Why should CPAs understand ChatGPT?

In the spring, two technology-focused accounting educators wrote an article about ChatGPT’s ability to answer hypothetical client questions. Now, they’re joining the JofA podcast to look into the future of generative AI.

Mfon Akpan, CGMA, DBA, an assistant professor of accounting at Methodist University, and Scott Dell, CPA, DBA, an assistant professor of accounting at Francis Marion University, explain why CPAs and other professionals can be left behind professionally if they don’t spend time learning about tools such as ChatGPT.

They talk about productivity gains, the competitive divide that could develop, and how, in the short term, more training is needed to fully harness generative AI.

Akpan and Dell also note that although the power of generative AI tools is exciting, these tools are far from perfect. In particular, accuracy remains a concern.

Editor’s note: The interview was recorded in August.

What you’ll learn from this episode:

• How Akpan and Dell got to know each other.
• A forecast of how generative AI will look in 2024.
• Some of the other tools competing with ChatGPT.
• Akpan’s “aha moment” when giving a recent presentation.
• Why Dell said — twice — “the sky’s the limit” with AI tools.

Play the episode below or read the edited transcript:

To comment on this episode or to suggest an idea for another episode, contact Neil Amato at Neil.Amato@aicpa-cima.com.

Transcript

Neil Amato: Welcome back to the Journal of Accountancy podcast. This is your host, Neil Amato. Today, we’re going to have a conversation about ChatGPT. It is a popular topic, and we have two experts here to talk about it. They’ve written an article recently in the Journal of Accountancy on the topic, and we’re going to have them explain a little more about their knowledge of ChatGPT, their excitement about it, and what you as a listener can get from ChatGPT.

Our guests are Scott Dell and Mfon Akpan. I’m going to have each of them introduce themselves a little bit before we get into the conversation. Mfon and Scott, welcome to the Journal of Accountancy podcast.

Mfon Akpan: Hey, thanks. Thanks for having us here, Neil. My name’s Dr. Mfon Akpan. I’m assistant professor of accounting at Methodist University in Fayetteville, North Carolina.

Scott Dell: I’m Dr. Scott Dell. Excited to be here. Neil, thank you for the opportunity to share insights. I am also a professor of accounting and a CPA. I have been a full-time academic for 15 years and excited to share insights about this technology.

Amato: Well, I’m going to get right to the why of this conversation, why we’re having it to start. That is, why should CPAs understand generative AI, which includes but is certainly not limited to ChatGPT?

Dell: I’ll take a stab. The power that this technology has — bigger than the printing press, possibly bigger than written language — has changed society as we know it, is changing business as we know it. If you don’t hop onto it, especially accountants, we will fall behind.

The technology itself has the ability to make amazing strides if you’re professional. If you know what you’re doing, it’ll 10x your efforts. If you’re an amateur or a novice, or some of our clients, or some of our students, when they use it, they won’t get the full benefit because they won’t know what’s real and what’s not real. But as a professional, when we read it, we should be able to decipher that and be able to have that opportunity to take advantage of what this technology has to offer.

Akpan: That was good. I think I’ll piggyback on that with a movie analogy. There’s a Netflix movie called Me Time with Mark Wahlberg and Kevin Hart. They’re the two main characters. They’re best buddies. I mean, it’s a good movie and there’s a scene in that movie where Mark Wahlberg is talking. They’re at a party. He’s telling everyone in the party about how good friends he is with Kevin Hart’s character and how Kevin Hart — he’s a musician and he recorded an album. Mark Wahlberg keeps talking about this album. Everyone at the party is like, well, how can we get the album? Kevin Hart, he looks around, he said, “Well, it’s on DVD.” So, everybody is looking and they’re saying, “Well, how can we play a DVD?” I think to tie into that, you don’t want to be the CPA in Kevin Hart’s position with this technology as it’s being implemented.

Dell: I’m just going to piggyback on that really, really good insight. But I’ll tell you, like I said, our clients are using it. Our staff is using it. If we don’t take advantage and start using it, we will fall off the wagon. I think it was a Mark Cuban quote that says, “There are going to be two types of companies in the future: Those that are good at AI and succeed, and those that aren’t and don’t.”

Amato: Probably the first Kevin Hart reference we’ve ever had in the Journal of Accountancy podcast. Thank you for that. Let’s step back for a bit before we go into the topic. How do you to know each other, and I guess what led you to reconnect in the past year? And what’s it been like tag-teaming this topic of generative AI?

Akpan: Well, we’ve known each other since, I want to say 2018, 2019 from AAA.

Dell: Actually, I’d say even a couple of years earlier because I came out to South Carolina in 2019. We were both in the Midwest; I was in Wisconsin and you were in Chicago. We hit it off [at] AAA, you’re right. We’re both tech geeks. We both have fun with the technologies. I had a computer consulting company and Mfon has been doing YouTube videos and all kinds of technical applications in and outside the classroom. We hit it off, and we’ve been keeping in touch career-wise. Just when this AI thing hit, I think we said, “You know what, we’re both doing some cool things.” Then I think that’s where we flowed in and said, “Why don’t we combine efforts?” — which we have been, and good things [are] happening.

Akpan: Yeah, exactly. Scott has a newsletter. I follow his newsletter.

Dell: On LinkedIn, yep.

Akpan: He did one on ChatGPT. That prompted me to make a video based on his newsletter, and then I posted the video on LinkedIn. And then things snowballed from there. We said, “Hey, look, this is working out good. Let’s combine these efforts and put some things together.”

Dell: We were just at AAA. We did a combined presentation at the Teaching and Learning Conference, CTLA, on the weekend. We did a combined presentation during the week at the AAA conference; we’ve been doing presentations for universities; multiple articles — Strategic Finance, Journal of Accountancy — have been published. Mfon is a sharp guy.

Amato: Again, AAA in this case, American Accounting Association for those not aware. These guys are not into car repair at the side of the road. You’re the co-authors of a popular article on the Journal of Accountancy site. That article published in the spring, so it’s been a few months. What I’d like to know, for either one of you really, is what does 2024 the near future look like as it relates to generative AI?

Dell: I’ll tell you — 2024, the sky’s the limit. As of two months ago, they maybe had 20 plug-ins, 30 plug-ins. Now there are 500 plug-ins.

And let me explain a plug-in for a minute. What that is like is an app on a phone. They have a whole bunch of apps that you can use, and it gets past a September [2021] deadline of data that was in there. You can now access current data. You can access Expedia, you can access current PDFs, you can make videos. You can access all kinds of tools, and those tools are going to grow exponentially, I feel, and make it much more useful and much more current than you had with the limitation of the past.

You used to be able to say, “Well, what’s a good stock to invest in?” It will tell you in the summer of 2021 what a great stock to invest in, and I can tell you that, too. But if you say now, “What’s a good stock to invest in?” There’re actually financial apps or financial plug-ins that you can use to analyze current data and current environments.

Amato: Mfon, anything to add there?

Akpan: Yeah, I would say things are going to move forward, but it’s hard to say. We really don’t know exactly what it’s going to look like. And there’s a book by Steven Johnson called Where Good Ideas Come From, and he talks about this 10/10 rule. The 10/10 rule states that it takes about 10 years to develop technology and then about 10 years to implement technology. There’s examples of the color TV. It took about 10 years to develop, about 10 years to implement. Then, I guess the example I gave with Kevin Hart, where you think about the DVD, the time to develop and then to implement. Then we have streaming.

Tying into what Scott was saying, that doesn’t exist anymore. Especially when we talk about ChatGPT and you think about this: It’s been around for less than a year. Moving forward, I mean, we don’t know, but we do know things are going to continue to grow and implement and improve.

Dell: ChatGPT-3.5 and GPT-4 are not the only games in town. Bard has been in development a long time. Even though you’re hearing ChatGPT and OpenAI and Microsoft invests $10 billion and that’s got some good press, it opened the doors for exposure for AI, for the masses, I’ll say.

But it is open right now and accessible to anybody who wants to for free. There’ll be more platforms, though, expanding. And we’ll be using — not necessarily doing prompts in ChatGPT or GPT-4, but in our applications. Everything from Microsoft Office to Snapchat, AI’s built in.

Amato: Going back to that 10/10 rule and, I guess, the mindset shift that has to occur with these products. Everything’s moving faster than it was before. The iPhone took a few years. I don’t know if it was 10 and 10, but it definitely took years, not days, weeks, months. How are professionals, CPAs included, going to have to change the way they think as they adapt to these products coming into play in their businesses?

Dell: Tax rules are changing every year. Audits change every year. And [with] the acceleration of technology, we need to change and adapt. It’s all about lifelong learning. There’s a good reason why we have continuing professional education in the profession. The tools we need to be kept more comfortable with, and learning them is going to be crucial to long-term success and even short-term success, because you’re right, it’s happening faster and faster.

Akpan: Yeah, and I think one tip is to look for ways to add AI into your workflows, to try to find ways where you can incorporate it into what you’re doing and in ways that you’re comfortable with. That is helpful, but also it is constantly changing, and you have to keep up with the learning. But I think putting your toe in the water and finding ways to use it would definitely help, and it will keep you on track as it develops. Because if you’re not using it and it continues to move forward, you’re going to have a gap in your knowledge base and then ultimately your productivity.

Dell: Mfon brings up a great point. Practice, practice, practice. How did we learn Excel? Did we say, “You know what, today I’ll learn Excel and three months from now, I’ll use it again.” No, we were in it, we’re constantly using it. If you can dedicate 10 to 15 minutes a day — even in the morning, afternoon, any query you have — take advantage of it. That’s the personal growth and lifelong learning we all need to exhibit.

Akpan: Yeah. On that note with Excel, the big elephant in the room is Microsoft Copilot is coming. It’s going to be in [Microsoft] 365 for Business. How exactly will that look? How that will impact things we do not exactly know, but it’s coming. So to shorten that learning curve by using the technology and becoming familiar with it, that can help you to ramp up when that does come into place. The other thing with that — you have to think about if Microsoft is rolling that out, like Scott was saying, other companies are rolling out similar products and will be embedding similar products as well for productivity at different levels.

Dell: Copilot is cool. My biggest disappointment is they announced it’s going to be $30 per month per user in addition to Office to really take advantage. But just a little background on Copilot because a lot of folks haven’t heard of it. It’s going to take your Excel, it’s going to take your Word, it’s going to take your Outlook, it will take your LinkedIn, and it will propose emails. I will say, “Give me a graph,” and it will give you a graph on Excel.

It will say, “Forget the PivotTables, analyze this, present that,” and it’s going to change [and] modify dynamically. You’re creating a proposal. You say, “Take my last proposal, here’s my new client, and here are the new objectives, here’s a new scope of service.” Then, it’ll just change the whole letter for you and be done without you having to go into the weeds and go into detail.

You still have to review it. It still hallucinates. Or as many people know, it lies, makes stuff up. But you need to, as a professional, go through that and say, “Boy, that saved me a whole bunch of time.” One of those things, at the American Accounting Association conference, I was talking to my alma mater EY, [to] a partner there. And they were saying, “I took an eight-hour audit activity and condensed it to less than five minutes.” If you’re not using it, the folks that are going to take advantage of that, and you’re going to be left in the dust.

Amato: Right, because even that five-minute audit activity, if it takes, let’s just say, an hour to review the accuracy of it, you’re still at an hour and five minutes?

Akpan: Exactly.

Amato: In presenting, whether at conferences or CPE, webinars, etc., I think one of the things you’ve asked people is, “How many of you — raise your hand or press the button on the meeting — how many of you are aware you’re using AI already?” What are people saying about that?

Dell: Overall, we’re seeing more and more people aware that they’ve been using AI from Netflix, from Google Maps. They’re using AI when they are using Siri or Alexa. There’s a greater awareness and more and more people have exposure, except so many people still are not using it. Anybody listening to this that says, “I really haven’t played with it at all,” you are not alone.

There are so many people out there, you hear it every day. Go log in, it’s free. Go to openai.com and you can access this stuff, ChatGPT, and just experiment. One other comment I will make, I used to call it Google on steroids. I’ve since changed my tune. Google’s a one-way street — “Hi, let me ask you a question.” — You’re going to give me a whole bunch of options, it can be a whole bunch of sources, and I can go in and do my digging. It’s different.

This is not a platform that you ask one question and you’re done. It’s not a one and done. You’re now going to ask a question, then you’re going to have a conversation. You go back and forth and your prompt is key. We’ll talk about prompt engineering, I think, in a little bit. But the prompt is key, how you ask the questions, the level of detail you give it, and what you want in an answer, you indicate what the tone you want it to respond.

The power in this is you say, “Well, actually, item No. 3, could you write a blog post about that?” Item No. 2, it remembers the conversation. It remembers what you’ve asked, it remembers what it has responded, and you can take advantage of that as you progress.

Akpan: I’ve had similar experiences. At my last presentation this past week, I had an aha moment where many of the participants, they said they’ve used 3.5, so they’ve used ChatGPT. However, I realized they’ve used the 3.5 when it first came out in 2022. The aha moment was really for them and for myself realizing that OK, there was 3.5, but that original 3.5 was discontinued.

At one point, there was 3.5, 3.5 Turbo, 4.0, and then the 3.5 was discontinued. Many of the perceptions and really a lot of complaints around hallucination with ChatGPT [were] based on a version that is discontinued. It’s been outdated. That goes back to talk about how fast these things are moving. This is less than a year, and that version that rolled out November 30 in 2022 doesn’t exist anymore.

Amato: You mentioned that word earlier, prompt engineering. That’s a phrase that probably none of us had even thought about three years ago, maybe even a year ago. But what are some of the jobs that are going to be created through companies adopting generative AI?

Dell: Lots of jobs will be lost and lots will be created. Let me quote a World Economic Forum article that came out in 2020. It said 85 million jobs will be lost and 97 million created. That’s a 12 million gain. You don’t have to be a CPA to calculate that one. Twelve million gain in jobs. You say, “Oh, that doesn’t sound so bad.” Back in May of this year, they revised that and said 82 million will be lost and 69 million gained. That sounds like a net loss. Frankly, they don’t know either. They’re throwing darts, they’re playing with it. But they’ve got some interesting stats and interesting analysis.

But you look at that and say, “OK, yes, jobs are being replaced. By the way, there are a lot of layoffs going off in high-tech right now.” I won’t say those are AI layoffs. Some are, but the majority of them, I think, were over-hiring during COVID. The fact is, we need to pick up more skills. The idea of being able to understand, AI is not going to take your job, but someone who knows AI just might.

That was one of the famous quotes that’s been around and circulating and I agree with that wholeheartedly. My concern is the timing. In the short run, we’re going to probably lose more jobs than gain. New skill sets are needed. It’s going to take a while to get those skills and build up those skills to recoup those jobs. But we’ve had revolutionary times. We’ve been through all kinds of industrial revolutions to go with it. This is more of a white-collar than a blue-collar industrial revolution, or a revolution, I should say. There are different kinds of jobs that’ll be gained and lost. But again, if you’re keeping up with it, if you’re on top of it, then there’s no reason why you can’t take advantage of these tools. It lets you do so much more. You can do the fun stuff. You’re not stuck in the mundane.

Akpan: When you brought up prompting, I would say prompting, in my opinion, is the secret sauce when we talk about generative AI. An example of that comes to mind is Dave Wood and some other accounting researchers.

They put it to the CPA test. So they [tested whether] ChatGPT could pass the CPA Exam. And I thought this was interesting, but in my opinion, it did, but it didn’t. And this comes down to the prompting. So when they fed it the questions, [versions] 3.5, 4.0, it did not pass, and there wasn’t that much of a difference between the two.

However, when they did something called 10-shot prompting, and this is more AI jargon. A shot means they’re giving an example of what they want the ChatGPT to solve. In other words, they had to give it 10 examples of the problem that they want it to solve, and by doing that for those questions, then 4.0 was able to score 86% and pass the CPA Exam.

So that’s why I said it did, but it didn’t. You had to give it a lot of examples to get the higher score. So that ties in with that importance of prompting. The better the prompts, the better the output at this time. They do say, as the platforms get better, there’s going to be less and less need for prompting, or advanced prompting, as they call it.

Amato: Just to be clear, prompts, you would say, are simply the instructions submitted to the AI when asking it to do something. Is that a simple explanation?

Dell: Prompting is how you set up the question, and you set up the example, you set up the background information, you set up how you want the output, the tone. Give it to me in the style of Dr. Seuss. Any anything you want, you can ask it and it will provide. If I may piggyback on the prompting ideas of that though. Neil, you were talking about when has it been around.

Many people have seen the article, yeah, $300,000-a-year-plus job. That was before November 30, when ChatGPT came out and that was a real job. The catch is that Sam Altman [OpenAI’s CEO] said that’s probably going to be obsolete in the next year or two, and I think actually faster, because you can say, give me a prompt to answer this question in the artificial intelligence and it’ll reformat the prompt.

People were doing that for Midjourney. Give me a prompt to draw this picture and you can do that in GPT-4 or ChatGPT-3.5, and it’ll give you a prompt. What’s the best prompt? You ask it and then you can have a dialogue. Ask me what questions you need. I’m making a résumé. What do you want to know? Give me a cover letter. I’m taking a job description and I’m taking my résumé. What am I missing? What kind of keywords you’re missing?

Those are things in the career space, in the professional space. This tool is amazing at doing that kind of analysis. It’s not really good at numbers yet, by the way. Excel is great at numbers and this computer, this is a word generator. The technology says, here’s my last word. What’s the next word with a good probability? Number-crunching is challenging right now in the AI space.

Akpan: That’s a good point because a lot of this right now, you can learn yourself. By working with it, you can begin to understand prompting and have conversations with ChatGPT, so you can begin to understand it. But as things move forward, if you’re not using it, then there’s going to be a bigger and bigger gap or learning curve that you’re going to have to overcome.

Amato: This kind of harkens back to Scott’s comment about the Mark Cuban quote, but Mfon, in a preliminary call, you had said as adoption increases there’s going to be more of a divide. I think we know what that means, but could you tell me more of what you mean by that divide?

Akpan: Well, there’s going to be a divide as far as skills, so you’ll have a learning curve, so if you’re starting at zero to get to a certain point, either you’re going to have to spend a lot of time to learn up, or you’re going to have to pay for some sort of course or some sort of training. Or you’ll have to outsource in order to take advantage of the AI tools available. That’s one.

The other one is access. Scott was mentioning, there’s $30, so you get 365 for Business. On top of that, it’s another $30 per user so now there’s another cost barrier there to have access to these productivity tools. I’m presuming, if Microsoft is No. 1, you may not be able to afford it, or your business is not able to afford it, there’s going to be a gap in productivity. So you may have to use the No. 2 or the No. 3 version of AI. So there will be some sort of divides there just in the workplace — not to talk about the education space as well.

Dell: We talked about the digital divide and the AI divide. Those are real divides and some ethical considerations. But let me bring up the education space. There was a study done, 1,000-plus HR folks. Eighty-six percent said that ChatGPT experience could and is more important than a college education in a number of instances. The survey questions were a little funky on how they asked it, but almost 90%? Better than a college education?

But the actual tools themselves, think about it. We can bury our head in the sand, or we can embrace. Academics were in the same boat. Professionals are in the same boat. But think of high school. A lot of school systems — this started with New York City, Baltimore, Seattle, L.A. — a lot of school systems were banning it, saying, “Well, students are going to cheat like mad dogs and they’re not going to learn.”

But if it’s that important to have ChatGPT experience or AI experience when you get out of high school, or when you go to college, or when you go on the job, you need those skills. Well, think about the haves, the folks that can say, “Mom, Dad, my Chromebook can access this AI stuff, but I got to do a homework assignment. Can I borrow your computer?” What parent is going to say no?

How about the folks that have a Chromebook and don’t have a parent or access to the internet or another computer at home? They’re going to graduate high school without having to use this stuff? Talk about an extra divide, talk about an extra level or layer that they’re going to have to catch up on if they haven’t used it. We, as educators, need to teach those critical thinking skills and adapt to this technology and help our students use it effectively. That’s where we have to go with it.

Akpan: Yeah, and Scott — well, actually both of you — gave the great example. If you’ve got an employee who can do eight hours of work in five minutes, and then maybe you have to check it for an hour, that’s an hour and five minutes versus eight hours. That’s a valuable employee, right, because of the productivity. Then you think of it from a business perspective. How are you going to compete if you’re not using AI and you don’t have employees that can use AI? It’s going to take you eight hours to do the same work your competitor is able to complete in an hour and five minutes, so the landscape shift is there.

Amato: That’s great. This has been a fantastic conversation. I have learned a lot. I know there’s still more that everyone’s going to learn day by day, week by week, but I guess I’ll ask you Scott first, maybe your takeaways and on what’s ahead and then also Mfon after him.

Dell: Again, Neil, thank you for the opportunity to be here. Obviously, Mfon and I thrive on this technology. We have a good time. I have a good time with it and a good time sharing, because it is changing capitalism, changing the economy. It’s going to be influential in our futures. The technology is going to get better and better.

There’s a Bill Gates quote that says we overestimate the next 2 years, underestimate the next 10. I’ve revised that. We overestimate the next six months, but underestimate the next two years. So that was a 1999 quote from Bill Gates. That is a much more appropriate, as we got to — you know, Mfon talked [about] the 10/10 — we’re condensing it. We’re speeding up.

There’s a study put out. It’s not Moore’s Law, which is every two years, [microchips] double, cut in half price. They’re saying AI is doubling every six months or quadrupling every year, and that’s been going on for the past decade. Not just since 2022, but go back 10 years, and that’s continuing. It’s accelerating. When we get to quantum computing and apply AI to that, the sky’s the limit.

Akpan: That was a good point and I wanted to say thank you again for having us here. This is really a great opportunity. What I would say a takeaway, one is Scott mentioned this before, but you got to have a mindset of lifelong learning and being comfortable with change. I would say that’s No. 1. Then based on that, the good thing about it is, because things are moving so fast, you won’t know everything.

Then three, I would say, spend some time once a week, once a month, looking for ways to add AI to your workflow. So to figure out how you can add an AI tool to what you’re doing to increase your productivity. And by doing that, you’ll figure out, OK, I can use it in this area. You’ll find out what you’re comfortable with. You’ll find out what you can use it for. You’ll be able to measure your productivity, see how much faster you can get things done.

Then you can look for other tools, because there’s so many tools that are already out there and then so many other tools that are on the way. Those would be my three things to take away.

Dell: You bring up a great point, Mfon, about the continuous learning and the continuous need. Your colleagues are using the tools. Your staff is using the tools. As we said before, your clients are using these tools. Have those conversations. Inject that into your meetings and discussions of all how are we applying this, and see what other people are doing with it because lot of people have good ideas. Take advantage of this technology.