Small Business and AI Transcript — Sept. 16, 2025
Rachel Jones/NPF (00:00):
For this next session of the NPF Local Business Journalism Fellowship, we’ll be exploring how small businesses can harness the power and the potential of artificial intelligence. We’re joined today by Professor Balaji Padmanabhan. He is the director of the Center for AI in Business and the dean’s professor of decision operations and Information Technologies at the University of Maryland’s, Robert H. Smith School of Business. And I think I’ve earned a degree just by saying all of his titles. So his current work addresses the intentional design and governance of artificial and augmented intelligence solutions that combine AI and human capabilities to generate value. Balaji is joined by Norma McCowin. Norma is an executive MBA graduate from the University of Maryland who serves as chairwoman of the Master’s in Information Systems Advisory Council. In her leadership role, Norma’s dedicated to securing valuable internships and employment opportunities for graduate students. And if I’m recalling my conversation with you, biology early on, Norma has actually developed a company that helps small businesses learn how to adapt AI strategies. So as I told them before we got started, we are so inundated with scary news about artificial intelligence and what it means that my work and preparing this session has opened up a whole new universe to me and I’m fascinated by this subject and I know you will be too. So we’re going to start with Balaji’s presentation and then we will go from there.
Balaji Padmanabhan/University of Maryland (02:06):
Thank you for doing that. Thank you. Thank you so much. So thank you. Thank you all for being here. And frankly, when they’d asked me to come to your audience, I didn’t have context, so I didn’t know who you were, what you were doing. And I just said, look, this is what I do. I’m happy to talk about it any day, right? But I learned more. And I think given the roles that many of you are playing in the different regions when it comes to business and reporting and telling the story, I think there’s a very, very important role that medias playing. And I think that itself is itself is a separate topic for another day to look at AI’s impact on media and what it’s doing to everyday life. So information and reporting is even more important given the amount of AI generated slop and other things. We are saying just separating what is real from what is not, what is important from what is not is becoming increasingly harder. And so this is where I think the role of traditional journalists and journalism become, in my view, even more important than it’s ever been. So I think, is this still semi? No. A little bit maybe. So you know what? I wanted to touch upon a few things in the first part. Talk about, there’s a little bit of echo somewhere.
(03:44):
Is it good now? Yep. Yeah. Yeah. So I wanted to talk a little bit about the capabilities of AI before we get into AI and small business because one of the things I always tell people is that it’s very, very important to understand this thing that we are working with. And it’s very important to understand what it can do and what it cannot do. And many times when you look at people talking about AI, you get a sense that it’s magic that AI is magic. It can do everything under the sun. I used to tell my students, look, if you want magic, go and watch Harry Potter. AI is not magic. It’s something very different. But in that sense, I thought magic is a very bad metaphor for AI, but I’m not so sure increasingly when you look around at the kinds of things AI can do, it turns out I think magic is actually a very good metaphor for AI.
(04:34):
Some of you may have been around when this happened, right? When Watson played Jeopardy, and this was really the, I think there are a few times in history we saw when the world saw AI in a big way, and this was one of the early such times and what happened there was spectacular. If any of you haven’t seen this, definitely look this up on YouTube. IBM Watson beating the world’s Best Jeopardy players at Jeopardy was just unbelievable at that point. The closest we are seeing to it now is happening in the last two years with Chad, GPT and Claude and so on, having the kinds of conversations that it’s having now.
So when AI works, I think magic is a good metaphor. It seems almost very magical. But the reason magic is a good metaphor is if many of you or any of you talk to magicians, they will also tell you that magic is not magic.
(05:34):
There are a bunch of very specific things they do under the hood that create the illusion of something magical. But there is a mechanism, there is a logic, there’s a reason for how they’re able to do what it is. And AI is very similar. When great AI works, it should look like magic. But under the hood, there’s a very deterministic logic that’s playing out, which is also important for all of us to understand because that’s how we understand what it can do and what it cannot do. So this is very, very broad and high level, but one of the things I wanted to point out was AI has been around for a very long time, and I started my journey in AI maybe about 30 years ago when I did my undergraduate thesis in genetic algorithms. Then I went into a PhD in machine learning and AI in business.
(06:28):
So I’ve been in this area for a while now doing more or less the same thing. I write papers, teach courses, and work with companies. But for the last 25 years, been in academia doing all of this. So if you look at the examples that we used to show for AI, many of these examples are 40 to 50 years old. AI is a chat bot. AI is doing medical diagnosis, AI is playing games. In fact, the earliest paper is 19 50, 75 years ago. You fast forward 70 years, you see the same use cases. AI is a chatbot, AI is doing medical diagnosis, AI is playing games. And why is this more or less the same? I think there are two points which are important to understand from this example. Number one, I think AI has always pursued this amorphous, vague concept of intelligence. And how do you get there?
(07:31):
If you ask people to define intelligence, you may not get the same answer from different people, but AI has always been about pursuing intelligence. Initially it was human intelligence. And so the kinds of use cases AI was going after were the kinds of use cases where humans are known to show intelligence, which is playing games, having conversations and things like this. And so that hasn’t changed over the years, but if you look under the hood between what powered the old AI and what’s powering new AI, this is night and day in a sense. The previous generation of AI was learning from here. I’ve said pre-programmed logic. What that really means is it’s learning from what we code the intelligence as, right? So think of human designers explicitly coding the intelligence. That’s what we used to build before. Now if you look at that concept, you can only get as good as our own intelligence if the best we do.
(08:33):
I mean, if all we are doing is putting our own knowledge into the system, you can get as good as what we know. The modern systems on the other hand, are powered by massive amounts of data as well as this learn by experience paradigm called reinforcement learning. Just like humans learn from experience, these algorithms are learning from experience by playing in simulated worlds. These two are game changers because potentially if there’s stuff in the data that even humans cannot fully process, these systems can process it. This is why AI is doing amazing things like things like medicine where we are generating so much data and knowledge every too much data on a daily basis, but humans don’t have the time to completely process. All of these systems are geared to sing stuff from data, and it’s not surprising that they’re starting to do even better than humans.
(09:27):
Same thing with game playing. They’re learning from experience by playing against themselves millions of times in order to get better at playing games. So I tell my students and others that, look today, if you play a game with a computer and you win, it’s only because the computer let you win. There’s no chance anybody in this room is ever going to beat a computer because this is how strong that reinforcement learning paradigm is. It’s near impossible at the levels it can get. So that’s the kind of capabilities that we are looking at here. Now, this is not super important from the point of view of what you’re doing in small business, but I want to end with capabilities four points. And if you’ll see there’s a commonality here. So what are the major things that brought us today’s AI and who was responsible for it? So what is the big developments here? So one of the things, and we forget about this, but one of the things which is very important to understand conceptually is that anything to an AI system is a number.
(10:39):
Just think about that for a second. You’re having all your conversations with Chad g, PT or whatever. All of that gets converted into a number or numbers, and the AI system has input a bunch of numbers, and from the numbers, it’s creating numbers that then get mapped towards. So this is not how we do stuff. So anytime I talk to you, you’re not converting my words into numbers and then converting it back to words, you’re understanding language natively. But AI systems, everything has to be numeric. That’s how these systems are built. So the first really biggest advancement in this space was how do you create numbers itself in an intelligent way? How do you take a word? How do you take a sentence and convert it into a number? So Google was the one that came up with one of the first methods called embeddings, and that essentially is an embedding technology called Word to W.
(11:36):
How do you convert a word to a vector? It’s about 15, 10, 15 years back. But that’s the very seminal advancement in the field where they said, you know what? We can convert words to numbers in an intelligent way, not in an obvious way. So you might say, look at a dictionary. If a word is the 18th word in a dictionary, map it to the number 18. But that’s not a sensible way to map words to numbers. You want to map words to numbers, so that similar words map to similar numbers, and that’s what Google did with embeddings. So that was one of the big things. The second thing that Google did was in order to train AI systems to do whatever they’re good at doing, you need input output pairs of information to train these systems. Before chat, GPT, probably the best AI example was in translation on how you translate language from one language to another language, English to French and so on.
(12:32):
And most of you probably use Google Translate and other services. In order to do this, they had to get input output pairs. And where do you get this from? People like you but who work for organizations like the United Nations. United Nations had a mandate where every document they processed had to be converted into all the languages of member countries. Now you have a rich corpus in which you have input output, and that’s how it translate was born. But the key thing to this again, is to remember that if you have enough input output pairs of anything, these systems now get very good at learning these input output pairs. The third big development came about seven years ago when a team at Google came up with this very wonderful technology called attention, which is now a key part of all AI systems. And it’s a very simple thing for most of us.
(13:28):
So for instance, if we were looking at a sentence which said, Mary got up from her sleep to look at and left the word blank, what would you fill it as? Now suppose in context you knew that the discussion was about Christianity and so on. So then Mary now obviously means something very different because the previous context was about Christianity and you might want to say looked at something else related to religion, but suppose the previous context was nothing about Christianity. It was just a bunch of people and Mary was just another person. So you might say looked at something else, right? So in order to fill in the gap, we would figure out which part that came before was important. Was Mary important or was something else important? And so humans are very good at paying attention to different parts of what came before in order to speak this seven year back, Google came up with a technology where the algorithm also was doing that.
(14:36):
It learned to pay attention to the right words that came before it in an intelligent manner. And that was a paper they published called Attention is All Your Need. And this paper really brought us modern AI because when they implemented the ideas in this paper at scale and trained it on all the information on the web, we got chat, GPD. And so the very last thing here is the feedback and how you can get better by trial and error, how you can get better by doing the same things many times that humans are very good at doing. There was a team based in the United Kingdom called DeepMind, which Google acquired that built technology to do this really well. And this is part of a lot of the game playing systems. And now if one of the common things you look at in all these four, like the big advancements that led to modern AI being audits, you see Google everywhere.
(15:35):
And this is in a sense the modern AI that we have. A lot of the innovations happened here and Google deserves tremendous credit for it. But now of course everyone is using it, right? That everything here is public. There’s nothing proprietary. So today’s AI has lots of things. It’s a very complicated world. There’s the good, there’s the bad, there’s the ugly. And every day you see examples of this that AI can do amazing things to help people. There are also some extremely bad things it can do. And one thing that I’ve learned being in technology for a long time is that whenever you have very powerful technology, it sometimes feels as if the bad guys use it better than the good guys, almost always. But then when you think about it, it’s not really bad guys or good guys. It’s whoever gets the most return on investment.
(16:31):
And it’s just that the bad folks for them ROI means something very different. And that’s how it’s been all the time. And today we actually should expect that when we have powerful capabilities, it can be misused and it will be misused and it is being misused on a daily basis at extremely dangerous global scales as well. And so this is something for us to watch out for and collectively figure out how do we defend against really bad things that can happen. The ugly is other issues. There’s the environment. AI is very energy hungry. There’s this issue of copyright. AI learns from input output data and how did it get that data? AI is great at programming. How did it get created programming by looking at US program and now we are out of jobs, and so how do we think of this? How do we decide that it’s okay to do certain things and it’s not okay to do other things? So these are the ugly side of AI that needs a lot of attention as well. So last point, and then everything else is small business, I promise.
(17:46):
There are two questions about humans and AI, which are very important. And this does also make a very important, I mean this obviously matters to all businesses and small businesses as well, which is how do you react to certain things out there? So one of the things that you keep hearing off is AI will replace humans at pretty much everything and the United States has moved towards an information economy in the last few decades, and AI is exceptionally good at information processing tasks. Now you put these two things together, it looks really bad because that’s what people are doing. They’re processing information and that’s what AI is doing. This morning I was at this event where I have a very nice tag by Anthropic. It’s now 160 billion company, so they can give out nice tags
(18:40):
And in downtown, but these are one of the big AI companies. You have the big players, you have open AI with its Microsoft partnership, you have Gemini with Google Lama, with meta and Atropic is Amazon, not Amazon, but partners, closed partners. Interestingly in the conference they didn’t mention Amazon even once it was all just clawed. But you hear that and this team keeps coming out. They presented data this morning to show how at programming at things like this, this is vast superior to humans. And they had quotes from their own folks saying as I’m waiting for Claude to finish this task, my main question is, what should I be doing? These are the things that software engineers are asking saying that, look, this is capable of incredible things at information processing tasks. So what does this mean for us? So one of the things that I tell people is that, yeah, look, anything is possible, but we’ve always known that look for these types of models that learn from data tend to be good when the future is like the past.
(19:54):
This is where they excel. Humans over time have become good at what I call as edge cases where something unusual happens and we know what to do. And I don’t think AI is there yet in a way in which we’ll let it handle edge cases. At an extreme example, sometimes I tell people that, look, many of you have doctors and you go to your doctor, you see what they say when you show up. And nine times out of 10, I feel I could have done the same thing. And you probably feel the same way as well. And if you and I could have done the same thing and AI trained on data can also do the same thing, but that 20 minutes a week, then that doctor, doctor is making a decision that you or I could not have made the doctor is probably saving some lives.
(20:45):
If you know exactly when that 20 minutes comes, you can fire the doctor for the rest of the time and hire the person for 20 minutes, but you can never tell when these edge cases show up, they could show up at any point in time. And so these types of scenarios is where often human intelligence tends to be vast superior. Now does it mean AI will not ever be good at edge cases? Maybe, right? I’m not sure, but I don’t think we’ll ever be able to trust it on edge cases. And so therefore on this question, I think we are seeing AI do a lot of information processing tasks most people can do. So the message that’s going out is that it’s right now it’s probably better than an average human being at any information processing task, but that’s the point where it’s at. So what does it mean for us? I really think we have to keep ourselves where we are better at the edge cases and so on.
(21:46):
So essentially, I think some people say humans make mistakes too, which is true, but I think we’ve come to understand that we know how humans make mistakes and we can train them so that they don’t make those mistakes. It’s very hard to know where AI makes mistakes. When AI makes mistakes, it makes ridiculous mistakes. And I don’t think I really like this mindset of how to think about modern AI. Now, Mustafa Suleman who is the chief AI officer at Microsoft, this is his metaphor and he’s trying to build AI at Microsoft, and his book is a very scary book that if any of you haven’t, if you’ve read, he’s the guy who’s built DeepMind and one of the big thought leaders in the AI space. I think I forget the title of his book, the Coming Wave, look it up by most Offa Suleman. It’s a fantastic book, but it’s pretty scary in terms of what all AI can do.
(22:46):
And he’s the guy who’s been building it for the last 20 years, so he should know what he’s talking about and he’s now the chief AI officer. But his metaphor now is, look, rather than thinking of automating people out, et cetera, maybe the mindset should be how do we give everyone a superpower in their pockets? And this I think is what we port to small businesses too, which is if you think of AI as this tremendously capable thing that can do a lot of information processing task extremely well, if we design it well for each one of us instead of trying to do our jobs, it should be our superpower. It should be something that’s helping you doing your job the best that you can do. This is not how the tech companies are approaching building AI systems necessarily, but as enterprises start using AI, this is the mindset which will be very good for them to start adopting.
(23:41):
And this is the same mindset now that small businesses will also benefit from what kind of superpowers do they need to do their job well that they did not have earlier that they now can have with AI? Lots of many such examples. And you spend five minutes, if I get you in breakouts groups, you’ll come up with 20 more examples of things that modern AI can do that can help these small businesses do their own tasks easier. Now, I’ll give you, at Maryland, we did this AI for small business workshop, and I may show you a link there in a couple of slides. Last year, last summer we did it in 2024, we ran a workshop like this. We had 50 small businesses come and we reached out to the AI company saying, do you want to come talk to them? Show what you had.
(24:41):
We got zero responses and essentially a year and a half back what it was was these big tech companies wanted to sell to the big companies. They had no interest in selling to the small business because they did not see them as a viable market yet. And that was an issue because these guys are almost 45% of the GDP in the United States this summer when we did it in June, there were a ton of startups that had small business AI solutions. So in a year there’s a huge uptick in the kinds of things we are seeing. And these are the kinds of examples we are seeing in our students work with small businesses to do some demos for them to take their use cases and help them use AI. And so this one is for instance, there are many apps that now fall into this category.
(25:34):
These are no code apps that can create, these are no code solutions that can create apps and websites automatically. In one sentence, you get your app almost something like this, give me an app to help customers make reservations for lawn mowing services. If you use one of these products, which is called Bubble, it’ll create this app for you, which is mind blowing. Earlier, you would have to hire people, do all kinds of things to get this done. But now you can do this. There’s this very popular social media, the app called Jasper, which is aimed at small businesses as well to create tons of marketing content for them automatically. There is this thing called Sprout that if a small businesses wants to run stuff on Instagram and so on and TikTok, this thing can automate that for them. So you just focus on your job.
(26:29):
Social media is taken care of, marketing is taken care of, and AI chatbot is answering your emails. Life is good. Are we there yet? No. But these types of solutions have started coming up. And if you think of these types of solutions, these are the examples of superpowers that small businesses need. They’re walking around doing a million things. And I tell people when we do these workshops, people tend to say Mom and pop, right? Small businesses. I say, no, it’s not mom and pop. Sometimes it’s only mom or only pop, right? This is a team of one trying to do everything and every additional piece of help they can get is going to be a productivity boost. So the biggest boost that will come to the economy will come from individual users of AI and small businesses using AI. But this sector has had the least formal support from the tech companies or from the government significantly.
(27:25):
SBA has been doing a lot of good things, but I feel small businesses need a lot of support to make this happen. Right now we have the tools, but the kinds of things that Norma is doing is going to actually make it more possible. So the previous examples, just as a disclaimer, these are not recommendations by me, but these are examples, right? You’ll find many of these things everywhere. So the one thing that I wanted to note here is what we have done here. So I was telling you that this small business workshop we did was in collaboration with Prince George’s County in Maryland, and they approached us a year and a half back, and now we’ve had two workshops with them. So in this slide, the first one, I’ll share these slides with you. In the first link you’ll see what this workshop looks like.
(28:19):
There’s a small YouTube video where we interview some of the people who participated in, it’s like a two minute clip that you can see. What we did after we did this, we followed this up with a free course course on AI empowerment for small businesses that we launched in last year sometime. And this is free open to anyone in the world. And the most important and really impactful thing I think we did was with this one. So this has a different history. In February mid of this year, we started seeing a lot of layoffs in the DC area with the federal workforce and others. And at the University of Maryland, we wanted to do something to help them. I got together with a team of 10 faculty and we put together a 10 module AI and career empowerment course in order to help them transition into the private sector.
(29:17):
And as part of, we have 10 modules, we have content from faculty, we also have interviews from industry experts, chief AI officers at firms and so on. So it took us two and a half months and it’s completely free. We wanted to do this to help the community. We launched it May 1st, and as of I think today we have over 30,000 people from every federal agency who’ve gone through this program, and it’s one of the best things we’ve done. The content is very broad, it’s more than, I mean, this was meant for mid-career people trying to figure AI out, but this too, I think could be very useful for small businesses in addition to what we did with the Coursera course. In the Coursera course, we have about 700 people who’ve taken it. This one, we’ve already had 30,000 who’ve gone through this program, but these are very exciting times.
(30:12):
What we want to do, our vision, we haven’t done all of these yet at, but at University of Maryland right now I’m partnering with the SBDC, the EDC and others as well. Our goal is to try to build a one-stop shop, a digital platform for small businesses so that whoever you are, you can come to this one place and get everything you want where you use AI to talk to this platform and it’ll tell you if you’re a hairdresser, if you’re running a equipment company for agriculture, what are the AI tools that you can use and how do you get it? So we are in the process of figuring out how to build this because if we build it, we feel it’ll be a resource for the entire small business community that’s out there. So I think we’ll go into our conversation with Norma, and Norma can tell you the amazing things she is doing now in her newer capacity helping small businesses, and both of us will be here till however long you need us to, right? If you have questions later, and you can always follow up on email as well. But
Norma McCowin/Master’s in Information Systems Advisory Council (31:26):
So hello everybody. I’m Norma McCowin, and as Balaji said, I started a company Creative AI Solutions as I sit on the advisory board at University of Maryland Information Systems. And I like what biology was saying, was sitting there thinking this is an amazing amount of information and amazing tools that most small businesses or medium-sized businesses wouldn’t have the advantage to be able to take advantage of. And so as I’m sitting there, I’m participating in all of the courses that he mentioned, all of the events that he mentioned, and I’m sitting there and I’m thinking, how can I help? I’m also responsible as the advisory chairman of the MSIS advisory board as the chairman, I’m responsible for trying to find opportunities for our students. We have some amazing students who are super knowledgeable in technology in business. And what I was finding is it was hard to be able to find them opportunities.
(32:23):
And so I started putting together what I thought was the right mix to be able to capitalize on what’s happening in AI and what’s happening for our students at Maryland trying to find jobs. And so I launched the company, these are my first two interns that are graduates from the University of Maryland. And what we do is we sit down with small businesses who are afraid of what they are hearing about AI and show them exactly how AI can make their life easier. Just like he said, what we’re finding is it gives them the ability to be able to remove redundancy in what they’re doing, create efficiencies for what they’re doing, which then be able to create the business value that they need. So what we’re doing is we’re showing these small businesses what tools are out there in open source for them, allowing them to then be responsible and have more ability to be able to control their technology and then also be able to create for them that business value that they’re looking for. So I actually have my employees here, the students too, for you guys to ask us questions about what we’re doing, how we’re doing it. I think the right thing to do is to use this as a platform to be able to answer your questions because just like Rachel said, there’s a lot of fear out there about AI and really and truly it’s that thing in your pocket that’s actually going to help you to be more efficient and effective as we go through. Why
Rachel Jones/NPF (33:52):
Don’t we get some introductions from the two of you right now? So lemme hand it to
Divya Mohan/Product Manager (33:56):
You. Hi, my name is Divya. I’m a May, 2025 graduate from M-S-I-S-A, Smith UMD. And this is an exciting opportunity and an amazing platform to talk about AI. And this is something I think which we are hearing from the last two years, generative AI and the things that it can do. And it is simply amazing how we are evolving not just in terms of businesses, but also how there’s an infusion of technology. Even senior citizens are now more tech savvy as compared to what they were two years back. I think this is an exciting discussion and thank you for letting me back part of it.
Rajat Nirwan/University of Maryland (34:49):
Hi everyone, my name is Rajat. I’m from India and I recently graduated in May 25. So I have been in technology since I passed out from my high school. And I love thinking with technology, either it’s hardware or software. And ever since AI came into picture, it gave me another superpower, which is to, I can code, I can do anything I want to, even if I don’t know the tech stack. So it has become a different experience altogether now I create even AI systems, I create AI models and also work with nor mine giving these capabilities to small businesses. Great.
Rachel Jones/NPF (35:31):
So at this point, do you, let’s open it up for questions because again, as I think about this issue, what it boils down to is how does this benefit small businesses who may be strained in their capacities, who may have budget issues, et cetera. How can AI fill some of those gaps? So I see we have a questionnaire. Introduce yourself and ask your
Diara J. Townes | Sandhills News (35:56):
Question. My name is Diara J Townes. I’m a reporter with the Sandhills News and my question is regarding small business owners who leverage AI and then hit hurdles. For example, there’s a small business owner in my town and they leveraged something similar to Bubble called Lovable, built themselves a website, and then they had to use other tools in order to keep the website running. So they ran into engineering issues. This is the edge functions. They had no idea what that meant. So I want to know more what does the, after they leverage AI issues look like?
Norma McCowin/Master’s in Information Systems Advisory Council (36:31):
You want to check that one?
Diara J. Townes | Sandhills News (36:33):
I’ll take,
Norma McCowin/Master’s in Information Systems Advisory Council (36:34):
Go ahead. So I’ll say that’s some of the challenge that we’re trying to also solve for companies that use technology and then don’t have the resources themselves to be able to sustain. That’s one of the things that we just recently talked about. How do we also expand our capabilities to be a servicing company, to be able to help small businesses that run into that exact same problem.
Balaji Padmanabhan/University of Maryland (36:59):
So lovable I heard of as well before, I think Bubble and some of the others are now getting more attention. So if you ask the companies, they’ll say the same thing. They’ll say, that was version one, wait for version two. We’ll actually continuously be the product that will take care. I think we’ll get there for sure. Now in the short run, what is probably good is that you’ll need this and then you’ll need some help. And that help could be graduate students, it could be others who like Norma’s companies and others who can fill that need, get you up and running quick so that you say, look, I have this problem two days later you have an app and you have somebody who can help you. So I think that’s still a step up, but you’re right, I think these issues will maybe go away and version two comes
Rowan Hetzer | Dayton Business Journal (37:47):
Microphone. Hi, I’m Rowan. I’m with the Dayton Business Journal. My question is a little more UX UI centric in the sense of I use an AI chat bot when I have to deal with at t and stuff and it’s not good. It makes me really infuriated. I’ve
Balaji Padmanabhan/University of Maryland (38:01):
Had the same experience. Exactly.
Rowan Hetzer | Dayton Business Journal (38:02):
And it’s
Balaji Padmanabhan/University of Maryland (38:04):
Not at t Verizon.
Rowan Hetzer | Dayton Business Journal (38:07):
And then I think you also get some people, especially in my generation who are just a little turned off by the usage of AI and then whatnot. Do you have any recommendations for small businesses when it comes to navigating how their clientele, their customers are interacting with their AI if they find those types of issues in there?
Balaji Padmanabhan/University of Maryland (38:25):
Can I ask you to expand a little bit when you said my generation is turned off by using a,
Rowan Hetzer | Dayton Business Journal (38:29):
There are certain, I’ve just seen it in the circles I am in and the things that I’ve seen on the internet. Some people are turned off by companies that over rely on AI. They would rather talk to a representative than talk to an AI chat bot. They would rather see a social media campaign designed by a person for people rather than an AI designing it, things like that. So that just kind of inherently turns and that just some people, because I can’t speak to everyone or every generation or anything like that, but that’s a trend that I’ve every generation hates. She can, every generation hates it. So I’m just curious if you have any advice on something like that.
Balaji Padmanabhan/University of Maryland (39:03):
I moved from an apartment to a house. I had Verizon in the apartment, I had Verizon in the house and I called Verizon in the apartment to say, I want to cancel it because I already had Verizon in the house and it took me 50 minutes to do that because the chat bot thought I was leaving and it was really frustrating. Eventually it got it right, but this would’ve been five minutes for a human to do. Now if you think of why Verizon, and I don’t want to put them in the, you said at and t, let’s just say providers, why these folks are doing it for them. They’re seeing cost advantages and unfortunately some of them are in industries which are not super competitive. They can afford to do these things. And what we are seeing is that instead of humans now by using these systems eventually that which will get it right, they’re saving money, but they’re passing on a lot of costs to us. So if you take 10 million people talking to these chat bots times 30 minutes of wasted time, do the math on our time, that’s lost in this process and it’s very frustrating. So I’ve seen reports that now are saying that many of these large companies have started hiring back customer service representatives and so on. And my most recent Verizon call, I spoke to a human in five minutes.
(40:19):
And so it is frustrating to everyone. And so this for small businesses is a bigger risk because they’re in a very competitive space. Verizon and at t can afford to do this, but if you are a small business and you have 20 people who can take your service, it’s super dangerous for you to rely on this chatbot that upsets people and then now you lost a business. So I can certainly see that as a cautionary tale for some of them to say, don’t go overboard and fully automate this process. Be a little more thoughtful on where you want to do it. But amazing point. I think this is all generations though,
Amelia Schafer | ICT (40:59):
Amelia Schafer, ICT news. So I just wanted to ask a bit. So I think this work is definitely very beneficial for small businesses, especially those that don’t have a lot of staff. However, I worry the more small businesses we have that use this, how much more you’ve mentioned water consumption, what impact is this going to have on the environment? The more businesses we have relying on this, the more people we have using this, even just for speaking on the phone or a chat bot, I just wonder is there any way to mitigate that or I guess I’m not familiar with the technology you guys are specifically using, but with chat GPT, that’s a lot of water for the cooling. That’s a lot of energy. How do we mitigate that?
Balaji Padmanabhan/University of Maryland (41:38):
Talk about that. So I think there’s always the plus and the minus, right? So I think if you look at the plus side of how small businesses most of us have seen people was swamped by think of they’re running a restaurant, phone is ringing off the hook, they have orders coming in that they cannot take because they’re running in and they’re cooking stuff themselves. There’s the AC mechanic who’s fixing mechanics, ac, and his phone is ringing from customers who want to schedule his appointment and he is losing business. These things AI is very good at taking care of, right? So I’m busy doing my job, my AI talks to the person saying, look, I’ll make an appointment for you, et cetera. So boom, I haven’t lost sales, right? The person is cooking at the back, AI takes the order efficiently, now there’s another order. So I think that’s where I feel we can help the small business now on the energy use.
(42:27):
Absolutely right. So I think every use of these systems is now burning GPUs somewhere. I don’t think there’s been a good handle for things. I mean these tech companies haven’t released energy consumption details in a very transparent way. They don’t report this as much as maybe they should and the systems have become very efficient, like deep seek. For instance, for all the negative press it got, one of the amazing things Deep Seek did was fantastic technology that made the cost of producing output tokens a fraction of what Chad GPT and others were using. And that automatically means energy consumption drops by a 10th. And so these systems have two things on the technology side, two things are happening. One is there are newer technologies that are making it easier to produce output tokens, which means lesser energy cost. There is also a shift towards smaller models in of large language models. So now they’re talking about small language, small large language models, SL lms, which are not so big. They’re good at one specific task, but because they’re smaller, the energy consumption is significantly lesser. So that’s what’s happening on that side. I don’t have a better answer on how we can control people from going crazy by just using as much energy they can get access to.
Rachel Jones/NPF (44:07):
Sorry, question
Omar Mohammed | The Boston Globe (44:12):
Omar Mohammed from the Boston Globe. I have a couple questions. One was actually a follow up from the previous one. Talk a little bit about why exactly these data centers require so much energy and then what was the answer to the question that you heard from your anthropic meetup today? What do people do if AI can do everything? What’s the answer to that question?
Balaji Padmanabhan/University of Maryland (44:47):
So the first question I’ll make Norma answer the second question. I think on the first question, AI’s energy used by and large is in two parts. One is for training these models and second is for using these models. And so these AI models are constantly being updated. I mean they released this as version 4.1, 4.2, et cetera. But in between those releases that are periodic sub releases as well. And so they’re constantly retraining these models and that’s very expensive from energy use. The bigger energy use comes in usage. Every time you and I go to chat GPT and say, I’m having a hard time sleeping, tell me a story, read it out to me. We are actually producing output tokens that’s running on a bunch of GPUs in a data center producing that song that supposedly put me to sleep. But that’s how we are using energy, right?
(45:49):
Each time we talk to this like we are the data center, these models when we are using it, the difference is that these are trillion parameter models. And so actually producing a single character on your screen is a process that’s running a neural network that’s trillions of a trillion parameter model and that is expensive in energy cost to produce that output. So these data centers are doing a combination, they’re storing a lot of data from which these models can be trained. They’re also being used to produce output to serve to us, and that’s what makes it very energy rich and I as to what humans should be doing. I think Norma can the event there was no answers they released yesterday, this anthropic economic index report that is now widely available and people are starting to talk about it in the media everywhere. You can look up Anthropic economic index report and they are looking at usage of cloud and providing statistics on how cloud usage is reflecting what’s going on in the economy. And their claim is they’re the first AI companies to do this. Chad Open AI and others should also look at usage and start reporting stats because only then we’ll know what AI is being used for. So
Rachel Jones/NPF (47:12):
We probably take two more questions.
A J Johnson/Bridge Detroit (47:16):
Wait, there was the second part of the question.
Norma McCowin/Master’s in Information Systems Advisory Council (47:21):
I was going to answer the second part of the question and say, when I think about technology, it’s always evolving. So those people that are sitting there, they are the brain trust and they should be the ones they’re working on. How do we continue to make it even better than what it is? Like the version 1.1 to the version 3.0 right there is always the way technology is, it’s always evolving. So the thought that it’s creating for people, a space to be able to be even more creative to me is how I would’ve thought somebody would’ve answered that. But it’s interesting that they’re silent because when I think about technology, it’s moving and growing and stuff so fast that it’s almost at the speed of light before you know it the next version or the next generation. And so there’s that opportunity for them to continue to be creative and to create what is the next amazing thing technology.
Rachel Jones/NPF (48:11):
OK, so we’re going to have two more questions. Remember that I’m going to be around,
Norma McCowin/Master’s in Information Systems Advisory Council (48:14):
Oh, that’s right. We’ll be around next. Can we just also say we’re hearing you guys talk a lot about stories about where you’re hearing challenges in AI, but we actually have some real success stories that we would like to share. So please come up to us so you can hear some of those because we want to answer your questions though. Let’s
A J Johnson/Bridge Detroit (48:31):
Go back here and then we’ll come back up. We can go right up there. I’m Alina Johnson from Bridge Detroit and I am also a small business owner. So I would like to know three things all related. So one, what is the name of your company? Two, can anyone use the service but also three, do you take advantage of the PTECH or the procurement technical advisory centers that assist small business in things they don’t know how to do, such as use and take advantage of procurement and contract opportunities with state local governments?
Norma McCowin/Master’s in Information Systems Advisory Council (49:20):
So first question, my company is Creative AI solutions and that is also the website. Second question is yes, we do take any and everybody. And third question is no, we have not done that yet, but we would love to get some information to be able to help. Thank you.
Rachel Jones/NPF (49:43):
And one more, was there one more back here? Oh yes. OK.
Scot Refsland | West Virginia Daily News (49:52):
Only do. Thank you. My name is Scott, I’m a publisher for West Virginia Daily News and Virginia Review and a couple other publications in West Virginia and Virginia. So far you’ve only been talking about content and content for everybody here writing it, but everybody in this room requires eyeballs on their content. The issue that’s happening right now is AI has been destroying web traffic. And lemme give you some statistics real quick. So Forbes and Huffington Post has lost 40% of their web traffic. Daily mail has plunged 32%. CNN has failed 28%. Fox News Corp has gone down 24%. Basically in terms like New York Post, they drew 433 million page views in June 24, falling to 381 million in 2025. What this means is that people are going to Google and then immediately the AI is in the front and it’s going to zero clicks, which means that this is an imminent threat on every single newspaper in this room. What are your suggestions in how newspapers can pivot and use AI to actually generate revenues instead of having revenues taken away?
Balaji Padmanabhan/University of Maryland (51:22):
So I had earlier this year in summer, I had in a very short trip to India, I had given a talk on AI to the Hindu, which is one of the largest newspapers there in the country. And it’s a different market because there for instance, still they tried going digital, it was not, digital doesn’t work in all markets the same way. But one of the things that I had recommended to them was that was around the time the day I was giving a talk there the same day or the previous day, I think one of the AI companies had a big partnership with a media company in the US for content does exactly right. And so there are two ways of looking at it. One is that, look, these AI companies may become the next Google. That’s the point at which most people are going to interact with information.
(52:16):
And so to find out what you all are creating, they’ll go through that channel. And so you have to also deliver the content through that channel. Now, the new media company is used to that because you’ve seen this game play out before. You’ve had to do it through the social media sites previously, and that was the partnership with Meta and so on. Now, the AI companies are very, very hungry for high quality information in version one. They could get by with a lot of crap because people cut a lot of slack. Now I think it’s not the case. So they also understand that having high quality content is making a big difference in output. So this is an opportunity for media companies to get together and maybe start forming larger groups. I see consolidation as a natural outcome of this, and then to partner with AI
companies so that the content you produce becomes input to them as a licensing stream as opposed to selling it directly through your existing channels.
(53:14):
This is an option. I’m not saying that’s the only option, but this is an option for the media industry to look at. The second would be to not rely on eyeballs and the digital advertising that comes from that is just shrinking, right? As you rightly pointed out, there’s no easy solution to that because the attention is being diverted to some of these other channels to focus more on other sources of revenue if it’s possible. In terms of, we have a business of journalism conference at the University of Maryland, which happened last summer where we had a big discussion that we are highlighting firms like the Baltimore Banner and others that are exploding newer models, but we can talk offline. But
Rachel Jones/NPF (54:08):
Well, at this point I can see why the motto for the University of Maryland is feared the turtle because what we have here at Dynamic Group of Visionaries, really in this field of AI and small business. So I’d like us to take this opportunity now to thank Balaji and Norma for joining us today for this illuminating conversation.
###
