22. AI is Not a Fad – Tech Field Day Podcast

The current hype about building massive generative AI models with massive hardware investments is just one aspect of ai. This episode of the Tech Field Day podcast responds to last week's episode and features Frederick Van Herrin, Karen Lopez, Marian Newsom, and myself, talking about a different perspective of the larger world of ai. Welcome to the Tech Field Day podcast, where we bring together a group of IT experts to discuss a single idea about key concepts in our industry.

This podcast features a variety of perspectives from members of the tech field, a delegate community. We usually record it in association with one of our events, but this episode is recorded sort of in between events. So we've just brought together a group of interested delegates to talk a little bit about AI on this episode.

We're kind of continuing last week's episode. If you listened in, then you heard me say that ai, as we know it, is just a fad. And I talked about the, uh, fact that we are building bigger and badder AI supercomputers and spending billions of dollars on hardware for what exactly.

Well, this episode, as I said, is a companion In this episode, our premise is that AI is not a fad. It's just a little different than we've all been talking about. Before we get to the discussion though, let's meet who's on the panel today.

I'm Frederick Van Hern, the founder of hyen, a consulting and services company active in the HBC and AI market. And you can find me on LinkedIn as Frederick z Aaron. I'm Karen Lopez.

I'm data check on almost all the social media on LinkedIn. I'm like slash in slash Karen Lopez. And I like to talk about AI in terms of data because that's who I am.

Hi, I'm Marian Newsom and I'm the founder of Ethical Tech Matters, and I am a ethical technologist. Uh, you can find me on LinkedIn at Marian Newsom. And I'm Steven Foskett.

I'm the organizer of the Tech Field Day events. Uh, you'll hear me on this podcast on the gestalt. It rundown on, uh, tech Strong Gang pretty much weekly.

And of course, I'll be leading our AI Field Day event and our AI data infrastructure Field day event, uh, coming up in the next month or so. So, as I said in our last episode, we talked about the fact that, um, we've been sort of led to believe that AI is all about large language models. It's all about generative ai.

It's all about building bigger batter, billion parameter models on these supercomputers that cost billions of dollars. But the truth is that AI is actually not that at all. That is ai, but it's a lot more than that.

And actually one person that knows a lot more about this subject than pretty much anyone and has been doing AI for a very long time is Frederick Van Herrin, who co-hosted the utilizing tech podcast focused on AI with me for a long time. Frederick, I want you to kick kick this off. Tell us about AI beyond generative ai.

Right? I I think, I think when we look at, at ai, it's important to understand there are really two components. There is a, a training component.

This is the place where all the data is where probably a lot of GPUs are needed in order to build a model. And then you have the flip side, which is using those models, which we typically call inference or production or hosting. Those two components are, are really important to understand because when you look at today what's going on with a large language models or what people call large language models, they're actually produced by a handful of organizations.

So all the complexity, all of the GPUs is typically sitting in a small group of organizations. Us, the consumers, we sit on the other side and we're consuming all of that technology. And our requirements for this are not that high.

And so where are we today? I would say we're building up use cases, finding new ways, building new software infrastructures to build it all together and allowing all of us as consumers and us as enterprises to build applications around all of that complexity that is really hidden from us. So I would say that one of the reasons that AI is not a fad is just what you said, Frederick, is around the consumers.

Uh, the companies have pushed this to the point where, um, there's just a demand for it. They don't know what they're asking for. Uh, but everywhere you look, there's ai, ai, ai, and so the consumer is actually looking for, uh, AI and to make their lives simpler.

And so that's why I'm thinking AI is not a fad. I agree with you guys, but I also wanted to point out, I think a lot of the way, like I'm guilty of it as well. We've talked about this, I use the word AI now more as a giant bucket name.

And I know technically when I say AI and I'm really talking about a large language model, I'm cheating. But it's just easier to talk to generalist crowds about it without getting into all the details and the acronyms. I think one of the reasons people think it's a fad is because of what you talked about Frederick, is that they are users of ai, and that's what's hiding the complexity.

So when I use generative AI to generate a short script or to make a funny image or to tell me three great jokes, I'm kind of using it the same way I use email and that's hiding When I use email, even as old as that technology is, I know it's hiding a bunch of complexity that I don't even know as exists, even though I'm admin on some email servers, I don't know what's going on. And I think we're making that same sort of, we as the whole, as the world are confusing uses of AI focused applications than actually building or training or doing all the other things that happens with people working on ai. Yeah, it's interesting because to the world, um, you know, if you look at mass mainstream media, if you look at sort of the mass consciousness and the mass understanding of AI out there in the broader world beyond the technology space, I think that most people think that they are seeing ai.

Most people think chat GPT is a, you know, equals AI and chat GPT equals sometimes ludicrously wrong, failed, untrustworthy, et cetera, et cetera. And I think the same thing happens like, you know, with AI images, um, and especially now that we're seeing this stuff being embedded pretty much everywhere. I mean, uh, everybody's going nuts because, uh, Twitter, uh, or I'm sorry, X now has, uh, image generation built in.

And so you can have it generate like a ludicrous image of like anything and, and share that on the platform. And, and there's this, this incredible backlash, a deserved backlash to some of these applications, these kind of half-baked applications of generative ai. And I think that people are thinking, oh man, ai, it just isn't what they're, what it's cracked up to be.

And then they hear about all this investment, all this money that's getting poured into it, all the energy that's required to run this, because I think people think that it's all running on those GPU clusters that are being built and like take up as much energy as Indiana. And you know, I, I think that people hear all this stuff, and I think that that's the real problem, is that you've got all these, uh, you know, idiots in the media out there saying AI is just a fad. Yeah, I know that's the title of the last podcast.

Uh, you've got all these people like me saying AI is just a fad. And then they hear that, and then they're like, oh, AI is just a fad. What a joke.

Nah, but it's not, right. I mean, there is there, there's useful stuff happening here, Right? And I think it's so useful that the market, meaning the consumers are asking for more and more, which is kind of problematic because the, the technology behind large language models is kind of maturing as we go, and it's maturing a lot faster.

I mean, if we compare inva the speed of innovation today of technology compared to a couple of years ago, it is going a lot faster today than it used to be. And that's a concern because people expect a lot more, you know, look at self-driving cars, right? I mean, when, when AI came out, people were expecting self-driving cars in, in, in a few years, while in reality it might never happen.

Uh, and just based on expectations. So I do think that, uh, there will be more and more use cases, certainly when new innovation comes out, where, where people feel more closely and can consume it on a daily basis. And by the way, today, a lot of people are already consuming AI without really knowing it, right?

So I would say that the expectation and the speed of innovation are two things that are kind of helping. Um, the, getting the message out that AI is not a fad, but there will always be, of course, people that want more and where it doesn't go fast enough. And so I also think when you talk about, uh, people are using it, uh, kind of brings in that trust factor, right?

So we've conditioned people to say, oh, um, uh, to trust the cloud, oh, my data's in the cloud. Oh, my phone, everything on my phone is in the cloud. And so that trust of AI is just, it kind of is piggybacking on that, but people are seeing that I can't trust it.

It has hallucinations, it gives me incorrect data. Uh, but as companies continue to integrate it into technologies, our hardware solutions, software solutions uses in the medical industry, um, I think we're, it's just not a fad. It's gonna be around advancements will continue.

Uh, and so I know Karen, we agree not to use the word responsible, uh, but we have to start looking at responsible ai, uh, and marketing it responsibly also. So I'd like to hear from Frederick because it's my understanding that the non generative ais generally can't halluc, like they're not designed 'cause they're not designed to generate things. They generally don't hallucinate the way we talk about it.

They can make mistakes and they can do things wrong, but is that true or not? That can non generative AI stuff hallucinate? Well, so let's start, how does AI really works, right?

So the neural networks, they don't come up with things on their own. So it's, it's all based on historical data. So any AI related engine who uses historical data, there is the possibility that the output, well, it is guaranteed that the output is heavily affected by what you provide.

Um, in, in, I would say that hallucination in the sense that, in the definition that it could give you the wrong answer based on historical data. Yes, of course any AI engine is actually could be affected by that. But I do think there is a difference between the traditional AI and the generative ai.

I feel like with generative ai, people trusted more out of the gate than a, a more traditional, uh, generative AI application. And it's, it's kind of a, a psychological thing, but just to answer your question, any AI engine can produce a bad result, not bad results, the wrong results based on historical data. I just think that with generated ai, because the impact can be a lot bigger than with, with, with traditional ai.

But, you know, banks use traditional ai. I mean, hopefully no mistakes happen there, but you know, you could also have issues with, with your bank account when AI is involved. Yeah.

And as a data person, like I totally get this right? Even regular reporting systems can make errors and make up stuff if the data going in is skewed, biased, wrong, missing all the other things. My whole thing is, is I don't think I'd ever generate a report that had good data that said, glue your pizza toppings on to keep them from sliding off.

Um, that's what I'm talk, when I say hallucination, I really mean the chaotic, weird, wrong answers that certainly, unless the data's been poisoned, it didn't learn it from any data. Well, I think in the case of the pizza glue thing, they did learn it from the data, but, uh, because people are goofy. But I think that, um, you know what, what we're getting at here, and this to me is the thing that makes, that's kind of infuriating to me is that on the one hand chat, GPT and stable diffusion and these generative AI algorithms, they're so good in, in that they're so convincing, they're so compelling.

They're so, they generate such incredible truthy stuff that we fall in love with them. And at the, on the other hand, they're so bad because they're just totally not doing what we think they're doing. I think that's the thing that kills me.

It reminds me, I mean, it seems like ancient history, but remember the, the, the guy who got kind of in trouble in the news because he decided that Google had developed, uh, an artificial general intelligence and, uh, you know, he had to, you know, run away and quit his job. I don't blame him because essentially he was interacting with, um, um, a generative AI model that was compelling enough to fool you, to fool me, to fool any of us into thinking that it was something. It's not.

But it's not that. And I think that that's the thing that makes this whole thing. So, you know, the whole world of generative AI so troubling is that it looks and sounds like it, you know, it like it's, it's right.

Even though it's not. It's, and I, I likened it a couple of podcasts ago to, um, that smart Alec kid in third grade who like knew everything and you'd be like, you know, oh, I don't wanna say that it was any of us. It probably was all of us.

But you know, the, the smart alec kid who, who, who just would be like, yeah, I know all of this. Let me just tell you about all of this, right? And, and everybody was like, well, I'm not gonna question, I don't know the, I don't know what kind of like whales and, you know, dinosaurs and stuff, so sure.

I guess he's right, right? Well, that's, that's chat GPT, right? It, it gives us these con these, um, confident answers in a compelling way and who are we to question it, right?

But yet That's not it, it's not as good as it thinks. It's Right. I mean, I, I think at first we all, we have to understand that AI is a tool and what kind of a tool it is.

Well, it's a tool that augments our knowledge and our decision making. And augmenting doesn't mean replace the moment you assume that AI is replacement for your decision making. I'm pretty sure everybody at some point point will have an issue with an AI engine, and it's the bad way to look at it.

And so it's very important, it's a tool and it's augmentation. And augmentation is helpful if you already have some knowledge, right? So Karen used the example where she already had some knowledge, so she understood that the information given to her by the generative AI engine was wrong.

If she didn't have base knowledge about it, it might have sounded weird, but maybe not considered wrong. And, and I think that's where we're going, but it's, it's, I think AI is, is, um, is a technology that continuously learned. So if it's doing things that today, hopefully we, the human will help and, and fix the tool so that your augmentation will be less weird and more on target.

I think that, um, a lot of people don't, uh, general probably, and it, and data professionals don't even know how AI has been used like even 10 or 15 years ago. So like in manufacturing, detecting anomalies, in retail, analyzing video, and a lot of those are mostly, now in my understanding, of course, we get a lot of AI washing about all this, but it's my understanding a lot of that, those usage usages that have been around, been tested, been tweaked, those operate mostly unattended. So we're not give, make, allowing them to make important, important decisions.

But the uses of AI have been around a long time. We just didn't hear about them because the average person didn't use those, didn't see the results, didn't hear about the tests, or they heard about it because a part wasn't manufactured with intolerance and cars were crashing or things were falling out of the sky. But even then, no one said, oh, it was an AI thing that went wrong.

They were just told it's a computer thing that went wrong. And I think we are missing, like, Marion, you brought up trust. I, I mean, it's not that I overly trust AI or don't trust it, it's that I want people who design these systems and put the safeguards around them are competent enough, not rushed enough that I can trust it.

And therefore, because of all that, I don't think it's a fad because ai, well, depending on who you listen to, AI was created in the fifties or the 18 hundreds, like whatever you wanna say. And that, you know, that whole, like, again, we come back to where we started that most people now think AI is just chat GPT. It's interesting, isn't it, because, um, you know, Frederick and I were talking about this when we started.

So we're, we've done now, uh, seven seasons of our, uh, utilizing tech podcast. And when we started it, it was before chat, GPTI mean, it wasn't really a chat, GPT did exist, but it was like, like the earliest version, it wasn't very impressive and everything. So generative AI kind of wasn't a thing.

And so if you listen to like episode or season one, we talk about the, the use cases that Karen is describing. We talk about finding needles in haystacks, and we find, you know, we talk about security and we talk about exactly what you're saying. Like, you know, oh, like, like this camera's gonna see if this tomato is ripe or if this tomato is rotten, and things like that.

And, you know, those use cases, they're still with us. In fact, those are really compelling use cases. You know, we were hearing about that at AI field day last time.

Uh, we had the Nature Fresh foods guy, and they're talking about how they have cameras in their greenhouses, um, looking at tomatoes, literally that's not gen ai, that's not, you know, stable diffusion. That's, that's, you know, regular AI doing regular stuff, improving crop yields, and frankly, um, you know, to Marian's point, that's responsible, that's responsible ai, it's doing good stuff, and it's not burning down the rainforest with a GPU supercomputer to pretend that it's a human. It's saying, is this tomato ripe?

Yes. No. Right.

And so Karen, I'd like to go back to one of the things you mentioned about it. How do you think we go about educating them to understand, um, the different pieces and parts that AI actually plays in the enterprise? Um, boy, that's a tough one because that's a good question for any topic.

Like, I want it people to understand why garbage data leads to garbage outputs no matter how great your ETL per is. And it's the same thing with ai. We're learning, again, that transactional data has issues and challenges, and we learned it with data warehousing that we had to go fix data to get it to work in our data warehouse.

But because we're going back to the source systems, again, we're learning it all over. And a lot of, uh, coworkers I have that are new to the field, haven't even heard that we have to fix data or we have technical debt with our data, or we never updated our systems to know about new things with data. And I don't mean quantum mechanic, quantum physics or any of that stuff.

I mean that now we have more than two genders, but our system still only records two. So what are we gonna do about that in the ai, because we're telling 'em there's only two. So all that complexity, but I think the biggest thing is if an organization is working on both generative and non generative ai, just they need to let people know, like, here's what you can do.

Here's what you can't do. I saw one stat that said that probably 20% of your IT workforce is using generative AI in your jobs, even if you have a policy against them doing it. And I think that's, I think it's probably a lot higher, and I think we won't know because if they're producing their work outside the system and bringing it in or typing it in from that, it's not just a mouse jiggler we're looking for.

I also think that the challenge from an education standpoint and a validation standpoint, uh, when you see people coming out of college, they understand that data is important, but they teach more about the software and the hardware, right? And so they get in the industry and they start building a, a framework where they might already, there might not be any kind of data, right? And the challenge there is is, you know, everybody knows the more data, the better and so on and so on and so on.

So there is a risk that people go for the low hanging fruit where they buy some data or they use some data that doesn't fit the, the criteria. We all, we all understand. I mean, I, I think what I see also happening in the industry is that, you know, people that are using AI and generating AI models being a traditional or generative generative AI are also starting to work on other AI models to validate the existing ai.

So in other words, using AI to validate ai, I mean, today you can already read articles about a, an AI that can figure out if AI wrote a blog or an article, right? So I think, you know, maybe that's, that's kind of a another risk. But I think it's, it's a great idea to, to kind of use technology to validate technology.

But I could see generative AI helping document my system again as a paired thing, not instead of me, you know, while documentation typically belongs to the organization and isn't mine, if I had a team that had to produce, I don't mean a whole bunch of texts, but documentation about their work, I'd want them to use generative ai. If we, if it was producing good enough stuff that we could produce technically accurate stuff as well as if it freed up their time, if it takes more time, then it takes more time. If it's inaccurate and they're not checking it, then I wouldn't want that.

But there are certain, there, like, there's all this stuff about copyright and what were the models trained on? I kind of like this idea of pointing AI to a SharePoint document library and building an internal only list that allows me to better search and find documents I like. Those are the stories that make me really excited about generative ai.

And I probably don't care if it made up a couple of things because the cost benefit and risk turns out better. Because if it says this document is about data quality issues and it's really about what everyone ordered for the office birthday party, what have I lost? Well, it's interesting, you know, Karen, because you know, you just brought up basically two tech field day presentations in my mind.

So, um, you know, the first one, Marian, uh, you know, we heard, uh, BMC talk about how they're using AI in the most, I, I mean, pretty mundane en environment, but a pretty important environment, basically analyzing COBOL code on mainframes to do, uh, to figure out what that code does and to create, um, commentary on the code, uh, and figuring out what the overall mainframe system does and how the whole system works. Um, so, so we just heard about that at Tech Field Day at Share. And then Karen, um, what you described with the, the rag and the internal documentation, I mean, that was basically click answers.

And, um, remember I was blown away, so impressed when they asked click answers questions that weren't in that document set in the, in the, in the, the, the set of information that it was fed. And its answer was, I don't have that information. I am not gonna make something up.

I just don't have it. That's not here. So we, you know, we've got two companies and they're not like, these are not like, you know, venture funded Silicon Valley hot startups, right?

These are, these are like kind of practical companies doing practical stuff and, and they're doing useful things with ai. That's not a fad. And I think what I liked the most about the BMC presentation was it really did bring human interaction into the ai.

So I do think the, the hype is around, oh, AI is gonna replace everyone, but I think for trust and responsible there, that word is of the inherent, uh, ai, it really is about human inter human interaction. Uh, like Karen said, your teams sitting down with the AI for efficiency, not for replacing a human. And that's one place like the naming of the product.

So not taking the, um, full self-driving vehicle badge and slapping on something that isn't full self-driving, but copilot and Gemini, like, they're really putting that forward. And I think other names are doing that too. Someone who sits beside you or above you and helps you do your job.

It's not full self programming. Someone's gonna trademark that now. Go Steve grade, mark it.

Well, it depends on how much of an idiot they are. Um, yes, this is full self programming version one beta, uh, now, okay, so I wanna go around the corn, around the horn here. Let's, let's, let's have y'all around the table, the virtual table.

Um, I'm gonna propose the premise again, and I want your sort of summary reaction. So no, I don't think AI is just a fad. I think ai, uh, is part of it.

It's here to stay, uh, doesn't have issues. Absolutely. Are they addressable?

Uh, most definitely. But I think what we talked about here, for me, uh, the future of AI is really the human interaction. It's not replacing anyone.

It's actually making us more efficient, giving us a, a better handle on complex issues that we may not be aware of in our enterprises. Boy, I agree with all of that. Thank you.

It'll make mine shorter. Um, I don't think AI is a fad. I think it is chock full of fatty things, and I said that with a d not a T and that we all have to be aware and ask more questions when like, you know, early on we found everyone saying, oh, it does ai.

And when you asked 'em about it, they're like, oh, I don't know anything about it. It's just on our spec sheet, our data sheet that we say it does ai and no one in our company knows what AI is being used. So all of that.

So I think we just have to be careful. It's kind of like social media ads. They all look great, but be careful bringing out your credit card to do that.

But I think AI's almost as old as I am, so I want it to be around a lot longer. Yeah. And as somebody who has been working on AI for a very long time, um, I'm, I'm just surprised and positively surprised on how many use cases there are and, and the complexity of the use cases that I've seen so far.

Uh, the challenges are still the same, right? So Steven talked a little bit about mainframe, so some of the problems are, are just timeless, and I think that's something we need to keep track of. Uh, one of the big challenges with ai, it's going really fast.

Um, sometimes it's driven by computers, which are a lot faster than we are. Uh, but it's always important to understand that AI is a tool, it's there to help us, um, and to augment our, our capabilities of reasoning. Uh, definitely not replacing it, which means if chat PT or some other AI tool tell you something, it doesn't necessarily, uh, be completely true.

You just need to understand what to do with that information. Um, and, and I also believe that that ai, because it's the new AI, is relatively, um, recent, and that also means that now there are capabilities of hiding the complexity. And hiding complexity means it's easier for people not only to build applications, but also to consume it.

And then finally, it basically means that AI not only is, is here to stay, it's already here. There are many, many applications that we're probably already using and we're not aware that AI is in is involved, but of course, we have to make sure that we, that it stays safe and that the information that we provide to those engines, um, meets the criteria we have for it. Yeah, I, I, I think that the challenge, the biggest challenge here is that we've, we've had such a fad, such a tulip beanie baby moment here with generative AI that everyone is almost hoping for the bubble to pop, right?

I think that, I think, and I mean like the general public is out there saying, oh yeah, I tried it. It wasn't all that cracked up to be, everybody's waiting for the comeuppance. But the truth is that maybe, you know, if it wasn't for those fatty, as Karen said, uh, applications, maybe we would realize that there's a lot going on here that really does have legs.

And I think that overall, I think that's what we're gonna be seeing in the next year. I think we're gonna see a big pivot from this idea of it's all about creating the biggest, baddest generative AI model to wait a second, where can we use this? How can we use this?

How can it help us? And, and frankly, you know, as Frederick said, I mean, we, you and I on, on the, you know, the utilizing podcast, um, ev we asked many of our guests, when will we get full self-driving that can drive anywhere, anytime? And the answer from anyone who knew anything about AI was never, that's what they all said.

You'll never get full self-driving. You'll get co-pilots, you'll get partial self-driving. And it's not just about self-driving, it's about, it's about everything we've talked about.

We'll never get there because that's not what it is. But we will get lots of productive and useful applications in all sorts of places that will help people do things better. You know, I love what Frederick said about essentially a tool that augments what we can do ourselves.

That's exactly what it is. And that's exactly why in my mind, AI is not a fed. So thank you all so much for joining us.

Uh, before we go, I wanna give you a chance to, uh, say one more time, where can we find you, where can we connect with you? Where can we learn more and, uh, where can we interact? com, as well as on LinkedIn as Frederick v Herron.

And I'm actually looking forward to being a delegate at AI Field Day five in two weeks in San Francisco. Same. I'll be there too, travel gods, hopefully.

Um, but you can catch me as Data Chick almost everywhere. And I'll be speaking in Prague about data privacy and security with AI as well as Shanghai, so September for Prague and October for Shanghai. So hopefully I can see you there and You can catch me on LinkedIn as Marian Newsom and also as co-host of the Tech Aunties podcast with G Gina Rosenwald And, uh, Marian, I think we're gonna see you at Security Field Day as well in October, right?

I, fingers crossed. Excellent. Well, I would really love to see you there.

Again, thank you so much for joining us at Share, uh, for mainframe, our first ever mainframe field day. That was awesome. It was.

And, and, and eyeopening. Thank you. Well, thank you for coming.

Uh, thank you all for being involved in the Tech Field Day community, and thank you everyone for listening to this episode of the Tech Field Day podcast. If you enjoyed this discussion, please do subscribe. You'll find us on YouTube as well as your favorite podcast application.

And I have to say, a lot of people are joining. Our YouTube numbers have just been going up and up and, uh, you know, that's really phenomenal to think that people like to tune in and and and hear these things. So thank you for listening.

We would love to hear from you. Uh, maybe give us a rating or a review or just a shout out on social media. If you're listening, uh, let us know.

Um, this podcast is brought to you by the Tech Field Day Events series, uh, home of IT experts from Across the Enterprise, which is part of the Futurum Group. com/podcast. Or you can view this podcast streaming on Text Strong TV as well.

Thanks for listening and we will catch you next week.