AI Issues and Concerns – Techstrong AI Podcast – EP3

Hello and welcome to the Techstrong AI podcast. This is week three of our brand new podcast addressing all things artificial intelligence. I'm Amanda Ani and I'm here today with Mike Vard.

And together we're gonna talk about some current issues and concerns regarding ai. How are you doing, Mike? I'm doing great.

Uh, probably better than a lot of the folks trying to make sense of AI these days, but, uh, we will endeavor to do what we can to help. Yeah, absolutely. So let's start with large language models.

There is a plethora of large language models across the market for all sorts of things. I think Hugging Face has 300,000 models that they have on file. And so these companies, as they're trying to decide which large language models to use, they're really finding that they need more specific models for specific tasks and not these big generalized models.

So what are your thoughts on this? I think Bill Orfield has a piece up on tech strong data AI that really nails it well because he dived into what all the challenges really are. And it's a subject near and dear to our hearts.

'cause we had a three day hackathon here where he invited some of the folks from the DevOps Luminar is how we say. Um, and they spent three days sort of through all these different issues as well. And we encountered everything that's in this article.

Um, the challenges are fundamentally, do I wanna build my own LLM in the first place? Do I wanna customize an existing foundational LLM or do I not need to do that at all? And I can use techniques like, um, RAG to expose my data to an LLM and then hopefully it will come back with some interesting insights and compelling, uh, capabilities.

So I may not need to build an LLM at all. And then other folks are like, Hey, I just wanna use this stuff, so maybe I'll wait for some application provider to do that and I won't need to do it myself at all. I think a lot of folks are sitting around scratching their heads about how to operationalize all this stuff and my suspicion is the right answer is gonna be all of the aboves, but perhaps not in equal measures.

I don't know. What do you see happening? Yes, well, I'm seeing a lot of business leaders really scrutinize the different options as they all try to harness the technology in some form or another.

And it really does come down to more specific, um, uh, you know, models per the vendor and their needs that are maybe a little bit more to the task rather than these large ones. Mm-Hmm. And I think it's gonna take a while.

I think we, if I had to characterize last year, I would say it was the year where irrational AI exuberance. 999% of the time. Um, I think a lot of organizations are running into challenges with how to marshal all the data that they need to train in something because their data management processes are not exactly up to snuff.

So this might be the year where we quote unquote operationalize ai, but it may not see the actual benefits of all this till next year or even beyond. I, I wonder if people are starting to get the sense that, hey, this is a journey. It's not something that we turn on when a light switch.

What do you say? Absolutely. Well, I know that a lot of executives realize that the return on investment's gonna be a few years out before they really see it, but most of them do tend to believe in the technology and that it is going to improve processes and, um, being able to manage the data is gonna be a lot easier.

But they know that also it's a long process to implement and that, um, it's, it's gonna be a few years before they really see the outcome. And also, you know, it's great to have all these ideas. There's a lot of ideas and um, big visionary dreams for how to use it, but only about 30% I think it is, or 35% is really gonna stick when it comes down to it.

So being really strategic when it, when it comes to implementation and these projects is important. I'll tell you who is doing a happy dance? It's all those CIOs out there for a while.

They've been feeling maybe that, uh, they didn't get the level of appreciation that they should be getting given them how dependent businesses are on it. But a lot of business execs thought they could make those decisions on their own. I think AI brings the CIO and the senior IT leaders back to the center of the room because none of this is gonna be possible without them and they're the ones in charge of this.

So, um, there's a story up on Tech strong AI that reflects a survey that, uh, some of these folks who kind enough to participate in and it seems to indicate that they're, you know, rather excited. There are of course things to be concerned about, but overall the the general sense of things is, hey, it is back in town. Yeah, absolutely.

And it's important because there's gonna be a lot of training required. You know, there's a lot of people that need to catch up and skill up, um, to, you know, maintain this technology and the human element is still definitely important. And these are tools that are gonna help augment processes but not erase, you know, that human element.

So everyone working together is critical. And the other issue is just, um, to break down any silos that are, um, stopping the process flows. Alright, cool.

Of course, you know, I keep saying the same thing. Proceed caution, and I call that out because when you see, and there's a story we have up on the site talking about how Microsoft, Google, and Apple and others are joining together for an AI safety consortium. Anything that Microsoft, Google and Apple Act to actually get together on means that, you know, danger will Robinson.

'cause it's not exactly like these guys love each other. So if they're spending time trying to figure out how to make something safe together, it would indicate it might be pretty dangerous. So I think we need to carefully watch what's going on in these, uh, consortium conversations.

I'm hoping that they'll define best practices for using this stuff in a way that's responsible, but, um, we also gotta realize there's a lot of folks out there and who may not use it responsibly at all with malicious intent. And we need to actually understand how they're gonna use that and maybe depend ourselves against that. So are we overlooking the safety issues here?

What do you think? Well, I think that's a, a key concern of the consortium and the regulations that are underway is the safety issues. But I would say that while these big companies, their major players, and it would be good to listen to what they have to say from their experience, but I would hope that we're bringing in a lot of vendor neutral, um, players into this as well.

So these are not all defined by just a few big companies. Well, that is the concern is that AI is just gonna drive everything to the big five or six platform providers and, um, and the rest of us will become wholly dependent upon them. Uh, that's probably, uh, not a great model for the future.

So I think one of the reasons that we're gonna see folks investing in LLMs the way we were talking about at the top of the show is, um, they, we can't have that level of dependency. Folks are gonna go, Hey, I don't want my data to be used to train AI models that are gonna benefit Microsoft and Google and the like without them actually, um, sharing any meaningful revenue back our way. Yeah, absolutely.

And another big concern there, you know, with the safety issue too, is you have this big sprawl of all this tools and technology and AI technology and all these LLMs that are all having to communicate with each other in between companies and, you know, is all that data going back and forth safe? So how, how do we make sure that it's safe across all these different platforms? I wonder to what degree regulators clearly will wanna step in here, but I don't know if they know enough about the technology to actually, um, implement something that is relevant.

Maybe the folks in the EU are a little more prepared than others, but I gotta say a lot of this stuff is still emerging technology and some of it is flat out rocket science and other parts of it are just simple math. But, um, the folks running the regulations, I don't know, maybe they're gonna be dependent upon these AI safety consortiums to essentially give them the script that will become the regulation in the future. And that's not uncommon in other industries.

But I feel like given what's at risk here, to your earlier point, we need, um, some independent third party that can assess this. Yeah, definitely. And bringing in as many key players with the knowledge necessary, that's important to all work together and collaborate to come up with some good solutions.

Okay. And so next, Mike, I believe you recently had an interview with an executive from Cisco about how to utilize AI for a better remote work experience. Can you explain a little bit more about that?

Yeah, this is a fascinating conversation 'cause Jit Patel, who's a executive VP and general manager for collaboration and security business unit at Cisco was positing the whole idea that we're having this ongoing debate about whether or not people need to come back into the office to be more productive. And you're seeing a lot of edicts that are like that. And his point of view was that's a little heavy handed and it's okay to bring folks back into the office for meetings from time to time, but at this point the technology exists that we should be able to, uh, a, let people work remotely more often.

But b he was saying, we can use AI to make up for the gaps that occur when people are more distributed because AI will going forward, capture the sentiment of a meeting, it will capture the notes, it will, uh, give people a sense of what to prioritize and there won't be this as big a communication gap, which is allegedly the reason that folks are wanting to bring their people back into the office. And one of the things he pointed out is there's just a whole generation of managers out there who are, um, not trained to manage people remotely. So they like to go back to what they know and his point of view is, well, they need to evolve.

But the other thing that was interesting, he says this is a disservice to the national economy. And his point was ultimately it means that if you're pulling people back into an office, you're gonna have to do that in places like New York and San Francisco and Chicago, and it forces people who live in more, uh, rural communities that otherwise could participate to move to those places. Um, and it's bad for the general economy because those folks could be making, um, a really good salary that would go a lot longer in a particular town in, I don't know, pick Texas, Nebraska, wherever your favorite spot is.

And we shouldn't force people to always have to be in the office. And if on occasion if we need to fly him into meetings, we can, but, um, he's bending longer term, the workforce will become more distributed because of ai. What do you think?

Yeah, I believe so. I mean, we're in a, a day and time now where we should be able to utilize this technology to, to stay where we're most comfortable because statistics show and prove that people working where they're most comfortable, um, in the environment that they prefer is a better outcome of work. Um, so they're producing better work and there's ways to utilize the technology to feel, uh, more, um, like have a relationship with people in a better way and to feel more connected and to feel less distance between where all the employees work.

And we should definitely harness that technology that's available. Yeah. And most white collar jobs are not nine to five.

Um, you can perform them over the course of a 12 hour day if you're so inclined and kind of wrap your personal life around that. But I think people have gotten a lot better at thinking about the outcome rather than, uh, whether or not I showed up from the office at nine 15 on the dot and then I stayed there to precisely 5 0 5 and then I got in my car and drove home for an hour to two hours that are probably a waste of time every day. And if you add it all up, uh, just commuting time alone, what is that, you know, two hours a day.

So that's, um, 10 hours a week that we could probably give back to all kinds of folks, which frankly half of them will spend most of that time working anyway. So, um, it just feels like we're being shortsighted. Yeah, absolutely.

It's a lot of wasted time to, that could be given back to the employees so they feel refreshed and they feel happier in their jobs. And, and then like you said, that's, that's more time that they can put to more work if they want to. But, you know, people work in different times, uh, better than others.

So some people are morning people, some people are evening people, and so long as the work is done, there's a lot of more, um, deadline driven tasks that can be done anytime. So let, let's be more innovative and advance Elsewhere on tech strong AI this past week. Um, there's a story up about how Sam Altman is talking about $7 trillion to build AI chips.

Um, we're not exactly clear what kind of chips we're talking about here, but it does seem there is an issue with, well, you can't find any GPUs out there to train the AI model, so we clearly need a different approach. Some people say the GPUs were designed for graphics and are fundamentally inefficient both from, uh, uh, cost and from a carbon emissions perspective. So I think there's probably room here for innovation, but this won't be the only game in town.

But if he needs 7 trillion, let's just add up all the other potential chip companies out there. Maybe we're looking at, uh, $50 trillion investment to create processors for ai. Is what's the return on that gonna look like, do you think?

Yeah, that is a good question because that is certainly an astronomical number that made headlines the last few days. And I know that, um, whereas he was also working with some other players like Nvidia, but I guess, um, you know, if they're gonna go the way of making their own chips and having the factories, which is also gonna be costly to run to make the chips, then I assume they've crunched the numbers and feel that it's, um, going to be a huge return on investment because I couldn't see making that decision without feeling that it's going to be a massive return on investment. So I guess even more than $7 trillion.

Yeah, I mean, when you think about how crazy that is, it's probably larger than the GDP of half the countries on the planet, right? Yes. Yeah, it's, it's interesting.

We'll see how it plays out for sure. And there's probably a lot of folks wondering if we have that kind of money to spend on AI chips, why aren't we doing something else with it? But who knows?

Um, True. We shall see. Um, that kind of brings us to our final topic of the week, but it's this, uh, AI climate change initiative.

There's been a proposal from, uh, I believe a couple of the democratic senators in the US Senate, and basically they're calling for the EPA to, uh, more aggressively monitor how much, uh, carbon might be emitted through all these, these AI activities. And they want NIST to, uh, voluntarily start collecting and reporting data from folks who will contribute that. I don't think anybody's gonna voluntarily do much anything, so there probably will be a need for a law or regulation, but how much of this is just flat out wishful thinking because this is a congress that, you know, I think passed a grand total of 27 laws in the last four years.

So, um, probability assessment that, you know, a bill proposed by Democrats is gonna get any traction. What do you think? Well, it's gonna be difficult for sure, but I would hope that they would come together and see the actual facts, which is, there's a lot of energy consumption when it comes to this ai and there's a lot of technol technology waste to be considered when, um, old technology is thrown out.

So the carbon footprint is a big concern, and I would hope that both parties would, you know, set aside their, their party lines and, and, and realize that some regulations need to surround this, the earth and environment factor of ai. We may have to look to the EU once again for some, uh, laws and regulations here that companies will follow, uh, simply because they want to do business in Europe, but of course not everybody will. So, um, it's gonna be a, a challenging time.

I think fundamentally we need some sort of massive breakthrough both, uh, in terms of how we consume it, I mean, realizing that it being only one small part of the overall carbon admissions problem, um, as what somebody pointed out to me not too long ago, cows generate more methane than it people create issues, but it's, uh, just part of the whole thing. Um, I cannot help but wonder though, um, maybe this will drive some sort of, uh, advances in capabilities to capture carbon in ways that, uh, won't let it up into the environment in the first place. So we gotta figure out how to be more efficient across the board.

I don't think we need to assume that the current state of carbon emissions is static. I think maybe we can hopefully find some ways to innovate our way out of this issue because, um, even if at this point we started ratcheting back how much energy we're kicking up into the atmosphere, it might, it's pretty much too late already. So we gotta find another way out of this thing.

What do you think? Yeah, absolutely. And, and just as you said, AI can both be harmful and there's risk, but it can also be positive and useful.

So taking AI and using it to, um, change the outcome of, uh, like somehow trapping the energy or removing it or reducing the footprint somehow using the technology, I'm hoping that there's a couple of smart kids in college or who knows, maybe they're in high school or even grade school, who will figure all this out for us in time. Uh, in the meantime, historically innovation's done better for us than it's harmed us. And I'm hoping that with AI that will continue to be the case.

Not a big fan of the, uh, doom and gloom scenarios that are out there, but, um, like I said, proceed with caution. Most definitely. Well, I wanna thank our audience for being here today and stay tuned next week when we bring you more of the latest issues and topics surrounding artificial intelligence.

Thanks, Mike. Thank you all. See you.