Winfried?

Yeah. It’s a… I think it’s a it’s a big task, right, it’s a tall order to say, let’s rethink how we do things, first principles.

At the same time, I see a point of I came to the conclusion that there are three things in governance that we need to do, right? I’ll call it the data governance, try it. That’s very much the first principle approach to it.

We need to negotiate, all right? We need to negotiate and align around meaning, around expectations across stakeholders. We need to align between what our regulators want, what our business wants, between what the market demands and what our technology constraints are. And I think you can add on a couple more on those.

But that’s very much a role for governance negotiation. And then, something about direction, not just thinking shorthanded. And that’s a tough one, right? Because on the one side, you want to show your value to the organization, you want to create buy in, you want to have some quick wins, right? But you also need to look into the future and see how does it actually progress, right? Where are we going with our organization? And especially when you think about the third, accountability audit, ensuring that the outcomes can be explained, can be owned.

You need to have a certain direction, right? And you need to make sure that you align on your priorities.

You talked about that electricity industry history.

There was an experiment at a factory outside of Chicago in ’20s, Hawthorne, I don’t know if you heard about that one.

Yeah, yep.

And they looked into how can we ensure that everyone on the shop floor follows the rules, right?

So they talked about it. They talked about we’re going to observe you the way you work to see where are the mistakes happening and how can we ensure that they’re not happening.

And they realized that just the thought of being observed was enough to keep people in line. They didn’t really have to implement any measures except for saying, You will be observed, which is a kind of interesting one. And especially when you think of governance as for humans, I think that’s getting you somewhere odd. Just thinking about having something in place for observability development.

That’s that’s that’s kinda that’s kinda meta. I mean, that’s that’s like Heisenberg. Right? Like photons being observed versus versus not versus not observed. I’m seeing the same thing in… With with governance.

Where do where do you see the role of humans in all of this? I I mean, you know, Joe, for a long time, and and maybe this ship has sailed. I’m I’m not entirely sure that it has. But, you know, what’s… Where do you see the role of humans, particularly when it comes down to data modeling? You talked about, you know, v’s, you know, velocity and and volume. I mean, the scale at which these things, these systems are going to be operating and changing seems to suggest humans need a different role in data modeling than they’ve had in the past.

Yeah.

It’s something I think a lot about, and I don’t know that I have a complete answer to it yet. I think we’re still sort of in this, like, fuzzy middle of… I wouldn’t say middle, just a kind of transition between, you know, a phase shift, I suppose. Right?

Between the old worlds and the new. And and so it’s it’s a question that I…

When I talk to practitioners and leaders, I don’t think anyone has a sense yet of what’s the role of a human, what’s the role of an agent. Right? If I’m looking at my laptop over here that’s running, you know, codecs right now, it’s doing things on its own. It’s using computer use to debug something, actually formatting Winfried spoke right now.

And so, you know, so it’s writing code. I’m looking at it. I’m reviewing the code. Right?

Trying to be a good citizen here.

And then it just goes and does things, you know, clicking around like a like a person would. Probably better than it. Like, I had it running yesterday, working out some details with my book, and I I was like, okay, this is a better job than you think probably anybody I’ve seen.

And so it is a big question mark, actually, because of the questions I would come away with as I watch this is this is super cool. This just saved about two weeks worth of time, and it took about three hours and 20% of my weekly token allotment. That’s not that that…

It’s not that much money. Right? I can just buy more credits if I want to. And so they…

It did it did…

I guess, sort of just walking around, like, okay, this is interesting. It’s it’s a new astro model, obviously, just came out. But as these models keep improving, it it does make me wonder what what is what is the role of of a person? You know, I I think the institutional inertia of most companies, especially bigger ones, will sort of delay the implementation of the things that I’m able to do just on my own because this…

You have governance, you have rules and regulations and so forth.

But, of course, if you’re if you’re like following the zeitgeist as we record this, you know, there’s calls for, let’s just put a…

Let’s halt the the progress on the research and AI models. It’s it’s, you know, it’s all going to, you know…

There’s gonna be a catastrophe. I think was Dario or somebody has said that, you know, pretty much you’re looking at a, you know, a takeover of the Internet in a in a few months for a rogue bot swarm, and who knows? Right? I mean, you could…

Some people write this off as just marketing speculation and so forth. Others take it really seriously. So I don’t know that we even have a consensus view. But that’s sort of the world we’re in right now.

Right? I mean, if that…

If something like that does happen, say, hypothetically, we’re a rogue botnet swarmed…

And it’s funny because when I hear the the counter arguments, like, well, but they’re not intelligent. They’re not humans. They’re not like us. They’re not conscious.

Like, well, I don’t think they need to be in order to inflict damage. Like, that’s No.

Not No. No. Yeah.

Right? So it’s it’s sort of this… I I think it’s just this kind of weird kind of straw manning of the the arguments, but it it is forcing…

I know, you know, people like myself and Winfrey, we talk about this kind of behind, you know, in text or whatever, where it’s just, what is the role of governance in a world where, as you say, Malcolm, things are are a bit unmanageable by humans. Like, how are you going to manage a million bots? All of which are probably like more capable than you are at whatever you’re trying to stop them at.

I don’t know. Right? And so for modeling and stuff, I think it’s it’s… It it it depends upon, would say, the use cases and the systems in which we’re working along side the AIs, but I can definitely see a scenario where we don’t have humans working in these either. I mean, I I think we should be open to this possibility. And I think increasingly, as I see the capabilities of these models, it’s becoming more of a real possibility actually, so I don’t know, bottom line.

Yeah. I I I don’t know either. Winfred, I wanna come back to you to talk about accountability in a second, but but to the point about the, you know, the the swarms and the misguided swarms and the the wreaking havoc. In in my six months of tinkering with Claude Bot, one of the things that I’ve learned the most is is that is that guardrails can generally, not always, but can generally work, but you have to anticipate every potential eventuality. You have you have to give it very specific instructions on what not to do.

And if you just set these agents out on a specific goal, there are so many unintended consequences that we can’t even fathom because we can’t even think about what they might actually do because they’re doing it at a scale and a speed which is beyond our comprehension. It’s it’s like you you have to establish the guarders…

Guardrails and say, do this. Don’t do this. Don’t do this. Don’t do this. And it’s not enough to say, don’t be dumb or don’t harm humans.

Right? That’s not enough. You’ve got to be more specific.

And I don’t think we could we could ever… You know, this starts to touch on governance. Right? Around what the rules are. Right? Like, what can you do? But more importantly, what can’t you do?

And anticipating what all those are gonna be, I don’t know how we do that in a non production environment.

Right? Like, these these things just seem to go go loose, and and they’re, like, Monte Carlo ing themselves constantly try this, try this, try this, try this. And the things that they’re trying, I’m not even entirely sure that we could ever conceive of of building a rule around. It’s kind of…

It it it’s kind of like…

Yeah. I I won’t go into politics. Forget that. Win…

Winfred, what do you what do you what do you think? I I mean, you know, you’re big on accountability, and and I’m all and I’m all for it.

How do you make somebody accountable for a process where these agents are doing things that we could never have ever anticipated?

Right. And I think that that was interesting one here, right? I mean, you have to be very specific, as you said, on the guardrails. But I also realized by playing around with it that if you have a very long list of guardrails, some of these will overlap at a certain point, and some of these will be ignored.

So, I realize that the longer the list of Godware is, the longer the list of rules, the harder it is for the agent to follow those.

But that’s also something to take into account. I’m looking at it from a different perspective, just trying to redefine the role of governance and humans in general here.

Not looking at, Oh, we just need to be more rigid, right? Because we’re not going to solve it.

That’s just going to harm us in the long run. So, trying to think of the human role as more directing, right? Giving a certain direction, that’s something that we already do, but be more explicit about it. And then, that’s the accountability part.

First thing is, I think that the entire idea of human in the loop is a good idea, but it’s really hard to do.

At scale, yeah.

Yeah. And it’s also a too passive of role. I think as a human, we need to be in the lead to direct, to give a certain, this is the direction I want to go.

I don’t know, we had a conversation about this, Joe, a while back on security and AI, exploiting security for companies. And then I had a conversation about it with CISO, and he said, Well, I think we’re going to come to a point where you would place your own agents at your perimeter with the only task to diverge the other agent’s prerogative, right? And then, where’s the human there, and how do you keep control of that?

So in the book, talking about four different kinds of interactions, we have human to human interaction. That’s what we’re doing right now. We have human to machine interaction when we talk or chat with Claude.

We have machine to human when we process the output, and we have machine to machine interaction.

That’s what we are. And we’ve seen the last weeks happening. We have seen with JibberLink and Wobberg.

All of these need to be governed, right?

And the thing is, you need to have a certain human accountability at a certain point of escalation there.

If not, you’re basically saying that machines can be accountable for human to human interaction as well, and that’s kind of weird.

Well, I was going play devil’s advocate though, and ask you this this scenario where, say the machines become more integral in our day to day processes business or whatever.

Do you see a situation where the role is actually reversed? Maybe in machines, you said humans need to be governed.

I definitely can see that.

Yeah. And what would that look like then? Because we’re the ones coming up with these frameworks and then, you know, the bots are like, well, that’s nice. Except, you’re the ones that actually cause more problems than us.

You know, I’m already seeing this happening that we get pushed for in the data management sphere to use AI for data management. Right? With respect to some effort, you use data management, data governance to ensure certain AI behavior that you would like, but also the other way around. You need to have AI in data management to be more efficient. Then, if you have an AI agent looking into your data quality rules, for example, might come up with a better solution than you ever could, right? So, maybe the policy you are writing is not good enough, and you can get some input there.

But, at the moment we are now, it’s more of an augmented kind of way we’re looking But I I could definitely see this evolving.

There’s there’s examples out there already, Joe. I mean, radiology is one Yeah. Where AI is is already superior to to humans. And self driving or self driving cars.

Self driving cars are are are far superior from a safety perspective. But then you get into weird conversations around like bias and anti machine bias, we’ve got the, you know, all it takes is for one self driving car to get in an accident, and then nobody wants to ever get in a self driving car ever ever again, even though they’re infinitely safer than than human piloted cars. So I I I think there’s plenty of situations and plenty of use cases where we could be in a in a situation where the AI is telling us to, you know, pull up our socks and and where they may be get governing us. I’m kind

of intrigued by giving them professional certifications. Right? I mean, getting back to Winfred, your your point about…

Yeah. I mean, like, your your…

To your point about accountability, if if if we can measure the effectiveness of AI and say, like accounting, for example, and say, you know, and I bet QuickBooks has got this data. I bet I bet they do. But but whatever. That aside, we can just say somehow manage…

Magically produce this number that says AI is 98% accurate and humans are 95% accurate. Why wouldn’t we just give the CFA designation to AI and then have them be on par with the Charter Financial Account? Right? Like you say, you’re an accountant.

Now you can be…

Now the machine can be accountable.

That’s a little weird. Right? Because how do you sue a machine? Right? How do you how do you launch a lawsuit against a machine?

I don’t know. And and where is accountability there? Is it the engineer? Is it the the company?

I I don’t know. Things start to get really, really weird. But I could see a world where some of these professional certifications are are imbued on machines that are doing work better than we ever do.

Yeah. There’s actually a very interesting discussion in legal right now around this topic, in what kind of way can a machine be accountable. And self driving was the first step, right? This is what triggered it.

Because normally when you drive a car, the accountability is split between the car producer, if there’s anything wrong with the car, and you as a driver, right?

The car itself is not accountable, But what about the car itself driving, right?

And one argument I read that I found really interesting is that the entire idea of talking about AI’s action as human actions And, can help us to find a certain gateway into making machines accountable, which I thought was little interesting one.

Claude is thinking, like, no, it’s not. Processing.

There’s… What what what is thinking though, Winfried?

I mean, it’s literally been like a philosophical debate since like the dawn of time, and we still haven’t figured it out, so…

I I I I think it it appears that pure pattern recognition and pattern repetition, it it plays a much bigger role than, I think, than we’ve ever been willing to acknowledge. Right? Where, you know, when OpenAI came out, everybody’s like, oh, well, it’s a stochastic parrot.

Okay. But aren’t I?

Right? Like, I mean, that’s not… I mean, that’s that’s, you know, no response required.

But but in these patterns Yes, yes, you are, Malcolm.

Thank you.

As as I host a podcast, everybody’s listening just like, yeah. Give me some more of that stochastic parrot.

I I don’t know. So getting to your point about what is thinking. What… So what do you say, Joe? What’s thinking?

I don’t know. It it’s it’s it could be a lot of things. Right? And so it’s…

Because I… Because right now, it’s it’s interesting, and this comes back to the convergence I was talking about where we have…

Never did I think, like, software engineering, for example, would be intersecting with philosophy. Right.

Right. As I’m watching Astra go through and build Winfried’s book as we as we talk. I think that’s interesting. Right? But build. Right? It’s taking the action and it’s doing the thing it was set out to do.

Hopefully not leaking my entire hard drive into the internet in in the process, but that’s a different story.

But assuming it hasn’t done that, good product to the end. But, yeah, it’s it’s… The the philosophical debates are are back. Right? I would think in that backpack over there, there’s a book.

It’s the philosophical objections to artificial intelligence. It’s a whole book of the various arguments over the decades, you know, about whether artificial intelligence is actually possible or not. Right? And so, this gets back to the notion of what is thinking and so forth.

Are we are we dealing with with objects? Are we, know, if you were to go back and think about, you know, mathematical arguments about consciousness and so forth, you know, what level of abstraction are we talking about and so forth. So, it’s a very nuanced question and I don’t have an answer to it because it’s a question we’ve been debating literally since the dawn of time. I don’t think we’ve we’ve ever really come to a consensus on it, but we all do it every day, unless we are sarcastic parrots, as as we you know, as you’ve said, but…

Or as you as you use the term. I remember when ChatGPT first came out, right? And it was a it was it was a pretty dumb tool. Right?

I think the criticisms heaped on it were probably deserved. In fact, you know, one of my friends, he said he was on the team that released it, and he said it was they’re embarrassed. I think he used GPT 3.5. He’s like, why don’t we use that one?

That one sucks.

And they didn’t think more than 10,000 people were going to use it. Oh, interesting. Yeah. They thought it was just like, yeah, this is stupid. Nobody’s going use this.

And then people used it, and then things changed pretty quick. So it’s…

It is interesting in that regard where, like a lot of…

I think about this a lot, like that wasn’t even four years ago. Right? And and so, you know, a lot’s changed since then. Like the models right now, if you use Astro or Fable, depending on what you’re doing, like there’s no way you would say that as a sarcastic parrot at this point.

No. I mean, kind of literally is in the sense where like tokens sort of like the way they work, it does like find them the most probabilistically likely thing to put in there. But with like re…

Change…

Reasoning change and everything else, and, you know, and and chain of thought and then recursion, it’s like these are…

And these are becoming kind of different things than what you saw, you know, almost four years ago. Like, if you were to…

And Sam Allman had this point, it was at last year at Snowflake Summit when somebody asked him what AGI was, like…

I think it was Sarah Gowan who asked him this, like, well, how would you define AGI? He’s like, well, mean, you can maybe make an argument. You’re kind of already here. Because if you were to transport the models of maybe even last year to five years ago, there’s no way you could say, like, well, that’s that’s ********. Like, you know, that’s that’s a toy.

I mean, you’d be like, yeah, that’s pretty magical actually. So it’s funny because we keep like moving this goalpost of like what thinking is and thinking models and AGI and all this stuff. But like Astro was supposed to be, you know, OpenAI says it sounds really cool. Like, we have AGI now.

But as I use it, I’m like, well, it’s a convincing employee for sure. Like, it’s it’s very good at doing tasks, better than I’ve seen anyone do. And so it’s hard for me to argue that, you know, it’s it’s just a dud. But is it thinking?

I don’t know. Are we thinking? I I think so.

The way that you can get into the whole kind of like stoner argument in your dorm room at two in the morning, like Right.

Yeah. It’s the universe on the end of my pinky. Yeah.

Yeah. Exactly. So, I mean, since we’re we’re not in a dorm room at two in the morning ripping ripping… Taking bong hits and stuff, I guess we’ll just… Winfried is. I was kidding. But what do you think… What… I don’t know. What are your thoughts on this Winfried? I mean, it’s…

Yeah, I’d love to hear… And Malcolm too. Like, what do you guys think? No pun intended.

I don’t know. I think it’s an interesting debate. Don’t think it would be different if we didn’t need to fit AI models and AI behavior into our social world as humans, into our human world.

It would be an interesting idea to take away the human element and see how it would behave without trying to compare it to how we humans behave.

But that’s the reality, right?

Sociotechnical system, to use the term, right, we are co organized between humans, machines, technology, and it’s all tightly connected. And, that’s kind of the hard part, right? You can’t just separate one element and say, Well, the Amish can.

You separate it out One element and say, Well, we’re just going to do the human element. But, as us normal people, I think it’s a tough one. So, I’m trying to… You mentioned philosophy. I’m trying in the book to be quite multidisciplinary approach.

Not looking at data governance from one perspective only, but bringing in business analysis, risk management, corporate governance, philosophy, history, what else, political science.

What do mean by Really?

Wittgenstein and a few others like that, right? Well, I mean Yeah.

I think Metracitas, among other things. So these are, interesting because like, you know, he’s he’s become like sort of this figure that’s gone back into vogue now because of LLM. Sort of like, what is the nature of language and and our thinking? And, you know, is is the limits of my language or limits of my thought? And and and things like that. And so it’s, it is, you know, it is interesting and definitely…

I mean, Alan Turing did this, right, in the imitation game. Yep.

You know, it’s more… He he… If he noticed, like, he didn’t define, he explicitly didn’t define what thinking or artificial intelligence or any or any… What the goal was. He just said, like, think of this scenario, right, of, you know, trying to decipher is the person talking to a machine or not?

And I think that was really the…

It it feels like it was intentional where he didn’t explicitly say, like, here’s how you define machine consciousness and stuff or machines. It’s just like, yeah. I don’t know. Like, I think it it feels like it’s more of an exploration of this idea.

But it is interesting, right, because we’re on a podcast called CDO Matters, and if you if you bring it back, like, these are the discussions I’m hearing from people now in in the field. Whereas, like, if you rewind a few years ago, it was like, well, what data stack do you have? What…

How should we…

What what is the role of a chief data officer? And now it’s…

These discussions are a lot deeper, I would say a lot more interesting than they used to be. And so, you know, it’s…

I mean, I’m I’m intrigued by the perspective of, you know, us humanizing something that maybe can’t be humanized, which you were just gonna maybe suggesting at Winfrey perhaps.

Another way to look at it in a very simplistic way is us trying to enforce deterministic rules on something that isn’t inherently probabilistic. And to me, it’s the boundaries of those two things is where there’s some friction and and where things like governance get super super hard.

Because these things are behaving in inherently probabilistic ways and in ways that we can’t fully model outside of any sort of production environment, and I think that’s kind of defines humans to a t. We’re we’re… We’ll all do dumb things at some point in our lives.

We’ll all say dumb things at some point in our lives.

Yeah. Said like 12 of them in this podcast alone. It’s great.

Exactly. Yeah. We’re we’re…

I’m working on it. But to the point about to the point about AGI, I don’t think we’re there. I I think the the goalposts are being moved. As I understand AGI, to me, the very least, that’s the point at which AI is coming up with truly novel perspectives that don’t exist in the training data.

Right? Like, right right now what we’re seeing is just the ability to to chew on, consume, and re…

You know, and digest, and then republish a massive corpus of data at speeds that are well beyond a comp…

Human comprehension, but it’s still there. It’s all in the training data, and it’s just running all these simulations. It’s running it’s running all these permutations, all these combinations of data that already exist that starts to take on the appearance of being new. And and I think that if we keep pushing and pushing and pushing and pushing and and we fold collaboration into this, right, which is what all these forms are doing.

They’re they’re collaborating. Right? Because they they…

Apparently, they they believe that functional specialization is good, but with collaboration is also good, which is also a social phenomenon.

I think when we keep pushing and pushing and pushing and pushing at that boundary, I I think there will be an… At some point in the future, a leap. And the leap will be that they’ll be go… They’ll they’ll figure out something that was never in the training data to begin with. That’s AGI. Now the question is, do we have enough power and compute to do that? I I think that will be ultimately the the the barrier in the short term is the… Is power and compute. I don’t think it’s necessarily the capabilities of the models, but I could be wrong.

I mean, I was talking with somebody from Stargate about this a couple years ago. It’s right when it was announced in a, you know, $500,000,000,000 data center initiative, and I was like, I turned to my friend, Dahala Nelson, actually, was a mathematician Yeah. And wrote a few books on math and ML.

I just commented I like, you know, at $500,000,000,000, you’d think that, you know, given the fact that a human brain operates on 20 watts of power, we might dedicate some of that 500,000,000,000 to figuring out maybe better math or more efficient models or more efficient ways of using GPUs and so forth or whatever we decide to use.

Because the disparity between 20 watts of what we’re trying to simulate, and then something that’s, I don’t know, 40 gigawatts or whatever seems, you know, so many many orders of magnitude different of a problem that we’re trying to solve. And so, you know, hopefully things like what, you know, Yan Lakun is working on and others like that in terms of, you know, just kind of world models hopefully, you know, puts us on a better path. Because it feels like, you know, I live in Salt Lake City. They want to put up the world’s largest data center at the edge of the Great Salt Lake here. Thankfully, the public is really pissed about this because it feels like it’s also been railroaded through.

But it’s, you know, it it… But it comes at a time. Right? Because, I mean, you’re right. If we were to go on the current trajectory of models, we’ll need more chips, more power. But this is exactly against a backdrop where, like, the public has a lot of concerns about how these data centers are coming in to their communities, and suddenly everything’s more expensive, and there’s pollution, strangely enough. Nobody was told about this. And then, of course, you’re told, well, AI is going to wipe out your jobs, you’ll have no livelihood.

Have no life actually because you’ll be killed by AI in about a year So, you know, but let’s put a data center in your community. And, you know…

Or it’s at the time also, let’s see, where all social benefits are gonna be cut, you’ll have no safety net when you lose your job and you’re sick because of the data center. So, I mean, this is what people see. Right? I mean, this is just me being hyperbolic.

It’s like this is the conversation that’s happening in The US and around the world right now. Or people just see this as like, okay, who is this benefiting actually? Is it benefiting me and my community and my friends, or is it benefiting the billionaires?

You know, this thing… So this is, you know, not to get political, but it is a political thing.

But I think this is one thing that’s going to maybe stall AI, plus the fact that, you know, now the industry itself wants to slow things down for one reason or another. I can only speculate what the real reason is, but. I’ve got I’ve got a…

I’m struggling a little bit with this idea that, you know, a a I isn’t seeing the massive gains that everybody thought that we would get because I don’t know a single engineering team that isn’t 40 to 50% more effective and 40 to 50% more productive. For your your grimacing, you disagree?

No, no. I I do totally agree.

I I had a a three hour boot camp I did today at a conference, and the title was Make Them Care. How can we as data managers talk more about what we are doing and get more input, more buy in, more selling our ideas better. I think that’s part of the reason why you don’t see the impact because we are not that good at selling our ideas to the broader public.

But, I just wanted to put on a bit of a positive spin here on what Joe said.

Please do, yeah.

Good idea.

Thanks. You were talking about the data centers, and we do have the same discussions in Norway as well.

Norway, by the way, 100% renewables, right, when it comes to energy, so it’s all water power.

And, I’ve talked to one data center on my podcast a couple of years back, and they had they used the cooling water from the data center for a lobster farm to heat up a a lobster tank.

Oh, there you go.

I was like, fantastic. Yeah. She can get something positive out of it. Cute, isn’t it?

Who doesn’t who doesn’t like a lobster?

Yeah. Exactly.

Well, alright. Let’s let’s bring it back around in our last couple of minutes here.

Let… We’ve been talking a lot about hypotheticals.

Let let’s let’s make it a little bit more concrete. I I know this is still a bit of a thought exercise and it it falls into the land of a hypothetical here. But but Joe, let’s let’s imagine you’re you’re now the CIO of Acme Incorporated, and I’ve given you a magic wand.

And you can go like that, and and you can change one or two things in the data and analytics function overnight to make it more effective, more AI ready, better in whatever way you see fit. What what are the one or two changes as as overlord that you would make?

I’ll definitely zoom out from the data analytics function more to the broader organization and understand, you know, what can we do to impact both data and AI literacy?

You know, I I I think that as we move towards a world of, you know, self-service AI BI now, The data literacy piece is is going to be more important than ever. I’m saying with AI literacy. And so those are things I would do. I’d be… I’d I’d I’d definitely not operate in the old way that we’ve been doing things where we’re we’re just sort of… I describe this in my book as sort of the difference between, you know, there’s enterprise land and there’s product land. Enterprise land is the internal facing, you know, data team that that works with IT.

I think that that mode of operation is definitely facing some question marks about its future. Right? It needs to evolve. Product land is where data serves the end user.

Right? And so this is where, you know, you you you work from the customer backwards, data supports this. I think that’s the future. And and your end customers really are going to be both those in the enterprise and and obviously your end paying customers.

And so data hopefully supports all this as we, you know, basically from data centric companies, this could…

I think the companies that don’t do this, I I think that they’re just going to have a really hard time figuring things out. If you…

If you’re used to the old ways of working, I I would say broaden your scope and understand that the the stakeholders that you’re serving literally is everybody who has an AI chat interface or agents consuming the data that you’re stewarding. And so that’s that’s how I would wave the magic wand. I would definitely be an advocate for educating and empowering, you know, the company in every respect and as well as obviously making sure the data is well governed and modeled and and works and all those things. So I think, you know, making sure we have the the data, the infrastructure to support both our human and machine compatriots is how I would do it.

What would you do if you were sorry.

What was that? Welcome.

I I I I love the the second part, which is you were kinda suggesting that, you know, a a a a rampant focus on products that solve customer problems. And then that’s the kind of what you were you were hinting at. I I love that, but at the same time, I have done a complete one eighty on data products. I I used to be totally negatively on on data products because I thought it was only internally facing. And you could make a case that if you’re focused only on scalability, then you are internally facing.

But I’ve come one eighty on data products writ large because when agents are your customer, when machines are your customer, you you need a more traditional data product. So… But… Yeah. Winfred, what would you do? I’ve given you the magic wand now. You you can go and and and tap the magic wand and change one or two things in a data function, or in any function, or anything.

What would you be How doing often can I?

Once or twice. Can even give it a couple of taps, yeah.

Let’s do it, let’s do it twice. Okay. First thing, and this is something we talked about already, I would waive it for a bit of a mindset change.

Love it.

I’m thinking technical, organizational, human considerations together, but don’t split them up.

That’s the first. And the other one is, and that’s governance specific, right?

I would love to have data governance as part of corporate governance of how we govern organizations as a whole. This is something I’ve seen over and over again that we are very framework based on how we do things. Do data governance according to a framework, and now we have a new framework for AI governance, and maybe we have an old framework for our IT governance, and then we have a framework for our risk governance, and it’s not connected. And there are holes and gaps between it, so I would love to see it all connected to the actual purpose that the organization wants to achieve.

That would be fine.

Using Winfrey’s new framework.

Oh, the irony. It’s interesting. What you just said is, every time I’ve looked at companies that are actually doing governance the best, it’s not a separate thing.

It’s just kind of woven into how the business operates. And you… Think you can make a compelling case to say the thing started to go sideways when we when we moved governance into its own thing, gave it its own own label and its own organization and its own mandate that was independent of the business. So then I think you’re on to something. Yeah. Alright. We’ve only got a couple minutes left. Joe, when’s your book coming out?

Very soon.

And so, it… I’m I’m…

I expect sometime in September it’ll be out. Winfried’s…

Well, you didn’t know this, maybe. So, Winfried and I, I I’ve been his editor for his book. Excellent. And I’m actually working on his book right now because there’s a new publishing company coming out called Practical Data Media.

And so, by the time this airs, my book will have been out. I believe Winfried’s hopefully be launched as well. And that’ll count as two books making it an official company that’s not just a self published company.

But, yeah. And there’s many others on the way, actually. So, yeah.

Who else are you working with?

Can’t say.

TBD. Find out. Okay. Yep. I like to keep things very mysterious. But anyway, Winfried’s book is awesome. I’ve I’ve had the pleasure of spending a few years with it and Winfried.

And I I would say that it’s it’s going to be one of the best books in in the data industry.

Fantastic.

And in advance of the release of both Data Governance in the Wild and Mixed Model Arts, you can check Winfried and Joe out on Substack. Joe, you’ve been you’ve been dribbling some some nuggets on on Substack. Winfred, I think you’ve been doing the same on your on your your respective Substacks. And of course, there’s always our old friend, LinkedIn.

LinkedIn, great place. Yeah. There there you go. Alright.

If you’re still with us, thank you for tuning into another episode of the CDO matters podcast, Winfrey. Joe, thanks, guys. Joe, thank you.

Thank you.

Alright. Please find time to like and subscribe, comment, do whatever, provide some feedback. I read all comments and welcome all feedback. With that, I will see you on another episode of the CDO Matters podcast sometime very soon. Thanks, guys. Bye for now.

Thank you. See you, Malcolm.