In the latest episode of ADDcast, Steve Gent, New England Account Manager at ADD Systems, discusses the past, present, and future of artificial intelligence (AI). He starts with the meaning and history of AI and then mentions present applications of how AI is used. Steve also explains the ways AI affects and can potentially improve the energy and c-store industries.
You can subscribe to the ADDcast at any of the following locations, or read the transcript below.
Apple Podcasts / iTunes: https://podcasts.apple.com/us/podcast/add-systems-addcast/id1604865113
Spotify: https://open.spotify.com/show/1atj72i5AnIwTJDBFD9DzX
Amazon / Audible: https://music.amazon.com/podcasts/df7c9f4a-8a3f-431f-8787-50eee85b2bcd
Castbox: https://castbox.fm/channel/id4744519?country=us
You can also search “ADD Systems” wherever you listen to your favorite podcasts!
Brian Cohen: Welcome to ADDcast. I’m ADD Systems Multimedia Specialist, Brian Cohen. Joining me today is Steve Gent, New England Account Manager at ADD Systems. Steve, thanks for joining me today.
Steve Gent: Yeah, my pleasure, Brian.
Brian: Can you start by telling us a little bit about your role at ADD Systems?
Steve: My role at ADD Systems is comprised of being an Account Manager for New England and Alaska, where I get to travel to great states of New England and call and help our prospects and our existing customers, as well as attend trade shows and occasionally do a presentation or two.
Brian: Well, lately, you’ve been attending a lot of industry events and speaking about AI in the industry.
First, can you help us by defining AI and maybe talking a little bit about the history?
Steve: Yeah, sure. Well, let me back up a little bit. It started with a presentation on analytics, and then the producer of the presentation said, “Hey, can you add some information about AI?” And which was a natural segue, right? Because AI is basically using the data and analytics that you have. But to answer your question about a little bit of the history, technically, 1950s, a guy named Alan Turing was considered the godfather of AI. He was the guy largely accredited with breaking the Enigma Code that the Germans were using in World War Two. Mathematician, right? So he started with code that was recognizing speech and just kind of heavier lift in terms of understanding speech, and it kind of kept evolving to whereas they were using it for answering questions. And there used to be a thing called the Turing Test to demonstrate whether the computer was getting any smarter. And the areas that they worked on was kind of machine learning as well as vision and robotics. And initially, it was very singular input with singular output. What ended up happening was, in the 70s, IBM created a machine called Deep Blue, and the first kind of recognized AI challenge was against a world chess champion, Kasparov, that Deep Blue beat this world champion in chess. And then there was an AI machine that went up against the two Jeopardy champions and beat them handily. So those were considered like the first public examples that people kind of recognized in the limelight.
And I think one of the things that has struck me when I have been fielding questions and these presentations is really like an understanding of what is AI? People don’t really have a clue, kind of where it ends and where it begins. And, you know, I had a question yesterday from someone that’s like, “Is Alexa AI?” Like, no, not really. It’s more like an automated voice recognition into Google, right? Ask a question, get an answer. It’s not really building or learning from your questions. It’s not conversational. The true AI concept is more like a chat, where you ask a question, also known as a prompt, and then you receive an answer. But as you continue to ask questions and further define what information you’re looking for, you’re getting responses from whatever agent that you’re using.
I have an example of what AI is, or a couple of examples, as well as kind of my analogy for what the structure would be. So kind of like a car engine. You can have a car engine, and it depends on what chassis you put that engine in as to the application. So you can put an engine in a Mazda Miata or a Chevy or a sedan or a truck, and it’s the same engine, but it depends on the fuel that you’re putting into that engine and the framework around that engine as to what the application and where it will be most useful from an AI perspective.
Brian: Can you talk about some of the different types of applications of AI?
Steve: Yeah, sure. So a couple of different applications. I’ll just use an example. So self-driving cars, take a Tesla. Every Tesla that you see on the road is feeding all the driving information up to one central neural net, and what that’s doing is mapping so that the next Tesla that drives down this road knows that it turns at 18 degrees, goes over a hill, through a tunnel, and then it hits a stoplight. So when you’re driving down the highway and there’s construction and it literally bridges to the other highway coming in the other direction, or the other lanes, there has been training so that the car with its driving down the road knows to expect that jog, even though the lines may still kind of go in one direction, the highway goes in a different direction. You have a baseline of mapping for information, and then layered on top of it is more real-time information, so you’re going around a corner, and if you look straight ahead, there might be an oncoming car. That self-driving automobile has to be intelligent enough to know that that’s not a head-on collision, but you’re driving at 55 miles an hour, going around the corner to the right, and that car is going to go by 10 feet to your left. Another example of that would be you’re driving down the road. You’re going 55 miles an hour, and there’s somebody walking along the side. Now, is that person in the lane of traffic, or is it on a sidewalk that’s three feet to the right, the car has to understand its real-time inputs, make decisions on those inputs, and then continue to direct a self-driving automobile. So that would be a quick example.
The different types of applications relative to AI, I break into two general categories, generative and computational. And the largest application that we all see on a regular basis is more generative. So you can do it in the creative world, where people are looking to create graphics, they’re looking to have help on an email, they had blood taken, and there’s 20 different acronyms and ratios, and they’re related, but they have no idea what those individual acronyms mean, but you can go into AI and start to get real-world answers or layman’s answers and explanations. And again, it’s conversational. So you might put LDL and whatever level it is, and it’ll give you a definition of that cholesterol. But then there’s a ratio to the HDL that’s important, too. So you start to build the chat conversation, and from that you’re engaging with artificial intelligence, but it all depends on the prompt that you’re putting in there.
The other type of AI is more computational. What I mean by that is it’s creating code for software. It’s taking real inputs from external sources and computing information to change your decision about, you know, an example that I would use, as you type into your phone where your destination is, and it comes back and says, “There’s a 10-minute delay up the road. I’m going to reroute you. Is that okay?” And you’re like, “Yes.” That’s computational, using real-time inputs to affect your decisions. And you might say, “I’m hungry. Give me a restaurant along the new route.” So you ask it a question, and you get an answer.
Brian: So can you tell us a little bit about the events you’ve been to and the industry segments you’ve spoken with?
Steve: Yeah, sure. So the events have been, you know, on the fuel marketing and subsequently convenience store industry, right? And what’s amazing is there’s a huge disparity between people’s understanding, and then the practical applications of AI. And I guess the best example is that, you know, we’ve all had smartphones for a number of years, you know, and there’s the individual that just wants to make a phone call. And then there’s the individual that is mapping their vacation, planning their vacation, buying airline tickets and selecting the best restaurants, and, you know, really digging in on some of the current applications.
So the specific applications we’ve been speaking towards is first in the convenience store market where you have, say, you have a store and that has 10,000 items, and you have traffic patterns at that store, you have staffing needs at that store. You have what’s on the shelf, combinations of coffee and donuts or hot dogs and chips and soda. And at one store, you might be next to a ferry terminal and coffee and donuts are the big seller, but if you’re at the FedEx terminal, it may be all energy drinks. So totally different customer base, traffic patterns, etc. If you can imagine trying to manage all of that input for a single store, much less compound that to 30, 40, 200 stores, bringing in artificial intelligence and bringing in the computational power of such complex information has really started to potentially help people manage their convenience stores.
On the fuel marketing side, you know, it used to be the conventional wisdom that a driver would take a stack of tickets in the morning and go to the furthest location so that when he was done for the day, he was back home. And then we said, “Okay, we can optimize the route, and tell you a much shorter mileage wise, in shorter time-wise,” whereas that used to be your first stop is now your eighth stop in your day. But guess what? We cut an hour off your day. Now what do you do with the hour? Some people reduce the expenses, reduce the number of trucks, continue to deliver to the same group, and then some people have said, “Okay, let’s add additional people that we’re forecasting that are called will calls. Do some marketing to those will calls and try to migrate them into being more automatics.” So what used to be 17 stops in a day now is 23 stops in a day because you have the additional hour on your day. Let’s add in AI. So a couple different examples. You have the route, but starting in September, you have a school bus that’s always one particular location. So let’s add in real-time traffic patterns. Instead of driving down this road behind the school bus to make that delivery, now we choose a different route, one that’s more effective by adding in real-time traffic patterns. NOAA (National Association and Atmospheric Administration) right now has a significant effort into AI, so we’re going to be able to increase the accuracy of weather predictions and add that as a layer to forecasting and delivery in the fuel marketing side.
Brian: Now what are some of the areas companies are addressing right now with AI and how is it helping?
Steve: A big one is predictive analytics. So we have that convenience store and you have all the data. And what’s the difference between a Friday sale in July? Well, if Fourth of July is on a Friday, it’s a different traffic pattern and a different product usage than if that Fourth of July is on a Tuesday. So adding in additional information and be able to predict what that day is, taking that and knowing what your staffing patterns are going to be for each register at each store on any given day, helping with staffing needs.
In the fuel delivery side, when you have automatics, people that rely on you just to fill up your tank when it’s due, the more will call customers you have, the more uncertainty you have. You don’t know that they’re going to call you. They may call somebody else, right? So if you can migrate those people from will calls to automatics, your ability to understand how much fuel as a fuel marketer that you need to buy to forecast out your buying patterns with your suppliers becomes more reliable. So again, using the advanced analytics and having help from your intelligent software helps lower your costs hedge on what fuel and your fuel commitments are, as well as marketing, too.
So on both of these examples that we’re using, you have loyalty programs in the c-store market. It used to be the gas price. Now we’re getting emails or texts saying, “Hey, there’s a deal on an energy drink, right? We know you like them. Come on in.” Or on the fuel marketing side, we’re using marketing to say, “Hey, we’re going to be on your street next Tuesday. We’ve been forecasting. We see that you’re due for delivery. Would you like us to add you to next Tuesday’s route?” So on both sides of the equation, on the c-store with the proactive loyalty marketing, and on the fuel delivery side with turning will calls into automatics, you can leverage AI; you can leverage generative marketing to appeal to both of those markets.
Brian: So then, what are some of the key areas people have high expectations for AI in the future?
Steve: Well, on a global matter, I think, from what I have seen and, you know, at a larger level, one of the biggest areas that AI is being leveraged is in the medical industry. Being able to take all the information around Brian and predict what’s going to happen in one year, two years, three years. So I think there are high expectations around improvements in medicine.
I think there are high expectations around logistics because it’s a very complicated, you know, thousands of factors involved with stuff, logistics-wise. And then let me give you a different example of what’s coming down the pike. So, let’s imagine that next Tuesday, I’m going to the NACS show in Vegas, and, it’s not unreasonable to think that within 12 to 24 months, I have an app that says, “We’ve booked you on United Airlines’ 6:10 am flight from Newark, New Jersey, and then you’re going to arrive in Las Vegas. And I’ve booked a car with Avis for you, and you’re staying at the Marriott that’s off the strip.” And an agent would be able to consolidate all the various apps that you have out there.
Brian: So then, Steve, what are you sensing around our industry? Is there excitement, nervousness, a little bit of both?
Steve: Yeah, I think people don’t know what they don’t know, and they only know from what they have seen. So I think the excitement is around the potential and understanding that it’s increased computational power and the ability to help with large data sets and help with normal day-to-day activities that are now self-generated, to call and do something. And what I’ve seen is, so there’s an example, like I have a gentleman whose wife is a professor at Columbia, and he said that two students submitted a paper that was AI-generated, and he could tell because they were over 90% the same in the conclusions and virtually identical in terms of the content. And it used to be, you would think one student was cheating off the other, but now you kind of recognize that it’s really just people going in and using generative AI and saying: “Here are my prompts. Give me the answer.”
So if you are not becoming really well versed in some of these generative AI products, your output is going to be generically the same as other people, and when we’re doing marketing, we want to be distinctive. But if we’re all at the same level, or two different companies that are competing are kind of at the same level, it’s going to be generic and look the same.
Or if I send out an email and I ask AI to edit it or appraise it, if you’re a savvy reader, you’re going to say, “Wow, this is like, insert name,” and you know, it looks kind of really canned. But I’ve seen and what’s starting to happen is that there are people that are good at marketing, that refine and instead of taking 30 hours to create an image of a bear holding a heart, they can use the prompts effectively and take a couple of hours to get a really, really impactful graphic that’s cute and unique and doesn’t have human hands holding a heart or a pig nose on a bear or whatever it might be, the trappings, right?
Now, what people are seeing are deep fakes, right? They’re hearing of marketing programs. So I think a lot of people are nervous about the believability of what they’re hearing and seeing because they know that there can be false representation in the market.
And another thing that is a legitimate kind of concern is, you know, I may engage with a chatbot, for instance, and one day I’m asking it to help me define what it is on an AT&T bill. The next day, I’m looking for blood work, you know, definitions, and then asking about storing a camper van for the winter, and what some of the things I should be doing, or I’m looking for a part for that camper van. I’m putting this information in, it’s very personal information, into another company’s data bank. When I’m doing that, you know, these things aren’t forgetting. You know, when I’m conversing with a chatbot and further defining what I’m looking for and giving them more information about myself. That’s, I don’t want to say it’s a public domain, because it’s not necessarily public domain, but I have now given my personal profile to some ubiquitous chatbot, and that’s real. So I know some companies are leveraging AI to help with market share and marketing, and so they might put a profile of their customer or put some customer data in there, and that’s now public domain from a chatbot perspective, and that’s a very real concern – the information and intent of future use of that information because you’re helping the chatbot, on one hand, become smarter and more refined in the answers, but on the other hand, you’re giving away very personal and sometimes confidential information.
Brian: Well then, what are some of the best practices with AI or the do’s and don’ts, if you will?
Steve: I think the realization that it’s not a panacea. I could install ADDs at your location as a business, but if you don’t take the time to learn it and set it up for your application. . .it’s kind of the same with AI. It’s not a panacea. You can’t just pop a question in there and get an answer. I’ll give you an example. It’s all about the prompts. How you ask the question, and if you ask the question, who did the British fight for in the Civil War, your brain first says, wait, what, Civil War? That’s American Civil War. The British weren’t involved. That’s one layer of response that they weren’t involved. But at a lower level, yeah, the British supported the Confederates because for two reasons: One, they were the first, they built the ship for the Navy for the Confederacy; and, secondly, they relied on the textiles from the cotton industry, and so they were favoring the Confederates for the purpose of economics, right? So it depends on what level your prompts are driving at and what you’re expecting for the answers. It depends on the, they’re fallible, and they don’t always have the right answer, and you’ve got to kind of take it at face value and add additional prompts. And a lot of the information is only to 2022, so if it’s something that’s happened in the last 24 months, it might not even be applicable.
Brian: And what is your overall message about the future of AI?
Steve: I think it’s going to be here tomorrow. It sounds kind of cliché, but the future is now. So if you’re a software company, I think you better lean in now and help it, let you help it with the code. Whatever the application, I believe that an exponential increase in the computational power and the assistance and the different agents that are coming out and being produced, I think it’s good to learn about it and to lean in on it because it’s going to have a very real-world impact sooner than we expect.
Brian: Well, Steve, this has been absolutely fascinating. Thank you for taking the time to speak with me today.
Steve: Thanks, Brian.
Brian: To keep up with the latest happenings at ADD Systems. Visit addsys.com/blog, or connect with us on social media by following ADD Systems on LinkedIn, Facebook, Instagram or X. If you have any questions about ADDcast, feel free to reach out to us at addcast@addsys.com. Thanks for listening and have a great day.
Leave a Reply