Business Applications of Generative AI
In our fourth Fireside Chat, FlexPoint Consulting Founder & CEO Michael Daehne is joined by Rachel Lockett, Chief Information Officer at Pohlad Companies, to discuss the innovations and business applications of generative AI and unpack the explosive growth of tools like ChatGPT.
Read more about Pohlad Companies at https://pohladcompanies.com/.
After this episode, we are shifting from video interviews to audio on our new podcast, Inflect. We invite you to subscribe on your favorite podcast player.
Opening
Producer: Welcome to Inflect, where we discuss navigating the critical inflection points of business and life. These conversations were previously known as Fireside Chats.
In this episode, Rachel Lockett and Michael Daehne discuss the innovations and applications of generative AI [artificial intelligence] and unpack the explosive growth of tools like ChatGPT.
Rachel is the Chief Information Officer at Pohlad Companies and has worked with Michael and the FlexPoint team on several transformation projects. Here’s more from Rachel.
Michael Daehne: Hey Rachel.
Rachel Lockett: Hey Michael, how’re you doing?
MD: Good, welcome back.
Background
MD: I feel like we just did a Fireside Chat, but alas, technology has evolved and changed in just a few months, so I’m glad you’re back to talk about it.
RL: And I’ve always loved talking about technology with you, so I’ll come back as often as you’ll have me.
MD: Thank you for joining.
Just as background for anyone who saw our Chat, I guess it was several months ago, and might be wondering why we’re doing this again. In the time since we talked before, Rachel and I both have been going back and forth about ChatGPT, which has definitely captured the cultural zeitgeist and the attention of the world I think, with how quickly it has grown in its usage.
Both of us have been thinking about all the use cases and the ways this can be used in the enterprise setting, and we did a talk a few weeks back with some folks in your group and thought, hey, this might be good to add this into the Fireside Chat library. So that’s why we’re back to talk today about ChatGPT.
Introduction to ChatGPT
RL: Yeah, it’s been amazing, the explosive growth. I shared this story with you earlier: the way I came to learn about it was a little surprising to me. My son, as you know, is a college student. He called me up one day. He never calls, he always texts, but he actually called me. He said, “Mom, there’s this really cool new tool that’ll write essays for you. You can put in a prompt and tell it how many sources you need it to cite, and what questions you need it to answer, and it’ll write these amazing essays.” So naturally I responded with a lecture on academic integrity. And then I didn’t think much more about it.
Then a few days later, a colleague posted something on LinkedIn about both ChatGPT and Lensa. And I took a minute, they tagged me, to research Lensa. I downloaded the app, but I stopped at the point where you had to purchase a subscription in order to create a magic avatar, saying “that just sounds kind of gimmicky, not anything I need, and I don’t feel like signing up for something else and putting my credit card number in another place out there.” So, I just moved along.
That was all in the first or second week of December. The holidays came, and I took a little social media break, and I come back after the new year and open up LinkedIn. Oh my gosh, every third post was about ChatGPT, and I thought, “what did I miss?” So, I quickly researched it, and it was really easy. You just log in, put in your email address, create an OpenAI account, and start typing in prompts.
It was right as I was coming up with that first prompt that it occurred to me, “this is the tool my son called me about.” And then I went, “oh my gosh.” It was cool and exciting. And then the explosive growth – how it went from launch on November 30 to a million users in five days – is just absolutely stunning.
MD: I don’t know if you experienced this, too, but right when it started to really be in the news, all over LinkedIn, I went to try it out. And that was in the phase when it couldn’t handle the usage, and it was like, “come back; try again later.” I almost felt like I was waiting for the debut for a new movie or a fancy new restaurant: when can I actually get in? Then I spent some time over a weekend playing with it, and it was so fascinating, and as we’ll talk about, it’s bringing into the consciousness or into view this underlying technology that’s been there but making it a little more consumable.
RL: That’s the key that we’ve talked about.
Talk about where ChatGPT fits in, because ChatGPT itself isn’t the new groundbreaking technology. It’s that combination of other technologies that have been in development. Really what they’ve done is make it accessible and consumable to everyone. So, as you did before, talk a little bit about what is really the technology behind it and where it’s come from.
Generative Artificial Intelligence
MD: Yeah, I made this comment to someone recently. I think ChatGPT is the thing that CIOs’ kids are talking about; generative AI is the underlying technology that CIOs are talking about. And so, to your point, ChatGPT is a natural language processing AI model that’s trained on a massive data set of text, and it’s really able to, based on a prompt, generate net new text based on what it perceives to be the way sentences should be structured, and the words and the content and all that.
That is the cool shiny object that everyone is playing with, and saying “ChatGPT, will you write me a haiku about digital transformation?” or “will you write me a sonnet about XYZ?” But underneath that is just this broader discipline of generative AI, which we’ve talked about, is really the idea of saying, “it’s not just AI in a pattern recognition way, or clustering models, or some of these more traditional forms of machine learning AI.”
Generative AI is generating new content. So, it’s saying, “based on all the rules you’ve set, how you’ve designed me the model to learn, I’m going to generate new content based on the prompt.”
With ChatGPT, it’s all text-based. But you could think about generative AI generating new music based on a data set of past music, or images, that’s something there are some other tools like DALL-E you might have heard of or DALL-E 2, where they’re generating new images based on prompts. And that creates all sorts of trademark and deep fake issues, so that’s a whole other story.
All of that is to say, the underlying concept of a model being able to generate something new, I think that’s really the game-changing piece here that could be useful.
RL: And I think an additional feature or capability that’s really exciting is, of course ChatGPT been trained for several years, and that’s part of the capability, but also the ability just in my own conversation with it to refine the prompt. So, I put in a prompt and ask it to write something, and then it produces the output back. And I just did this the other day, I said, “that’s good, but can you make it a little less formal.” And it replied and said, “okay, how’s this?” Or another time I said, “I’d like you to add in a little bit of this,” and it actually replied and said, “that’s a great idea, here’s another try.” I felt very validated.
But what’s important about that, and this gets us back to the academic integrity comment from earlier. I think that this is going to change, as a lot of technologies do, how we think about learning and education, and just like spell check and some of those things decreased the importance of learning spelling and grammar, sadly, I think this is changing what are children really need to learn.
Now it becomes more important to learn how to ask the questions, how to write the prompts. They’ve got all the knowledge in the world at their fingertips. Why make them memorize and learn things that they can get so easily? But where the skill and the true knowledge really lies is in how to write the prompts and how to ask the questions.
MD: Well said, and I think that – what’s the term? digital dexterity? – I think of that here. How do we enable and train and empower the humans to interact with the machines or the models, the algorithms. How you ask the question or write the right prompt is certainly useful for individual users with ChatGPT. But especially as we think about how we use this technology for enterprise use cases, it’s not as simple as just saying, “we’re going to use AI to solve this problem.” It’s about: “what’s the interaction and the handoff between the thought leader or expert person who has been doing all of this manually with the algorithm that might be taking over 10% of the process, or running with a piece of it?”
Practical Applications of Generative AI
RL: You mentioned the practical applications or the enterprise usage. What are some of the practical applications of generative AI, not just ChatGPT but the concept overall, either that you’ve seen or you’ve started to explore and do work with for clients or elsewhere?
MD: Good question. There are a few that are almost “yeah, no duh” kind of use cases that I think make a lot of sense around chatbots for customer service use cases and content generation. I can speak to those in a little bit more detail. And then there are some that I think that I think most folks wouldn’t immediately think about around code generation and debugging. I want to come back to that in a minute, but I think that’s more of a nuanced use case.
On chatbots, I was flying home from Minneapolis last Friday night. My flight got diverted to Chicago because of bad weather. So, I’m on the plane and immediately open the Delta app to try to get rebooked or figure out how I’m going to get home. It immediately says, “do you want to chat with us?” So, I start chatting with the bot for like fifteen minutes before it handed me off to someone. You can think about that amped up, with ChatGPT not only doing the if-then statements and “give me this confirmation number” but really making it more of a conversation around the problem trying to be solved in the way that it's communicating with the user. So, I think from a customer service center / omnichannel experience, we’re going to see a lot more of that in the coming years.
Now thinking from a content generation perspective. The house next door to me is for sale. Someone spent time writing a paragraph to describe that house and why someone should buy it. There’s a generative AI use case for that, to say, based on all these inputs, go write a description that’s going to pull someone in and say they want to buy this house. You can think about that in a commercial real estate setting very similarly.
RL: Right, I was walking down the hall the other day, and our communications director stopped me because she’d read my LinkedIn post about ChatGPT a few weeks ago. She said, “oh my gosh, it’s the greatest thing ever. It’s going to make me so much better at my job.” She explained that in her peer group of communications directors, there are some who are worried and scared and “oh my gosh, what does this mean for our jobs?” and noting the copyright concerns. She said, “not me, this is going to make me so much more efficient, and I can focus on the things that only I can do and get rid of the things that can be automated.” I really think that’s the best approach.
But back to the practical applications, what else?
MD: You’ve seen this in the course of your career, and I have, too. That’s always the case with new technologies. It’s not how do you fight it; it’s how do you adapt and collaborate with the new technology in a way that’s good.
One last thing on the content piece, then I’ll talk about coding. I think there’s a huge play around knowledge management and knowledge bases here, too. At a lot of organizations, folks that are new and are onboarding or are in a new role, whatever the case may be, they have a question. They probably go look in SharePoint and do a search, or in some wiki or some training materials. You can imagine saying, “hey, when you’re new to this company, talk to our ChatGPT interface, and it’s going to go search through all the knowledge base out there and generate back an answer to your question of “how do I submit my expense report” or “how do I do XYZ?” So, I think that’s something we’ll start to see in the enterprise setting. There are already some tools that are sprouting up that are trying to play in that space.
And then the last thing that I alluded to: there’s some stuff already in flight around using generative AI not only for writing and generating code but more so for debugging code. You know this, so many developers spend much of their time not writing the code, it’s debugging it when it’s not doing what it should. There’s a real opportunity to use generative AI to say, “I would have expected this to be in the script or the code, and I didn’t,” go look in line 146 and see if you’ve got the right variable or the right syntax.
RL: Sure, or what about documentation? The reasons I got out of programming.
MD: Yeah, no more commenting in code, if you can get to it all through a generative AI interface.
Adopting Generative AI
MD: What about in the work you’re doing with Pohlad and your operating companies? You’re maybe not using it yet, but thoughts about how you can?
RL: The stage we’re at is I went to our leadership team meeting and just did that same introductory session, educating them. I even started back at what G, P, and T stand for [generative pre-trained transformer]. And then opened it up and had them write some prompts, and they were kind of blown away. Our Chief Legal Officer was like, “um, I might be replaced pretty soon.” She actually asked it to write up a contract for something, and it did a decent job.
Yeah, I think of some of the potential applications. As you know, we own a group of automotive dealerships, and I asked one of the leaders there, “how long before generative AI can replace service advisors?” and saw a little moment of panic in their eyes. Or you already alluded to writing the lease abstracts in commercial real estate and things like that. I’m pretty excited about the potential future. But, as with most things, it’s going to take time and it’s going to take that change management and adoption process and people getting familiar with the concept.
But that’s why I love what ChatGPT has done: it’s made it accessible to everyone so people can start to get familiar with the ideas. And being built into Microsoft Teams and other tools, people are going to start to see and get familiar with it. And then, and that’s part of that digital dexterity progress and maturity process, then they’ll start thinking about, “how could this actually make my job easier?” instead of being threatened by it.
MD: Absolutely. I was reflecting on this this morning. Even in the month or so since we did the talk with some of the IT leaders within your organization, I think we’ve already started to go from the peak of inflated expectations to the trough of disillusionment. And some of it is, all these enterprise software companies, whether it’s Bing or Teams with Microsoft, or Salesforce, they’re trying to get generative AI and the various tools into the tools as fast as possible, and some of them are falling on their face a little bit because it’s early and there’s so much to evolve. It’s interesting that it has the world’s attention, and the eyes are watching to see how it goes. I think it’s going to be really interesting to see over the coming years, as enterprises embrace it, as software companies embrace it, how they move out of that trough of disillusionment.
Lessons to Learn from ChatGPT’s Early Months
RL: I think there’s something for us to learn from ChatGPT’s rapid, explosive growth since launch. A couple of things I asked myself: How did they do this? How did it go from zero to one million users in five days, and then it’s up over 100 million now – probably way beyond that, I haven’t checked lately. How did they accomplish that when so many other things took months, even years, to reach that point?
I think there’s a couple of things. First of all, as we already mentioned in the beginning, the ease of access. You didn’t have to pay a subscription fee. You might have shown up on a day when it was too busy, but wait a few hours or wait until midnight, and you’ll be fine. That low barrier to entry: you don’t have to buy an expensive piece of equipment, and you don’t have to give them a bunch of personal data. You input an email address and you’re good. I think that was key.
The other thing, when I was pondering this and trying to figure it out, I called up my son – well, I probably just texted him, actually – “you remember that tool you told me about that writes really great essays.” He said, “no, I’m not using it, Mom, don’t worry.” And I said, “no, no, no, tell me where you heard about it.” And he replied, “YouTube, duh.” The viral spread, you know, it spread among the college students. Like you said, the CIOs’ kids, that’s how they learned about it. And I think there’s something to be learned there as well.
And then the fun aspect. For example, I asked it to write a love letter to my husband, and it came up with something really generic. Then I refined my prompt and gave it a little bit more information about how long we’ve been together, and how many kids we have, and what I like about him. And then it wrote something that brought tears to my eyes. I think that fun aspect is something we can learn and apply when we’re trying to drive adoption of technology tools elsewhere.
MD: Just to underscore, something you and I have talked about in the various projects we’ve worked on together. It’s really important that the technology is accessible, consumable, dare I say fun sometimes. That is not usually how technology transformation looks and feels to employees and customers. It usually feels more like uncomfortable, rushed to the finished line, forced change. We definitely can learn from that.
RL: And sitting in a two-hour training session to learn how to use it.
MD: Yeah, versus, “hey, it’s kind of self-explanatory. Go log in and try it.” That probably won’t work on a new ERP, but I think being inspired by that in the way we go about transformation efforts is certainly a great lesson to take from this.
Concerns around Generative AI
RL: Let’s talk for a minute about some of the concerns. What are the weaknesses, the ethical and legal considerations? You started to allude to them a little bit. What are some of the things that will drag it down into the trough of disillusionment and create problems and headaches going forward?
MD: I think the easier way to describe it, from my perspective, is there just aren’t a lot of guardrails because it’s so new. What always happens is that new technology creates the necessity for new counter-technologies or guardrails or whatever you want to call them to manage that.
The easy examples here are generating images. Say, “hey, DALL-E, make me a picture of Michael standing in the middle of the street with an advertisement in his hand for a competing consulting firm.” I don’t want that out there, I want FlexPoint all the way. So, there’s all that around how do we protect against those deep fakes or those images that are creating misinformation, so to say?
With the text examples, we’ve been talking a lot about ChatGPT. ChatGPT doesn’t really have a good filter so to say. It lacks some context sometimes. It has some bias in the data set it was trained on. If I said, “ChatGPT, what do you think of my shirt today?” it might say, “terrible, I hate purple.” If I asked a human, they might say, “it’s not my favorite, but you do you.” So, there are things like that where the lack of context, the ethical risks around misinformation, bias, and fakes, I think those are a couple.
RL: There’s even accuracy concerns. I asked it to tell me about the company I work for, and nine out of ten points were accurate, but it claimed that we owned a business we don’t own, and that we were still in a business that we sold five years ago. So, there’s still some accuracy concern around that as well. But I think people are really mostly concerned about some of the ethical considerations, and “how do I know that this is something that was created by a human versus a machine, and do I care?”
Closing
MD: Well, I’m excited to watch this and see it evolve and see it move from the fun use case that everyone’s looking at and playing with, moving toward how are we actually leveraging this in a business setting to drive value. I think it’s going to be a fun journey.
RL: For sure. I’m looking forward to where it takes us and new opportunities that it opens up and makes available. And looking forward to continuing the conversation. I can’t wait to hear about the first client that comes along and asks you to integrate or build something with it.
MD: Absolutely. Thank you so much for your time, Rachel. Probably right around the time we get this published, there will be some new technology or trend out there that will require us to do another one.
RL: We’ll do it again. Sounds good.
MD: Thanks, Rachel.
RL: Thank you. Take care.